Lenny's Podcast · Ethan Smith (Graphite)

The Ultimate Guide to AEO: How to Get ChatGPT to Recommend Your Product

SEO veteran Ethan Smith, CEO of Graphite, breaks down answer engine optimization — getting your product to come up when people ask ChatGPT, Gemini, or Perplexity for a recommendation. It's intensely tactical, sometimes counterintuitive, and mostly an extension of SEO rather than a replacement for it.

Creator: Ethan SmithRuntime: 72 minVideo published: September 14, 2025162.8K views when summarized

Watch on YouTube

TL;DR

The tactical version

AEO — answer engine optimization — means getting your product into LLM answers. Ethan calls it the second-biggest shift in SEO's history, an evolution rather than a reinvention, and points out Google isn't dying: the pie just gets bigger.
The core difference from SEO: LLMs summarize many citations, so you win by being mentioned most, not by ranking #1. That's why a brand-new company can win fast — a Reddit thread or YouTube video can get you cited tomorrow.
The payoff is real: Webflow saw ~6x higher conversion from LLM traffic than from Google search, and ~8% of its signups now come from LLMs.
The playbook: find the questions, track your share of voice, answer every follow-up on your own site, earn citations off-site (YouTube, honest Reddit, affiliates like Dotdash), and run real test/control experiments — because most best practices are wrong.
Pure AI-generated content doesn't work, and at scale it risks model collapse — answers converging on 'only vanilla.' The most underused move is optimizing your help center for the long tail of 'does your product do X?' questions.

01 · The shift

AEO is the second-biggest change in SEO — not a reinvention

People are moving from Google to ChatGPT, Claude, Gemini, and Perplexity for answers, and Google is bolting on AI Overviews and AI Mode in response. That's what creates AEO — answer engine optimization, or SEO for chatbots. But Ethan, who's done SEO for 18 years, is careful about the framing: this is the second-biggest change he's seen (the 2007 anti-spam Panda era was bigger), and it's a summarization of search with new inputs, not a clean break.

Ethan Smith (Graphite) with Lenny Rachitsky. His reframe: everything that works in SEO still works in AEO — there's just more on top of it.00:09:24

Google isn't the loser hereEvery couple of years something gets declared the Google killer — TikTok, Instagram, now AEO. They keep adding new channels instead of taking from search. Google's slice stays about the same; the pie gets bigger.

02 · The new rule

Get mentioned the most, not ranked #1

The core difference is counterintuitive. On Google, the #1 blue link wins. In an LLM, the answer is a summary of many citations — so the product mentioned most often across those citations wins, even when no single page ranks first.

"You can get mentioned by a citation tomorrow and start showing up immediately." Early-stage companies can win, and win fast.00:00:36

That flips the usual advice. Ethan normally tells startups not to do SEO early, because you spend months building domain authority first. AEO is the opposite: a Reddit thread, a YouTube video, or a launch post can get you cited tomorrow. And the long tail — specific questions nobody would have typed into Google — is far bigger in chat, full of questions never asked before that you can win.

03 · Why it matters

The traffic converts ~6x better

This isn't a vanity channel. The leads convert far better, because by the time someone clicks through they've had a multi-turn conversation that built real intent.

Webflow's data: LLM traffic converted at roughly 6x the rate of Google search traffic.00:00:48

Web Flow saw a 6x conversion rate difference between LLM traffic and Google search traffic.— Ethan Smith

And it isn't small: Webflow now gets about 8% of signups from LLMs — already a top channel, and growing.

04 · Where citations come from

On-site, off-site, and Dotdash

Split the work in two. On-site is traditional SEO plus answering the long tail — every follow-up about your features, integrations, use cases, languages. Off-site is earning citations, which fall into a few groups, each needing its own approach.

Video

YouTube/Vimeo — easy to control, and wide open for unglamorous, high-value B2B topics nobody makes videos about.

UGC

Reddit and Quora, heavily cited and heavily policed (see next section).

Affiliates

pay-to-play and controllable — "want to be the best credit card? Pay Forbes." Dotdash Meredith (Investopedia, Good Housekeeping, All Recipes) is the most-cited of all.

Citations differ by vertical

B2B leans Tech Radar; commerce leans Glamour/Cosmopolitan; marketplaces lean Yelp/Eater/TripAdvisor.

05 · Reddit, done right

Be a real person, not a spam farm

Reddit is one of the most-cited sources in LLMs, and the #1 thing clients ask about. The obvious growth move — hundreds of fake accounts posting automated comments — doesn't work; the accounts get banned and the comments deleted. Reddit polices itself, and ChatGPT trusts it precisely because it's policed.

The approach that works is almost boringly honest.00:00:32

Find a thread… say who you are, say where you work, and then give a useful piece of information. And that works really well… you don't actually need 10,000 comments. Even five could be great.— Ethan Smith

06 · The playbook

Find questions, track, earn citations, experiment

The loop: (1) turn your and your competitors' paid-search money terms into questions — ChatGPT will convert them for you; (2) put them in an answer tracker and watch your share of voice, since answers vary run to run, so you're measuring a distribution, not one result; (3) see who's getting cited and build a strategy per group; (4) build landing pages matching the page types that show up, and answer every follow-up.

On tooling: pick the cheapest answer tracker that does the job — there are about 60 and they're basically the same.00:23:58

Run real experimentsHold out 100 control questions you don't touch, intervene on 100 test questions, compare against the control, then reproduce the result. "Most best practices, most blog posts are not correct." Ethan's academic-research background shows here: reproducibility over folklore.

07 · The AI-content trap

Pure-AI fails, and risks "model collapse"

A rigorous study — running a validated AI detector against pre-ChatGPT common-crawl content — found only about 10–12% of the content in Google and ChatGPT results is fully AI-generated, and that purely automated, no-human-in-the-loop content doesn't work. AI-assisted, human-edited content is clearly where this goes.

Why pure-AI is self-defeating: if it worked, everyone would scrape, and the engine would just become a search engine for its own derivatives.00:52:00

If you ask what's the best flavor of ice cream, it will eventually say it's vanilla and it's only vanilla and there's no other flavor of ice cream.— Ethan Smith

That's model collapse: feed the derivatives back in and the wisdom-of-the-crowd that makes a summary good shrinks to one impoverished opinion.

08 · Help center

The long-tail questions nobody optimizes for

Ethan's most underused tip: chat users constantly ask "does your product do X / integrate with Y?" — which is exactly what a help center answers, and exactly what SEO teams ignore. Three fixes: move it from a subdomain to a subdirectory, cross-link the articles, and fill in the tail — the obscure use cases nobody else covers, so you become the only citation.

Open it to the community so they fill the long tail of questions for you.01:00:30

His example: "which meeting-transcription tool integrates with Looker?" None do directly — but Otter via Zapier to BigQuery to Looker does. Nobody had written that up, so whoever does wins that question.

09 · Myths & the honest truth

Misinformation, attribution, and playing the game

The space is unusually full of misinformation, partly because "everything is new and different" sells tools. The reality: SEO and AEO overlap heavily, Google traffic to publishers is flat-to-up rather than dying, and attribution is genuinely murky — people see you in an answer, then open a new tab and type your brand into Google, so it shows up as branded search or direct traffic.

Headlines like this come back every couple of years. They're usually wrong.00:01:00

It's not your choice whether to play the game. You are playing the game whether you want to or not.— Ethan Smith

You can block LLMs from training on your content (robots.txt, separate user agents) — but blocking indexing entirely just hands the citations to your competitors.