The Head of Growth at a $100M Company Explains AEO | CEO OS

tldr

Ravish Agrawal (growth at Gamma, ~$100M ARR / 100M users, profitable, ~50 people) treats AEO as a top acquisition channel and the fastest-growing one — more than 90% of it is ChatGPT. The playbook is not "spam the models." Execute like Google (prompt-shaped, comparison-heavy, always-on content), measure like Facebook (self-reported "how did you hear about us," because click-through undercounts). Reddit is #1 for ChatGPT only; Claude, Perplexity, and Google AI Overviews want different work. For a distribution-starved FeatureOS, the new part is not "do Reddit" — it is instrument ChatGPT as a named source, then ship one thick comparison page plus one comparison thread that names you next to Canny/Productboard, not as a pitch.

Key Takeaways

AEO behaves like Google in execution and like Facebook in measurement. Click-through in GA undercounts. Gamma added ChatGPT/Perplexity as first-class "how did you hear about us" answers; self-reported discovery kept rising after referred clicks fell. If you only look at utm/referrer, you will kill the channel.
Reddit is by far the most influential source — for ChatGPT only. Ravish's rank for GPT: Reddit → owned site ↔ other people's sites (PR, blogs, reviews) → YouTube → LinkedIn. LinkedIn citations are thin and unpredictable. X is almost absent from LLM citations even though it ranks in Google.
~30–40% of the AEO playbook overlaps across engines; 60% does not. Google AI Overviews ≈ SEO. Perplexity weights third-party sites more. Claude cites ~5× less; the lever is being in the Claude ecosystem via MCP, which retains users, it does not acquire them. Thesis: Claude looks like Play Store, ChatGPT looks like Google Search.
Prosumer vs enterprise ranking is inverted. Prosumer: ChatGPT → Google AI Overviews → Perplexity → Claude. Enterprise guess: Claude → Microsoft Copilot/Bing → ChatGPT → Google AIO → Perplexity.
Crawler-first, 80/20, not clickbait. On AEO-designated YouTube/LinkedIn work, headings come from real prompts, not human CTR titles. Still 20% human-digestible so a person who lands does not bounce in disgust.
Authority is a level playing field on the open web right now. A new startup site and Forbes start closer than they do in Google. On YouTube, 400k-view and 18k-view videos both get cited. The pattern that does show: being talked about by someone else, in comparison to something else. Listicles and "X vs Y," not "we are great."
Don't "spam content." Write thick content. Ravish's line: if models can write an app they can write a real article. Thin keyword stuffing is the thing Google already started penalizing (May 21 search update). LLMs will grow an authority/relevancy layer; until then the field is open for depth, not sludge.
Reddit tactics that actually move citations: talk in competitor/comparison frames; use an open-ecosystem wedge (import/export) so you belong in the incumbent's subreddit; own-community posts do get cited (how-to-export); comments change sentiment, they do not get you onto the citation list. Negative sentiment is sticky — LLMs appear biased toward it.
Niche down like SEO. You will not win "best CRM." You might win "best CRM for health-tech sales." Then climb. If a competitor copies the niche, keep moving. Always-on: there is a Citation Half-Life; Gamma tracks citation share in a spreadsheet because they do not know when a cite dies. Profound is not doing this for them.
Two games: training vs retrieval. Reddit = training-data intervention (even a citation sends the click to Reddit). Owned and third-party blogs = retrieval/citation into your ecosystem. Logged-in memory is a first-mover loop: once ChatGPT recommended you to a user, it tends to recommend you again. Claimed split: ~60% of ChatGPT queries hit the web, ~40% come from parametric knowledge — you can influence both, with different assets.
Next paradigm is marketing to agents. AEO is the MVP of that. Some growth teams are already putting "10% extra reward if you sign the user up now" into JSON schema / install packages. Gamma is not seeing agentic usage outrun humans yet (presentations are for people). ChatGPT Ads is the paid version of the same surface, still early.

[01:08] Gamma's GTM north star: don't be boring, find the unidentified channel

Hosts Subah Wadhwani and Arthur Zargaryan open Distribution Podcast with Ravish as growth lead at Gamma — AI presentations, prosumer, vs PowerPoint by killing the blank page. Two GTM principles:

Don't be boring. Company value, used daily, because decks are culturally boring. Creative is the GTM.
Identify what the world has not identified yet. Early on AEO and Reddit; now testing ChatGPT Ads. Test whether a new surface fits you, then extract the alpha.

They did not wait for an AEO agency market. Go in while organic is open (Facebook 2012 analogy). He flags the risk: these surfaces can shut organic in months, not years.

[04:23] How they noticed AEO: 8× ChatGPT referrals, then a measurement lie

The origin is not a strategy offsite. Ravish was in Google Analytics. ChatGPT-referred traffic grew, then dipped mid-last-year. During the rise they saw 8× ChatGPT referral traffic across Gamma channels — still <1% of mix, easy to ignore. Two moves:

Keep tracking click-through separately (the only thing GA can see).
Promote ChatGPT/Perplexity from "Other" into a first-class "how did you hear about us" option. Self-reported discovery kept climbing while clicks fell.

Mental model: execute AEO with Google playbooks; attribute it with Facebook-style view-through. If FeatureOS only watches GA referrers, ChatGPT will look dead while people are still arriving "because ChatGPT said so."

Common knowledge at the time: LLMs pick up Reddit. They started Reddit. Now the same engines also pick LinkedIn and YouTube. Several channels they invest in are AEO channels first, native-channel second — you may not get YouTube views, you still feed the citation graph.

Organic is still unusually open: a new site gets crawled quickly by ChatGPT, Claude, Perplexity (in that speed order, in his telling). Google AI Mode is "killing standard SEO practices"; growth peers everywhere report falling CTRs. Everything is in flux, which is why the channel is still deep.

[08:50] 80/20: write for crawlers, don't punish humans

Does AEO feel like gaming? Yes, and it feels bad, because gaming expires. Game theory still applies: if everyone games and you don't, you lose.

For AEO-designated initiatives they flip the ratio: 80% agent/crawler, 20% human. Example: YouTube titles and headings pulled from actual prompt phrases, not clickbait. A human who finds it should not think "what the hell is this," but you are not optimizing the thumbnail.

[10:53] Do LLMs do authority the way Google does?

On websites: basically no. Old Forbes vs a friend's startup is closer to even than in Google. The startup often has more AI-native, prompt-shaped content. Authority will arrive (he assumes someone at the labs is staring at the index the way Google did). Until then it is an open field.

On other platforms: no clean engagement pattern. YouTube citations at 400k views and 18k views. Closest pattern: are you being talked about by someone else, in comparison to something else? The listicle/"vs competitor" playbook, on video and on Reddit. Content that is only about you underperforms.

How long does this last? He does not know. Google's May 21 search update already punished sites that just dumped content. ChatGPT pulls a lot of queries through Bing, which behaves differently. Two wrong assumptions he hates:

"AI-written = spam." Models can write research-grade articles if you let them.
"Volume of thin pages." A glass-of-water page vs why steel beats plastic, with real comparison depth. Thick content is what he thinks LLMs will keep rewarding once they grow an authority layer.

The hosts joke that marketers are using ChatGPT to gaslight ChatGPT. His defense: the labs do the same thing when they train the next model on the last model's outputs. If you are actually serious (real comparisons, real product claims), you are not fooling — you are feeding the retrieval index.

[16:23] Ranking the platforms — and why Claude is a different sport

For ChatGPT, his stack:

Rank	Surface	Notes
1	Reddit	Dominant, relatively predictable after ~6 months of industry pattern-hunting
2 / 3	Owned website ↔ other sites	Switches. Owned = control of how you and competitors are described. Others = blogs, PR, reviews
4	YouTube	Transcripts get picked up; patterns still fuzzy
5	LinkedIn	Thin citations; articles and posts; no engagement correlation; "more bizarre than YouTube"

Reddit only works for ChatGPT. Playbooks:

Google AI Overviews: play SEO. This is where "good SEO = good AEO" is true.
Perplexity: more weight on third-party authority.
Claude: cites ~5× less. Best lever is being in Claude: ship an MCP server, get listed. Constraint: Claude will not discover your MCP and acquire a user for you. You still tell the user "we connect to Claude." Remote OAuth MCP setup is now as easy as old Facebook connect. Thesis: Claude → Play Store (ecosystem, retain, native agents). ChatGPT → Google Search (query, acquire).

Gamma numbers (directional, not a deck): AEO is a top channel, fastest-growing, top of mind, >90% ChatGPT. They have an MCP connector; once a user is in Claude, "create a PPT" can emit a Gamma deck — retention/expansion, not top-of-funnel. They have not cracked Google AI Overviews and now want to invest because AI Mode moved faster than expected. Perplexity is present, smaller.

If you sell prosumer (Gamma, FeatureOS-shaped): ChatGPT, then Google AIO (you get residual SEO), then Perplexity, then Claude.

If you sell enterprise (SupportWire-shaped): he would flip it — Claude first (that's where he's searching for tools), then Copilot/Bing (they've seen real traffic under the Bing umbrella; huge companies), then ChatGPT / Google AIO, Perplexity last. He admits he does not know how to make Claude recommend you, other than ecosystem presence.

[25:42] The Reddit playbook: comparisons, open ecosystems, ignore comment-spam

Tactics Gamma actually runs:

Own subreddit / community. How-to posts get cited ("how to export a presentation from Gamma" → PPT, PDF, images). Own-community content currently gets picked later; they are still building scale there.
Other subreddits: never "Gamma is great." Always "how is Gamma better than PowerPoint / Google Slides." Comparison is the unit of citation.
Open-ecosystem wedge. Because Gamma imports and exports, they can post in the PowerPoint subreddit as "I made this in Gamma and exported to PPT" — useful to people trying to align three boxes faster, not a brand dump. Analogy: if you are a CRM, import from Salesforce so you have a legitimate reason to be in that subreddit. FeatureOS equivalent: import from Canny/Productboard/Jira, export changelog/help center, so you belong in those tools' communities.
Starting up: other subs first. Long term: own brand subreddit. ChatGPT citation lag 30–45 days. Perplexity can pick a phrase the next day ("turn your Notion doc into Gamma"). Google AIO has been seen citing Reddit in a day on niche queries (host anecdote) — probably because the thread already sat at the top of Google.
Listicles work ("top 5 tools I've tested," paint one as favorable). Vanilla, everyone raises eyebrows, it still works.
Comments do not move discovery. If you are already cited, comments color sentiment. Comment-spamming will not put you in the citation set.
Negative sentiment is a liability. A peer prosumer company had one negative comment keep getting indexed for months; they could not kill it. Host theory: LLMs are biased toward picking up the bad comment. Treat it as a cost-of-doing-business risk, not a growth hack (they joke about black-hat "pay me $3k to delete my indexed dunk").
You choose which competitors to stand next to. Name the tools you want to be compared with (open, MCP, Zapier, export) and the ones you want to look closed. Do not accidentally AEO for a rival you would rather starve.
When the category is crowded, niche. Not "best CRM" — "best CRM for SMB / fintech / health tech / health-tech salespeople." Same as SEO topical authority. Competitor copies you → keep moving. That is growth.

Subreddit selection at Gamma's scale: ICP places (where a sales/marketing leader or founder hangs out — e.g. SaaS) and problem-universe places (PowerPoint, Canva). They skip tiny subs (thousands / low tens of thousands of members) because it cannot move Gamma-scale numbers. Host challenge: for AEO, a small sub might still get cited. Ravish: maybe, and he would rather put long-term energy into the brand sub; in practice he has never seen a citation from a subreddit he didn't already know existed in those two universes. Broad category (presentations) lives in huge subs; niche verticals (health tech) might need small subs because the model has to dig. Unproven.

Dirty plays they have seen: copycat sites cloning the Gamma landing page and name, showing up in paid faster than organic.

Other surfaces: X is not showing up in LLM citations despite Google surfacing tweets. LinkedIn: posts and articles, no engagement correlation. He would bet LinkedIn and Instagram start moving search more than Discord over 6–12 months (UGC plus a reputation signal). After Reddit: Quora (early, still cited, weaker), Medium (indexes well), Substack (weaker because gated).

The 4,000-drone SF stunt vs bottom-up AEO share a cultural root — let ideas breathe. He does not care about GTM vs "distribution" as vocabulary. Drive the right signups.

[36:53] The half-life of a citation

This is the part most AEO Twitter threads skip.

AEO is not one-and-done. It is an always-on engine. He does not have a formula for how long a citation lives. They log cites in a spreadsheet and watch citation share decay. Carbon-dating metaphor: there is a half-life. If it is ~3 months, you recapture the prompt when you fall off. Half-lives likely differ by engine. Profound (they are customers) is not giving them this; it is a feature-shaped hole.

Zero-sum: an LLM answer has a handful of Reddit references. You are competing with competitors the way you compete for top-3 Google. He could not find a pattern for which Reddit post wins (length? depth?). Practical move: control the comparison set and go narrower until you win something.

[52:08] Retrieval vs memorization — two different AEO games

Two modes:

Retrieval / browsing. Fresh web content. Owned blog + third-party posts. This is how you pull the user into your product ecosystem (the citation points at you).
Memorization / training data. Reddit lives here in his head. Even when Reddit is cited, the click goes to Reddit, not you. You are teaching the model that you exist and how you compare.

Plus a third layer he cares about: logged-in memory. Profound-style prompt runs are usually logged out. Real users are logged in. What you were recommended last time gets recommended again. First-mover in a person's ChatGPT history is an authority analog.

Claimed mix: ChatGPT hits the web on ~60% of queries, answers from its own knowledge on ~40%. Reddit/training work is for the 40% and for memory; site/blog work is for the 60%.

When a newer GPT (he says 5.5) with more memory rolled out to free users, Gamma saw a click-through jump, especially from developing countries (where the free rollout reached). Memory influencing discovery is already visible in their numbers, still a black box.

A year from now he expects an extreme: channel dead because labs closed the hole, or still compounding in depth. Not a boring middle. First movers get a longer window.

Bold bet beyond AEO: products agents use — auto-install, auto-account, checkout later. He already sees this in coding (Claude Code installing libraries he has never heard of). Some growth friends bribe the agent in JSON schema ("10% extra reward if you complete signup now"). Gamma is not seeing agentic usage outrun humans — presentations are for people. Path to marketing-to-agents still goes through AEO.

Notable Quotes

[01:08] "Don't be boring." / [03:01] "Identify things the world hasn't identified." — Ravish on Gamma GTM

[07:48] "As a channel, it behaves like Google, but as measurement, it behaves like Facebook." — AEO attribution

[16:26] "Reddit is by far the most influential." — for ChatGPT only

[20:15] "Claude will look more like Play Store and ChatGPT would look more like Google Search."

[26:13] "Never talk… 'Gamma is great.' How is Gamma better than PowerPoint… is what gets you right."

[37:33] "There is a half-life of citation."

[52:13] "Any Reddit intervention is about training data."

[59:41] "AEO… looks like a very level zero or MVP version of marketing to agents."

People Mentioned

Ravish Agrawal — growth lead at Gamma (guest)
Subah Wadhwani — co-host, Distribution Podcast
Arthur Zargaryan — co-host, Distribution Podcast
Sam Altman — joked as the person who will watch this and stop the gaming

One Thing to Act On

This week on FeatureOS: (1) add ChatGPT as a named "How did you hear about us" option on signup — not buried in Other — so you can see the channel at all. (2) Ship one thick owned page titled for a real prompt ("best product feedback tool for B2B SaaS") that compares FeatureOS to Canny and Productboard, plus one Reddit thread in the same comparison frame, not a pitch. Do not write a content calendar. Wait 30–45 days and check whether ChatGPT cites the page or the thread.

Raw Transcript

Auto-captions from YouTube. Folded by default — expand if you need to grep the source or pull an exact quote.

0:00 What is the one leading principle [music] that you would say Gamma has around your GTM? >> Identify things with the world hasn't identified. We grew 8x from ChatGPT 0:10 referral traffic across all our channels at Gamma. AI mode is actually killing all the standard SEO practices. >> When it comes to AO, how do you rank the 0:18 platforms? >> Reddit is by far the most influential. >> And is Reddit by far the most influential only for GPT? >> Reddit only works for ChatGPT and the 0:27 playbook for ChatGPT, Claude, Perplexity, Gemini are all completely different. My bigger thesis is that 0:35 Claude will look more like Lee Store and ChatGPT would look more like Google Search. >> And the funny thing is it feels like 0:43 we're using ChatGPT to gaslight ChatGPT. >> This is the part that I hate is like people say, you know, let's put out spam content. Firstly, you're assuming that if AI is writing a content, it's going 0:52 to be spam. If you're moving towards AGI and you know, if these models can do coding, build an app, of course they can write an article and of course they can 1:00 write a great article. >> What does the future look like when it comes to growth? >> I'm more bullish about 1:08 >> Welcome to the Distribution Podcast. Um, we're with Ravish Agrawal from Gamma and Gamma's been one of the fastest growing companies. So, we're really excited to 1:17 chat. We go really, really, really deep into AO because that's been one of the channels that's been the fastest growing for them. >> If you want to find out a bit more on 1:26 how they got 100 million users or at 100 million in ARR profitably, welcome to Distribution where we talk everything distribution. >> So, you lead the growth team at Gamma 1:34 and Gamma is killing PowerPoint right now, right? You're using AI and solving the blank page problem, helping teams 1:42 and individuals build like amazing presentations, right? I would say that in terms of the GTM world, Gamma is probably one of the leaders in 1:52 in tech, right? And you're playing in an even more difficult industry because you're going after the prosumer, not the enterprise deals, right? So, 2:00 there's a lot more creative involved. My My first question around this would be, what is the one leading principle that you would say Gamma has around your GTM? 2:08 What is the the North Star that you're always searching for when you are, you know, going to market, getting users on board on the system? 2:17 >> You know, two answers to this. I think one of the things that we discuss day in day out, or in we can is 2:25 also one of our company values is don't be boring. Right? So, as you go out there, you know, put put a content piece, you know, put anything out there, 2:34 you know, try to be you know, try to be not be boring. Right? Because your PowerPoint as a tool can be really boring, or presentations are generally 2:42 associated with the boring stuff. So, I think that that comes that comes a lot in our day-to-day discussions, you know, don't be boring. How can we go out and do this 2:52 in the most exciting and the fun way of doing things. Uh when it comes to doing growth, or you know, bringing new more 3:01 users, I think my personal approach has been to identify things with the world hasn't identified. 3:09 So, you know, what are the channels that currently is something that, you know, hasn't taken over, or there is a lot of things that you can do. We were very 3:17 early when it, you know, doing AEO, doing Reddit. We have just started doing ChatGPT ads. So, our like our principles 3:26 around there is can we, you know, can we explore a couple of things which the world is also exploring and understand, you know, if this suits us, if if if it 3:36 does, you know, what's the alpha, as you say, or you know, what's the method to go about this. And I think that that's one thing that we keep going after. 3:45 >> Not being boring is easier said than done. >> Of course, yeah. >> It's like it's one of the most difficult challenges ever, and you have to be so creative and do crazy stuff to like not 3:54 be boring often times. With that said, when it comes to Reddit and AEO, you mentioned you guys were very early on. 4:02 What did you guys do, and how did you catch on that AEO and Reddit specifically is, you know, where nobody is really doing it, and now it's become 4:11 common knowledge, everyone's looking into Reddit, and it's selling so easily as like an agency service, right? What made you guys look into that? And then, what's the 4:21 next paradigm of that? Is that ChatGPT ads or what? >> I will start with the next paradigm. I think next paradigm is is something that we all are figuring out together. I 4:29 still feel that there is a lot of depth in in AEO stuff as well. I think it's it's also about how platforms decide. I 4:38 think today, uh be it ChatGPT or Claude or Perplexity, like every one of them is giving a good organic discover like 4:47 they're giving chance for organic discovery to you. Put up a new website out there, you know, it gets really quickly crawled by any of these like you 4:56 know ChatGPT, Claude, or Perplexity. Like that's also the order, you know, the way they crawl and, you know, quickly index our site and do something. 5:05 So, it's also I think it's it's what Facebook maybe was early back in 2012, you know, before before they 5:13 before they decided to pull back on organic reach and, you know, ads ads became a thing. So, that that's one thing like we don't know 5:21 how these platforms because these platforms are also fast growing. Like when when you were early in Facebook days, you would still get 3 years time to, you know, build your organic 5:29 approach and continue to do it. Like today, because things keep changing every 3 months, you know, tomorrow if these platform leaders decide that, hey, 5:37 now we want to pull back on organic and, you know, push something else, that would be a completely different conversation. But for now, I do believe 5:45 that it's really deep and you know all of these AI you know, AI answer engines 5:53 have also prompted Google to you know, go more on AI mode and AI AI mode is actually killing all the standard SEO 6:01 practices like all all the growth peers that I talked to everybody is seeing click-through rates on their websites 6:09 falling and so on. So, everything is in a flux and I and that's where I believe that it's it will continue to be in a flux and that's where it will also be 6:18 very deep as you know, one of the biggest search engine goes into it. What led us to AO was I was I think just one 6:26 fine day I was just looking at Google Analytics to understand you know, where our traffic is coming from. I saw that you know, our traffic that is coming and 6:35 this was click-through traffic because Google Analytics cannot discover any other traffic. So, our traffic from ChatGPT kind of increase increase increase and then it decrease somewhere 6:46 in the middle of last year. So, it was very surprising to me that in that period of in like the traffic increasing we grew 6:55 8x from ChatGPT referred traffic across all our channels at Gamma. You know, we have never seen that growth. So, that kind of caught my eye. It was very small 7:04 part still like it was less than I would even say 1% of our part. So, did not really felt like you know, it's it's going to change something but then we 7:12 did two things. We started uh you know, we started tracking it in a different way. We like we are a SaaS platform consumer company. So, one of 7:22 the methods that we rely on is how did you hear about us right? How did you hear about us? And previously people were filling that there was an other option 7:30 and people would say, "Hey, I discovered you from ChatGPT or Perplexity." but we brought that in at the face and a lot more people started answering. So, we 7:40 realized that you know, while the click-through traffic is less or it's decreasing, people who are self-reporting continues to increase. 7:48 And I tell this to, you know, to every people. I don't like like as a channel, it behaves like Google, but as 7:57 measurement, it behaves like Facebook. Like, you know, the whole AEO stuff. If you want to go ahead, you would execute all the playbooks that 8:05 is required for you to come up in Google search, but if you're measuring, you should measure it the way you are measuring Facebook attribution, which is 8:13 a lot of view through. At that time, the common knowledge was, you know, LLMs were picking picking up from Reddit. So, we started doing Reddit. And now now 8:23 Reddit is, of course, one part, but LLMs have started to pick from LinkedIn, from YouTube. So, you can do a lot of stuff on the existing channels. And you might 8:33 not get a lot of traffic from those channels, but you would be building you would be building the core AEO engine, and you would be laying the foundation. 8:41 So, a couple of channels that now we invest on we primarily invest more from an AEO perspective than, you know, that 8:49 particular channel perspective. >> That's what I was I was going to ask, right? So, when you're building out for AEO, right? Let's say we're placing a piece of content on YouTube, LinkedIn, 8:59 what percentage of the thinking is I'm making this for a human to read versus I'm making this for a crawler to read? And then does it feel like sometimes maybe we're gaming the engines? 9:08 >> It does feel that we are gaming the engine, and to be honest, it it also don't feel good because, you know, like if if it's gaming, then, you know, you are going to lose the advantage at some 9:17 point of time. >> They'll notice that you're gaming it, and then they'll un-game it. >> But everyone's gaming it, so it's like game theory, right? You got to game it, or you're going to fall fall behind. 9:25 >> Yeah, exactly. So, yes, it's just kind of gaming. I think for a couple of channels, like we have initiatives where we say that this initiative is primarily 9:35 for AEO, and then like we would see that initiative as 80% from a very agent first or a crawler first angle. So we 9:43 would want to have for like let's take YouTube as an example, right? Whatever our headings would be is, you know, we'll we'll pick from prompts or phrases that 9:53 are being used rather than, you know, clickbait headline, you know, which will make humans click. So uh And so generally it's like I like all 10:03 the AO stuff when we do for a certain channel we take an 80/20 approach where is 80% is it's for the agents, for the 10:11 crawlers, 20% is it for humans? Like if a human discover it they should not just feel that what the hell this is, right? Like it should 10:19 still be digestible. >> Makes sense. And I guess a question on that one is, right? Like if we look at the market Google I would say is probably the most sophisticated about picking up signals, 10:28 right? That's why they built the the whole way SEO works is based on backlinks, right? Because they show authority. Right? Are you feeling that 10:36 GPT Claude they are looking at authority other than the platform you posted on? So for example are they more heavily indexing things if 10:44 it has more views, more upvotes, more comments? Is it looking at those kinds of signals or is it just really looking at the the content itself in almost a 10:52 vacuum? >> In terms of websites, I don't think they're really looking at authority. So you know, uh websites as as old as 11:02 Forbes and maybe, you know, your friend's startup. I think from an AO perspective it's the same. It's more about content and I think your friend's 11:11 startup would have more AI first content than what Forbes would have. Of course, Forbes you know, would come because they have authority and so on. But so I think 11:20 in terms of website, I think authority is a very very level playing field right now. I think everybody is starting from the same authority. 11:29 On other platforms, to be honest, I haven't seen a pattern. Like I've also tried to understand this, right? Like I've looked at our YouTube citations, 11:37 right? We're getting cited for, you know, for a video that has 400k views and also a video that has 18k views. So, 11:47 like there there's no pattern as such. I think the closest pattern that I've been able to figure out is that are you being 11:56 talked by someone else. Uh and then are you being talked in comparison to something other? Like if 12:04 it's going to be like it's it's just a whole listicle playbook, >> than X competitor? >> Yeah, exactly. It's the it's the whole listicle playbook, you know, going on 12:12 video, but are you getting talked in comparison with someone or like the video or the content should not be just 12:20 about you. I think that's the that's the broader direction that I've been able to pick up so far. >> Okay. And just around that one again, 12:28 one more question is how long do you think this is going to last, right? Because uh what is the playbook now? Does that mean just like spam content basically or 12:36 at least, you know, put out as much content as you can? And then do you think the the engines are going to start to count 12:45 like fight that basically and start looking at views and likes and so and so forth. And how much time do we have basically? >> To be honest, I don't know how much time 12:53 we have. I do feel that that's going to change. I think Google's update that came on May 21st, the search update, it did penalize a lot of website that was, 13:02 you know, just posting content out there. So, I think Google is definitely taking action. ChatGPT pulls most of its 13:11 I think queries from Bing. So, you know, Bing Bing ends up acting differently. On the content piece, I won't like, you know, 13:19 and this is this is the part that I hate is like people say, you know, let's put out spam content out there. I think there are two two wrong assumptions there. First, we are 13:27 assuming that if AI is writing a content, it's going to be spam. Right? I don't think so. I think models have become better today and with Fable 5 or 13:36 even Opus 4.8, and you can actually write content which is much much better than human, and you can actually, you know, do a much more research content. 13:45 So, I think that's the that's the first piece that I really hate about like whenever, you know, like I also talk about AI, people roll their eyes and they're like, "Oh, spam content." So, 13:54 it's not spam, right? You have the, you know, if we are moving towards AGI and, you know, if these models can do coding, 14:02 build an app, of course they can write an article, and of course they can write a great article. And I think the second one is what do you include in that 14:10 article, right? You can on one side you can say that, "Hey, this is a glass of water and you know, this is this, this is that and 14:18 so on." Or you can go deeper on, you know, why is this steel glass better than a plastic glass, right? And that is that is upon you to decide like how deep 14:28 you want that content to be, how meaningful, you know, content to be. You can always write, you know, five health benefits of 14:36 using a steel glass as compared, you know, just saying that, "Hey, this is great, this is great." over again and again. So, I think that's the second I 14:45 that's the second idea that the content doesn't have to be thin, right? All people assume that if AI is writing content, you know, let's just give it five keywords and stuff it and let it 14:53 repeat. The content can actually be very thick and you know, and that's what I think LLMs 15:01 would appreciate. I'm sure that, you know, as they scrape the websites, like as LLMs are training scraping website, 15:09 I'm sure there is someone who is looking at all of this index of content and, you know, just like Google at some point in time we will have, you know, authority and 15:18 relevancy and so on. But till then it's an open field and, you know, we we all get to take our shot. >> It'll all be It's going to be ever 15:27 evolving in that sense. >> Exactly, yes. >> When it comes to The funny thing is it feels like we're using uh we're using, you know, ChatGPT to 15:35 gaslight ChatGPT. >> ChatGPT, exactly. >> [laughter] >> You know, it's like which model does it better? >> It's It's like It's like Yes, please write uh content that is going to help 15:43 fool you thinking that my product is more, you know, important, you know. The best part is if, you know, you're not fooling and you're actually serious, but >> Yeah. 15:51 >> In a marketer's defense, the labs are also using their own models to train next model. So, I don't think >> Exactly, it's all it's all like that. 15:59 When it comes to like AEO, right? And what's been working for you guys? How do you rank the platforms? Like there's been Reddit, there's been YouTube. YouTube I know is like heavily picked up 16:08 on when it comes to their transcripts, right? And then there's of course LinkedIn content and then other blog {slash} articles, whatever. If you had 16:16 to rank them in order of how you guys have built out your AEO strategy, what would you say has been most influential? >> I think Reddit is by far the most 16:26 influential. >> And is Reddit by far the most influential only for GPT or for other elements of >> Just for GPT. >> Just for GPT. >> Yeah, and and that's a very good topic. I will come back to that as well. 16:35 Yeah, they have a data deal, but I think people also keep saying good AEO good SEO is, you know, good AEO. We will touch that. 16:43 But, I think with ChatGPT Reddit is really great. The second one I've realized is your own website, right? The kind of content that you can do on your 16:52 website. I think that gets picked up faster, that gets cited more, and you also, of course, have con you know, control over how do you want your 17:00 product to talk about, how do you want it to talk about competitors. The third that I have seen is generally, as a category, is other people's websites. 17:10 So, is it blog, is it PR, is it uh review? I think that's the third most important 17:18 cluster and to be honest, I'm always always switching between second and third, right? Second could also be, you know, how other people are talking about 17:27 you on, you know, the complete text or or internet. I think that it it always between your website and all the other websites that are talking to you. Uh it 17:37 always switches between second and third. After that, I think is YouTube. After that, I think is LinkedIn. LinkedIn is also very thin right now. 17:47 Like you don't get a lot of citations and to be honest, I haven't been able to see, you know, what gets cited. It's more bizarre than what YouTube is. I 17:57 think Reddit is much more predictable. I think people have spent 6 months across the world of, you know, important time of their life to understand, you know, 18:05 what's what gets cited on Reddit. And I think as as we continue to go ahead, I think we'll start seeing more patterns on what's happening on YouTube, what's 18:14 happening on LinkedIn, what's happening on other platform. Coming back to, you know, what you pointed out like Reddit and ChatGPT, I think 18:22 Reddit only works for ChatGPT and the playbook for ChatGPT, Claude, Perplexity, Gemini or Google AI 18:32 overviews are all completely different. They of course have some overlap. In my heuristic observations, I have seen that there is 18:42 generally 30 to 40% of overlap across all four, uh you know, LLM engines, but 60% is something that you have to do it 18:51 different. And I think if you want to make sure that you are coming top in Google AI overviews, then you have to play by SEO rules. And that is where, you know, good, uh you know, good SEO is 19:01 good AEO makes sense. Just in the whole Google context in the ChatGPT context, as I said, 40 20 to 40% of the playbook remains same, but then there are new 19:10 things that you can try. >> Like for GPT, it's Reddit. For Perplexity, it gives more authority to third-party >> Exactly. Right. 19:18 >> What about Claude? Do you know that? >> Claude doesn't cite. I think the best way so Claude's like and then there have been a lot of studies. It's It's all out 19:26 there on internet. I think the the number of citations that Claude does is I think 5x lesser than ChatGPT and so 19:34 on. The best way to be I think to crack Claude is being Claude's ecosystem, right? So create an MCP server, list your app there, and so on. So 19:45 And again, that's that's slow, right? Of course, you know, like Claude Anthropic needs to approve it. At least the MCP servers have evolved like before like 19:56 3 months before you would have to download a zip file, do a local installation. Today, you can just paste the link and a remote installation through OAuth can, you know, quickly set 20:04 up. It's just like, you know, connecting Facebook in olden days with any of your third-party website. And I think it will it will evolve more. My My bigger thesis 20:15 is that Claude will look more like Play Store and ChatGPT would look more like Google Search. 20:23 >> I think that's a very good analogy. I think we should, for example, just make an MCP for something or something kind of product we do just for the sake 20:32 of having a Claude pick us up as well. Can you rate um and review >> You can't do that. >> It will have to do it 100%. 20:40 >> It will have to do >> Exactly. I think a lot of this lot of this will evolve, but I think the biggest constraint with MCPs today is that you still need to acquire user on 20:48 your own, right? Like Like if you have an MCP hosted on your website, Claude will not pick you. Like it it won't pick you saying, "Oh, this is MCP. Why don't 20:56 you try this?" You still have to do do that heavy lifting where you have to tell your user, "Hey, we have an MCP connected with Claude. You You can 21:04 continue using it and so on." Which I think gets solved and that's why I'm saying it will evolve as Play Store where, you know, like earlier in the app, you know, the 21:13 discovery happening through a Play Store or App Store wasn't, you know, the biggest, right? You would have to ask people to download the app and then as mobile took over, right? 21:23 People started natively searching on these places, right? You would go to Play Store and and search for, you know, a maybe a weather app, a travel app and, 21:31 you know, some mobile native companies came in, right? We we know Robinhood Robinhood is one of those examples, you know, very mobile native that that got 21:39 picked up. So, I think and that's where I think these new category of products will come in. When will they come in? 21:47 How will they get adopted is is what we all have to see. >> I've got a another question for Gamma. Um 21:56 when it comes to um acquiring customers, how much of it is AEO right now? >> It's one of our top channels. Right. So, 22:05 like it it is a channel which gets talked about and it's a channel that we can't ignore. >> It is top of mind. >> Top of mind, of course top of mind. 22:14 >> what is the directional growth of that channel, right? >> Faster than any other. >> Oh, okay. >> And then when it comes to this AEO, right? How is that AEO broken down? So, 22:23 it's one of the fastest channels for Gamma. You guys are one of the fastest growing consumer companies. So, then which LLMs are >> ChatGPT 22:31 >> what percentage? >> I think it's north of 90% I would say. >> Because every normal user just types on 22:39 ChatGPT. They don't even know Claude. >> Yeah. So, I think >> consumers, right? >> Yeah. We have an MCP connector and people use our MCP connector. We have 22:48 also launched it. So, we have seen that once MCP connector gets built into Claude, like we have seen good amount of 22:56 usage, right? Because you say to Claude, like you have thought about something with Claude and then know, create a PPT and suddenly a Gamma PPT is created, right? Because you are 23:05 there in that ecosystem now. But I think a lot of new user acquisition is still happening from ChatGPT. Uh Perplexity is 23:13 is there. Google, I think we haven't been very I would say like top class when it comes to cracking the Google AI 23:22 overviews. Uh I I also feel that I haven't paid a lot of attention because I wasn't expecting that you know, it would move so fast 23:31 like as you know, the whole AI mode has come in. Now now I would want to invest more, but yeah, ChatGPT 23:39 is way ahead. >> So so for prosumer brands, right? How would you rank the importance of the different LLMs, right? Number one is 23:47 GPT, what would be number two? >> I would say number two I would still bet on Google AI overviews rather than 23:55 Perplexity. I think because if you do that correctly, you also have a uh you know, secondary SEO factor playing in. 24:03 And again, it's Google, right? So you still get a lot of distribution out out there. So it's ChatGPT, then I would say Google AI 24:11 overviews, then then maybe I would say Perplexity, and then I would say Claude because as I said, right? Claude will not help you acquire new customers. It 24:20 can help you really retain customers or >> Mhm. What if you're playing in the enterprise space, right? Um how would you think of the prior prioritizing? I I 24:29 know we're we're guessing here, right? Because >> Very interesting. I think enterprise space I would flip it and maybe I would say the first should be Claude because 24:37 you know, yeah, because I keep talking every day and if I'm searching for a tool, I'm searching for a tool there. So that's one and I have no idea how to, 24:46 you know, how to make sure that Claude is recommending you. The second I would say is Co-pilot, right? Co-pilot being I 24:56 think that's also a very interesting piece. I haven't I haven't gone much deeper into that space, but I have seen a lot 25:05 of, you know, traffic coming to us from Copilot or, you know, under the whole Bing umbrella. So, that that is big. 25:13 >> And those are like massive companies, probably. >> Exactly. So, I think you get you get that distribution advantage and then 25:21 then I think everything else is the same. I would say maybe ChatGPT, Google Google AI overviews, and then Perplexity maybe would be last in that 25:30 ranking. >> Because you guys are getting so much phenomenal growth from ChatGPT particularly and GPT is one of the most important sources is Reddit. What 25:39 exactly have you guys been doing? Like what is the Reddit playbook then? >> I think one we we have been doing different things on Reddit. We have our 25:47 own community where, you know, people talk about they ask questions and we have seen that, you know, our own community also get cited. Uh like, for 25:55 example, how to export a presentation from Gamma, right? A lot of people are asking questions. Like you can export it to PowerPoint, you can export it to PDF, 26:03 as images. So, that that's one good play. And as I said, like if you're participating in other communities, make sure that you are 26:13 talking in besides your other competitors or other players. So, never talk you know, the best way 26:22 don't talk about hey, Gamma is great, right? How is Gamma better than PowerPoint or how is Gamma better than Google Slides is what gets you 26:31 gets you right with SEO from a Reddit perspective. If you're just doing Reddit, then, you know, there are different approaches. I think we 26:40 take a mixed approach to Reddit where our overall idea is to be on Reddit so that we can be up on search results. 26:49 >> How would you say is the like waiting of picking up things which are in your own subreddit versus on other subreddits. 26:58 >> Same question. >> Yes, I still feel it depends upon the scale. We are we are still in the phase of building our own subreddit. So, our sub 27:08 like content from our subreddit gets picked later. What I have seen is that generally it takes anywhere between 30 to 45 days for a new content to get 27:18 picked up as citation from ChatGPT. I think Perplexity can pick up as quick as the next day. If you are specifically, you know, searching that phrase. For 27:26 example, turn your Notion doc into Gamma. If you if you're there, I think Perplexity will quickly pick that up. If you're starting up, I would say it's 27:34 more important to be on other subreddits as compared to your. Longer, I think the longer term potential is building your 27:42 own subreddit as compared to participating on that. >> I've seen a Google AI overview also cite in one day from Reddit. 27:50 >> Oh, okay. Very interesting. I haven't seen that yet. >> It was a very niche query, but then I was shocked that Google picked something up in a day. 27:58 >> Got it. I think with niche query, it's also that because it ends up showing on the search results at the top and that's why that's why it gets picked up. >> I think so, too. 28:07 >> Like the longer you go on the query, the you know, >> It was a specific one, but I am like, okay, it's getting faster and faster, >> [clears throat] >> the cycles. >> What would you say or like let's let's 28:16 go like really tactical with Reddit, right? I'm I'm going I'm going to ask some questions you might know how to answer. So, So, the first one there is how do we 28:24 actually post on other subreddits without getting taken down because Reddit is a bit of a >> Yeah, it's >> You know, the the moderators they are 28:32 >> yeah. I think Think it from a very, you know, very value-oriented perspective. Like I think what we have seen is as Gamma, you 28:42 can import a lot of things to Gamma and you can export a lot of things from Gamma, right? So, are not a walled garden. And that helps us talk in 28:50 compare like we can talk on a lot of other subreddits, right? Like, "Hey, I created a PowerPoint, but I actually created in Gamma and then exported it to 28:59 PowerPoint." That's one kind of content that you know you're not promoting Gamma, but you are saying that, "Hey, you know, Gamma is a good and a faster 29:07 way to make a PowerPoint quickly." And then, you know, this content can get picked up because if you go into a PowerPoint subreddit, people are trying 29:16 to understand, you know, what's the quickest way to make PowerPoint? What's the quickest way to, you know, align three boxes, right? And that's that's one way to, you know, that that's a 29:26 wedge for us as a tool to be a part of that community. So, I would say that, you know, make sure that your tool is open you like one one good thing about 29:36 us is our tool is open ecosystem, but can your tool support an open ecosystem? And then can you plug onto those conversations 29:44 that is happening around those those ecosystem? Like for I think Salesforce or Salesforce could be good, right? If you're if you're 29:52 building a CRM company and you can import from Salesforce, then it's it's really a good place to, you know, be in that subreddit. 30:01 >> And are you seeing things like um literalistic type comparisons working on Reddit or are they >> Exactly. Like yeah, it's 30:09 >> top top five tools I've tested and then you put top five tools I've tested and then you list out all the different tools, you know, painting one of them as 30:18 being, you know, favorable. And then you also kind of put that in the discussion. Do you want people to discuss that as well in the comments? Are comments picked up as well? 30:26 >> I haven't seen comments having a meaningful impact in changing the 30:35 direction of your discovery. So, let's say if you're already getting discovered via Reddit or you are in ChatGPT. I think comments are more for sentiment, I 30:45 would say. And but if you're if you're thinking that hey, I would just keep commenting and I would, you know, end up being on the top or, you know, my 30:53 ChatGPT citations would increase, I don't think that would work. I think like my mental model has been the comments are the good sentiment 31:02 sentiment way of, you know, looking looking at that ChatGPT answer. And to your question, yes, I think that 31:10 is like a pure play vanilla playbook. Everybody knows it, right? People raise eyebrows when you say that hey, you are doing it, but for 31:19 the good or for the bad, it works and, you know, it it gets you there. >> You know what's funny? I've seen another prosumer company, I can't name, but they 31:28 were also doing like a lot of Reddit marketing. And one negative sentiment style comment just kept getting indexed. 31:36 And they could not get rid of that for several months. They did everything. So, when it comes to Reddit, one learning is that like look, good sentiment is of course the goal, 31:45 but bad sentiment you have to avoid it all cost because if there's a bad sentiment, LLMs have bias towards picking that up. >> You think so? >> And showing them. I think so. 31:53 >> Oh, that's interesting. Like I I I always knew that, you know, comments are playing some role in sentiments, but I've never gone gone so deeply, you 32:02 know, thinking that, you know, negative sentiment would pick up more or >> I That's a theory. I don't like I can't tell you whether or not >> We're all theorizing here. 32:10 >> Yeah, you know. >> Like poking the black box that is an LLM. >> And then they did so much to get that comment removed and then eventually >> Guys, new business idea. It's black hat. 32:19 Go and comment on any like prosumer Reddit post negative comments and then reach out once it gets indexed, reach out to the company and charge like 32:28 $3,000 to remove your comment. >> Oh god, bro. You know, you you are getting canceled now. >> Good. We're here to get canceled. 32:36 >> [laughter] >> And an interesting one is that you know, um, I guess let's Reddit getting indexed, you know, from stuff you post on Reddit 32:45 is it is almost a zero-sum game because there's only, you know, four five um, references, right? In an LLM query. 32:54 So, you are effectively competing always against your competitors, right? On Reddit. It's similar to you know, SEO because there's only top 33:03 three results, right? Then you're being always compared on backlinks, basically. So, so what is the thing that the AI agent the the AI, you know, the 33:12 LLM will look at when choosing which one of the queries to to include in their citation from Reddit, right? What is ChatGPT looking 33:20 at? Are they looking at how extensive the article is? What's your thesis on this, right? How do you stand out from the competitors? >> I don't have a thesis on it. Like I I 33:30 did think about it, but I couldn't see any patterns. I think the only the only thing that I could deduce from it is 33:39 that like choose the competitors that you would want to talk about. And I think that's the 33:49 best way to you know, still be there, but at the same time, you know, not really get uh, 33:58 bogged down by a competitor, right? So, for for for example, you can like you can always talk about that hey our tool 34:06 is open, right? You can use MCP, you can connect it to Zapier, you can export, and so on. But hey, this XYZ presentation tool doesn't let you do all 34:14 of that. So, like you can just choose, you know, who who do you want to place yourself against? And then then then 34:22 accordingly you can play that game. >> To translate that a bit, I think what's been happening on Reddit is that for a lot of companies, you don't want to do 34:31 AEO for them. Right? So, by mentioning competitors you don't want to do act in their favor. So, it's very it's like a game. >> But my perspective on this one is more 34:41 around the fact that, you know, let's say let's take another like companies example. Let's take like CRM companies, right? Let's say, you know, HubSpot, 34:50 Salesforce, um Sugar, right? They're all doing Reddit marketing, right? And they're all putting out a lot of content there. 34:59 How does And then, you know, there's a general query, right? On ChatGPT. What's the best CRM system out there? How is ChatGPT going to decide which one 35:08 of the Reddit posts is going to pull? Right? Because Google decides based on backlinks, right? Maybe also not cons- not anymore really concentration of 35:16 keywords on the page, more like the length of the page, you know, the kind of the depth of the content. What are these these like LLMs ChatGPT looking at when they're deciding what to index and 35:24 cite on Reddit? Because the thing is no one is going to be doing this in a vacuum, right? Like, I'm sure you're going to have your competitors, you know, competing with you in this domain 35:32 as well, right? So, how do we compete with them? How do we beat them? >> As I said, I I really don't have a very strong thesis on it, but I would say is 35:40 that what you can do is instead of saying what is the best CRM, what is the best CRM for SMB? What is the best CRM for fintech? What 35:50 is the best CRM for health tech? I think can you go there and that is where you reduce competition. It's it's also like the SEO playbook, right? You cannot come 36:00 at the top in best CRM. So, can you start by going you know, going slightly I would say niche, like you know, best CRM for health care and then you might 36:09 end up ranking. You You will get some topical authority and relevance from there and then you can try to take another shot at, you know, like being up 36:17 on best CRM results. I think it I would think it is the same way. Like, I would maybe like if if I'm doing CRM and you 36:26 know, it's a competitive space, right? How much can I niche down? Um And can then I do like best CRM for 36:35 sales people in health tech. >> Okay. What happens then when your competitor starts niching down and just following you? >> Keep moving. That's what growth is, right? >> That's what growth is. You got to look 36:43 for the alpha. And And actually on that one is how similar is it to SEO where you have to consistently be putting out content, right? Uh is it a 36:51 one-time-and-done thing? Or or [snorts] you just got to be always >> It is a ever-on-running engine. You have to keep keep doing it. I still don't 36:59 have a formula on, you know, uh what is the lifetime of a citation, right? Like I've been Like I've been thinking a lot about, you know, what is the lifetime of a citation. We have also 37:08 We are also tracking a couple of citations. We know that when we get cited, but we don't know when we get removed from citations or when does that 37:17 citation, uh you know, loses importance. Like it's like I don't know if you have if you remember, you know, early in college or 37:25 in science school, you would you would have read about carbon dating, right? There's a half-life of everything. So, I think there is a Like my bigger theory is 37:33 there is a half-life of citation. And I I keep thinking about it, what is the half-life of citation? Can I say that, hey, my citation will only exist on 37:42 internet for 3 months? And that's like whatever value I can get in that 3 months is is all that I can get, and it won't matter, which means then 37:51 every 3 months I will have to come back and make sure that as I get removed from that particular phrase, keywords, 37:59 prompt, right? How do I make sure that I'm ranking again? >> Again. It's So, it's an always-on engine. I think also every LLM is 38:07 different when it comes to half-lives. So, it's like up for the grabs. What tools are Sorry, what tools are you using to track Is Pro- Like you guys are 38:15 ProFund customers? Is that helping you track the half-life? >> No, so what we It It isn't like what we are doing is we 38:23 are just manually tracking on a Excel sheet. You know, like this got cited here and then we are trying to look at its citation share and do we see a 38:34 change in citation say share somewhere down the line. >> So, product idea or feature idea for profile, right? 38:42 >> right. >> Half-life of a citation. >> Exactly. >> And I would also like the interesting thing about what we're discussing here is we're always discussing the frontier, 38:51 right? Of marketing GTM and all of these different channels. So, if anyone has insights, like please DM us. Please write in the comments, right? Let's 39:00 Let's talk about this because we're trying to learn as well, right? We're putting our thoughts out there. >> And you guys are hiding. >> We are hiding. I am hiding. Like Like if you have a alpha on Reddit, you know, 39:08 just just write to me. You can find me on Twitter. You can find on me LinkedIn. Like give me give me, you know, your hottest take. Tell me, you know, where I 39:17 am wrong. Where have you seen something else? And yes, you know, we would love to have you on the team. >> We should make a group chat called uh marketing alphas or GTM alphas. 39:26 >> Another question is getting more tactical in Reddit. How do you guys decide which subreddits is worth investing in? Right? Like which 39:34 subreddits do you guys seed any content on and then what's the good way to seed that content? How do you talk about it? And 39:43 how much do followers of a subreddit or like weekly active visitors of a subreddit matter? >> Because Gamma operate at a certain 39:51 scale, the last question is very easy for us. Like we do not go at any subreddit like which has, for example, thousands or 40:00 10,000s of active users or members, right? Because we know that that is not going to impact us at the scale that we are 40:08 playing. So, but if you are starting up just now, I think, you know, that's a fair game, too. We have We have thought about subreddits from two 40:19 perspectives. One is a very ICP perspective, right? Where is Where is, you know, a sales leader, a marketing leader, a founder going and, 40:28 you know, looking at the next content content for themselves. And then the other is I would say the problem 40:36 universe, for example, where are decks being talked about or, you know, creative generation being talked about and 40:44 uh So, for example, in first case, the subreddit would be you know, SaaS, right? In the second 40:53 case, it could be PowerPoint or Canva. So, these these are the two approaches that we have taken on discovering, you know, which subreddit to go after. >> Could I challenge something you said, 41:01 however? You said that you're not going to be looking at the smaller subreddits, right? >> Yes. >> Um but if this is a primarily AEO 41:10 um action, right? We're thinking about AEO here. Isn't AEO going to still pick you up even if it's a smaller subreddit? >> I think it will. 41:19 Um and it might makes, you know, it might make sense from a longer term, but then I would want to focus on our own brand subreddit on the longer term. So, I 41:28 won't disregard that idea. Yes, I think we should spend some time on doing smaller subreddits. In my view of all the citations that I have seen, I have 41:38 never seen a citation from a subreddit which I would not know existed in this 41:47 two universe. And I don't think Reddit communities also don't grow like Instagram channels, right? They don't go viral suddenly the 41:56 next month or 3 months. So, I think things would like they grow step by step, month by month, and you know, I think you will dis like if you're too 42:05 deep into Reddit, if you're doing it day in day out, you will rediscover that, you know, one of these one of these places is where the con conversation will have is happening and you will go 42:14 there. So, I would like my answer to that challenge would be I think there would be a certain threshold after which it will become important and 42:22 >> So, it's not going to even pick it up if it's in a small So, probably for everyone that applies actually. It's not just the even bigger companies. >> I think we are also a very big category. 42:30 So, for example, presentation is a very big category. So, you know, any anything that we talk about also has a very big subreddit and it's a very 42:38 horizontal thing, right? People are talking about in our PowerPoint, Canva, which again has millions of you know, millions and 10 millions of users. So, these subreddits are already 42:47 bigger. But, I think it would be an interesting play to see how you know, things in health tech are getting picked up. Are they getting picked up from smaller subreddits? 42:55 >> Maybe the way the AO AO works and these LLMs do it is if it's a broad topic, they're going to be able to find it in the bigger subreddits. So, 43:03 they don't bother digging. Maybe maybe if it's a more niche topic, then they have to actually start digging. Maybe it's a question of computer efficiency basically for this. 43:12 >> Could be. >> Have you guys seen any dirty plays by competitors? >> We do have seen, you know, some like dirty plays. For example, we have seen a 43:20 lot of copycats coming up, right? So, sometimes people like sometimes you know, people will just spin up a website 43:28 that exactly looks like Gamma, you know, has a similar landing page like Gamma. They would also use our name and they might end up being on the top and 43:37 sometimes users users would confuse you know, okay, you know, this is Gamma, this is Gamma. Both are Gamma. Which one 43:45 is the right Gamma? So, we have seen that happen. And how did they get indexed? They do, right? Like some of them also, you know, just build the website and run 43:53 ads, right? So, so that is also happening. Yes, we are figuring out, you know, how to tackle 44:01 that. They they generally don't show up on organic, but they end up showing quicker on paid. >> You know, when we're talking about the 44:09 different places that content is getting picked up, you mentioned LinkedIn. Um, are we looking at LinkedIn posts, LinkedIn articles? And then the next 44:17 question from that is, how is X being indexed by the LLMs? I guess like Grok is probably indexing X quite a bit, but not many people use Grok, right? 44:26 >> Let me go back and look how X is indexing. I have never noticed X >> Same. >> getting indexed. >> I haven't seen X. >> Yeah, but I like I think you're correct, 44:35 right? Because if you search anything on Google, X do come at the top. So, X ideally should get indexed. X content should get indexed. >> I haven't seen it. That's the thing. 44:43 >> No, but if you search a news, you would see X coming up, right? >> But not LLMs. >> Yeah, I haven't seen that, but I would I would love to maybe go back and check 44:52 because I've never thought about it. It's a very good thing to go back and see in my console. But I've I've never seen X or I've never heard about X. I 45:00 think on LinkedIn, it is still difficult to say. I've seen both getting indexed articles as well as posts. I have also seen them indexed like I've 45:10 seen these pieces of contents I either being an article or a post, you know, not coming from an influencer 45:19 or, you know, not having them one of the most engagement. Like you would you would feel that, you know, something that is getting indexed might have, you 45:27 know, 500,000 likes. People are commenting on it. But I even haven't seen that pattern. I also think it's too early for, you know, like in the earlier 45:35 phase of LinkedIn content getting generated. So, there is there is yet something to get evolved there. And I would love to understand from the 45:43 audience if they have seen anything particularly interesting with LinkedIn content. >> What do you think of Discord? Will Discord ever be picked up? I feel 45:52 like it's one of those again, what companies might get into in the future because Reddit is what? Discussions, online community discussions. Discord is 46:01 one of the other hubs for these discussions. So, as LLM try to get more data deals, etc. Do you think Discord can ever be indexed? >> I think before Discord, Instagram will 46:10 get indexed. And it's already getting indexed. Yeah, right? Like Instagram is has started coming on search results. And I think it 46:18 will end up getting indexed before also Discord. Like my biggest bet would be on LinkedIn and 46:26 Instagram. Um you know, influencing search results over the next 6 to 12 months because that's also very 46:34 like I see it from a very user-generated content perspective. Like, you know, where is, you know, new user-generated content coming up, 46:43 you know, get getting some reputation in some way. Like I'm like a Discord content doesn't get reputation. Like uh 46:52 you know, you don't have a I know you have a vote, you have a comment, you have reactions, but still it's not a very structured content to pull from. 47:00 But I do feel that Instagram is that like, you know, you you can explain a you can explain a new concept and suddenly it gets 20K likes. So, both 47:09 LinkedIn and Instagram has those classic user-generated content plus a reputation of that 47:17 content. And I think that will get picked up. And that's what also happened with Reddit, right? Reddit is a place where you get the most unfiltered 47:26 content, most true content. So, as this universe gets expanded, I think people would look for more content which is not 47:36 available out there. And where is that content coming from? >> What about Quora? >> Quora is getting Quora is still getting indexed. >> Yeah. >> I think it was it was the big one in the 47:45 start before even Reddit, wasn't it? >> Yeah, it I think it Quora got left behind, but it's still important. Like if you have exhausted Reddit, you should move to Quora. Medium 47:54 is another like Medium >> Medium as well, yeah. >> Medium is also getting indexed a lot. >> What about Substack? >> Substack Substack is getting is also 48:04 getting indexed, but I think with Substack it's also got a It is a run by journalists or, you know, people who are in the media. So, they they generally 48:13 prefer it gated than not gated. So, I think Substack has more gated content. That is why it's not picking up, but 48:21 Medium is picking up. So So is Quora. Like, I haven't seen Quora in influencing till now, but yes, it definitely gets cited. 48:29 >> Okay, interesting. And I mean, if we think about like, you know, this PLG motion, right? This word-of-mouth motion. I think the tool out there which was 48:38 the most influenced by was probably Calendly. Do you Do you remember when that just grew insane? I think they went to like $300 million in revenue. >> Powered by Calendly. >> Yeah, powered by Calendly or, you know, 48:46 you >> And you guys are powered by Gamma, of course, like in the >> Yeah, yeah, yeah. Made with Gamma. >> Made with Gamma. >> Made with Gamma. But yeah, like Calendly was one of those products that, you 48:53 know, every time you send your Calendly to someone. Now it's become a verb. Do you have a Calendly? >> Exactly. Yeah, yes, yes. >> So, it's really interesting if you can get that PLG in a product. 49:01 >> The other thing is, since most of your growth has been AEO-related in the last few months especially, or, you know, more bottom-up motions, what was the 49:10 campaign where you guys flew 4,000 drones over San Francisco? Like, how do you guys choose to do stunts like that 49:18 versus, you know, the company that you guys are going bottom-up? >> I think one thing that I really like about Gamma, and if you look at both of 49:26 these things, the underlying factor is you let ideas breathe. You let people You let people do things that 49:36 they are most passionate about, and then you trust that the best works get done. So, I think when the idea 49:44 came in first, I think it sounded very bizarre, but, you know, I think uh you know, And founders, the leadership has done really a great job uh in you 49:54 know letting some of these ideas breathe and you know some some of them see the light of the day and you know get talked about. We we also have so so many more 50:03 crazy ideas coming. But yeah, at the center of it is that you know can you let ideas breathe? Can you let ideas improve and then can you trust the team 50:12 to throw something out and then see what happens? >> How do you differentiate go-to-market versus distribution? Those two keywords. 50:22 >> Do you differentiate them? >> Cuz we often get confused. Like we need different definitions of these. People have different opinions of these. What are your perceptions about them? 50:32 >> I think for like go-to-mar- go-to-market for me always sounds like a B2B side and distribution 50:41 always sounds like a B2C. Like and that could be just me. But like I've always associated GTM with you know business-to-business distributions with 50:50 distribution with you know with primarily consumer. I also feel that I think distribution comes from supply chain, maybe comes 50:58 from you know all the FMCG companies which are also maybe you know they were consumer companies. Uh that's how I differentiate. Of course 51:07 you know both both could be the same, both could be different. >> Makes sense. I'm curious what you think about it. >> I I'm I I try not to get lost in 51:15 definitions. I try to just drive uh sign ups, >> That's a good answer. >> Yeah, you can call it the pink uh pony as long 51:23 as you're driving uh sign ups, right? I think uh I think at the end of the day, right? Um number one thing is like correct sign ups from the correct people. 51:31 >> Yeah. Reading from my research, there seem to be two different games when it comes to AIO. There's retrieval mode answers, 51:39 right? Which is like browsing AIO views where fresh content is better and better. And then there's LLMs actually memorizing 51:47 content, right? And this can bring us back to the half-life discussion of content. When it comes to this memorization mode, the model 51:56 recites from its training data. So, that data is the same data until the lab retrains a new model. Do you have different playbooks for these both 52:05 things in AEO? And have you thought about this? >> I have definitely thought about this. I think Reddit content is more about 52:13 training data. >> Mhm. >> So, if you're like any Reddit intervention is about training data. I don't think it's it's going to get cited. And even if it gets cited, you 52:22 you know, the click-through would go to Reddit, not to your place. So, I see Reddit intervention as a training data. Third-party blogs or our blogs is 52:31 something I see more like cited data or, you know, people who would you know, LLMs would bring this into 52:39 our ecosystem like, you know, our product ecosystem. I also feel that one of the very important things that have started 52:48 playing a role is also memory. Uh you know, how like a lot of these engines today or, you know, tools like profound 52:58 actually run prompts without logging in. But, an average user is always logged in ChatGPT or Claude or 53:06 Google. So, it matters, you know, what has been referenced before. And if you're playing in the same landscape, then, you 53:15 know, what gets referenced again. So, I think that I think training data is important for that 53:24 memory and referencing. I think they there is some study where it says that ChatGPT queries web for 60% of the queries and 53:35 you know, gives answer from its own knowledge from 40%. So, this 40% is where training is coming from and you know you're doing other 53:44 interventions. 60% is getting picked up from web where you have a chance to influence. >> So there is a first mover advantage here 53:52 because if you move quickly on a year, right? You would show up in people's um 54:01 queries and then those queries would be loaded in memory. And then you'd reshow up again basically. So would you say that's the case? >> Yes, I do yeah, I do think that and I 54:10 think that's going to and that also brings that back to you know what would relevancy and authority look like in a LLM domain. And like we we have seen 54:20 like when GPT 5.5 got rolled out to free users uh we we saw a jump in you know 54:29 people coming through click through from chat GPT. And our biggest thesis there was that I think GPT 5.5 had more memory 54:37 component to it as compared to previous models. And when when it got rolled out to you know free people we saw a lot of traffic coming from developing countries 54:46 because it was like the farthest it was rolled out was to developing countries. So I think memory would play far more 54:54 important role there. I I don't know how it will like it could be that hey I have already recommended 55:03 this uh you know this brand or this article or this citation to this person so I should do that again. Google SE also works in a similar way like a link 55:11 that you have clicked shows up you know slightly at a better ranking as compared to what you have seen it before. I would see something would happen here. Again 55:19 all of this is a black box we don't know where it will head but I'm very promising like I I'm very hopeful about it and I have seen promising results with gamma uh when it comes to you know 55:29 memory influencing discovery. >> I think it'd be really interesting to have this conversation again maybe in like a year >> [laughter] >> and see what See how like your channel's 55:39 velocity and channel distribution changes, and then also how does the black box evolve? Maybe by the next time we're having this competition, you know, >> It'll be completely different. 55:47 >> ChatGPT will, you know, Sam Altman will have watched this podcast and stopped us from gaming the system. >> Yeah, who knows? I feel that when when 55:56 we were like if if we are going to have this conversation next year, it would happen in either of the extremes, right? Either this channel would be dead, or 56:05 you know, like it it would be skyrocketing and we would still be talking more about depth. I don't think it will be somewhere in the middle. >> What happens is anytime there's an asymmetry, anytime something is working, 56:14 all the tech companies will abuse it. Once it's done, all right, maybe some companies will definitely have first mover's advantage where, you know, they're getting to abuse it for way 56:22 longer. And then once everyone catches on, the models have to evolve and adapt so that their model quality of their actual responses are of quality and not 56:30 gamed. Thus starts the new paradigm. >> It's an arms race. Arms race. That that >> That is what it is. What is While we're on this topic, what 56:38 is the future look like when it comes to, you know, growth? What are you What is something that you're taking like a bold bet on, right? 56:47 Anything experimental where you're looking out like 1 year from now? Or are you just adapting as things move on? >> Both sides, but I I'm more bullish about 56:58 agent using your product and, you know, maybe creating account on their own, or 57:06 maybe doing a checkout. Checkout is still far-fetched, but is your product agent first? I think it completely makes 57:14 sense. In developer tools, you can see it today, right? Can Cloud Code auto-install your, you know, your package and, you know, get it out there 57:23 and use like I I I do code coding using Cloud Code, and it it uses five other tools that even I don't know, and I don't care about. Like I simply trust it 57:32 to, you know, install the important libraries that are out there and get the work done. I think that would that would become more mainstream in other 57:40 knowledge work as we are seeing in Claude and it would become more and more important. Is is your product being used by agents? And 57:48 and like I want to move towards that direction, but I don't have a very good mental model of, you know, how it will evolve. 57:57 I have a couple of my growth friends who have actually tried experimenting where they have said that we would give, for example, 10% extra 58:06 reward to the model if model ends up signing right away. Like, they >> Like, you're negotiating with the model. 58:14 >> Exactly. >> Where are you putting that text information? >> I think you're putting you're putting it in your JSON schema 58:22 and you're also putting it I think in your install package or something like that that if you do it 58:30 right now, you're going to get this as compared to this >> That's next level gaming. I love it. >> Yes. So, I think there are a lot of these work really being done by a lot of 58:39 good teams out there. I I haven't done, in all honesty, I haven't done a lot of work on that, but I am particularly excited about it. 58:48 >> That's very insightful, to be honest. Actually, a quick question. I've got two questions here. One is is the growth on agentic usage 58:58 faster? Is there a bigger gradient than it is on human usage or not yet? >> I don't think so. I think uh 59:07 >> For Gamma specifically? >> For Gamma, no way. I cuz the >> Yeah, the master is a presentation. Like, people present to other people. >> Okay. >> And plus when GPT, for example, I feel 59:16 like this is the conversation that they will have when Claude is dominating over GPT because that represents a more sophisticated user base, in my opinion, 59:24 right? So, that's when these sort of agents, because right now think about the population of the world which is even using agents. 59:32 >> Yeah. So >> Very tiny. >> And the question is how do you market to agents? >> That's the next. I'm guessing though. >> Yeah, I think that that is going to be I 59:41 think the path will go through AIO because fundamentally what you're doing is marketing to agents. AIO that looks like a very level zero or MVP version of 59:50 marketing to agents if you you know, if you're doing all your interventions in AIO. It will it will evolve. 59:59 Again, I I haven't spent much time on it. It's very interesting topic for me as well. And I would love to spend some more time. 60:07 >> Well, that could be like a dedicated hire whose ROI will come years later but you're seeding all of that in like information for LLMs now. That's very 60:16 interesting.