How to Get Your Startup Recommended by ChatGPT (GEO for Founders)

A founder messaged me last month, half amused and half terrified: "I asked ChatGPT who to hire for my product design, and it named three studios. We weren't one of them."

That is the whole game now, compressed into one sentence. Buyers used to Google a category, open six tabs, and decide for themselves. Today a huge share of them — roughly 64% by the numbers I trust — start by asking an AI. They type "best [category] for [use case]" or "who should I hire for X" into ChatGPT, Perplexity, Claude, or the AI Overview sitting on top of Google, and they treat the answer as a shortlist. If your startup isn't in that answer, you don't lose the deal. You never enter the room.

I run into this from two sides. I'm a founder trying to get my own studio, Elysium Designs, named by these engines, and I'm the person other founders ask to make it happen for them. So this isn't a trend piece. It's the actual playbook I use, written for the founder who just realized the machine is now making introductions and wants to be one of them.

Why AI recommendations work nothing like Google

Here is the mental model to burn in before anything else: answer engines don't rank links — they retrieve, synthesize, and cite. Classic search hands you ten blue links and lets you do the judging. An answer engine does the judging for you. It reads across many sources, writes one confident paragraph, and names a handful of brands as the answer. There is no page two. There is barely a page one. There's the answer, and there's everyone the answer forgot.

That changes what you're optimizing for. You're no longer fighting to out-rank a competitor on a results page a human scrolls. You're fighting to be the source a model reaches for when it composes the reply — and to be trustworthy enough that it puts your name in the citation. That discipline has a name now: Generative Engine Optimization, or GEO. Some people call it answer engine optimization. Same idea. It's SEO's younger sibling who grew up in a world where the search result talks back.

SEO got you found. GEO gets you recommended. Those are not the same job, and doing the first one well no longer guarantees the second.

If you want the deeper contrast with old-school search, I get into it below — but the short version is that GEO and SEO share a foundation and then split. You still want good content and clean technical hygiene. But GEO adds a layer that most founders, and almost all of their competitors, have never touched.

The three things that decide whether AI recommends you

After running this for my own work and for clients, I've stopped believing it's mysterious. Whether an engine recommends you comes down to three factors. Get all three moving and you become quotable. Ignore any one and you stay invisible.

1. Clarity — write like you're being quoted, because you are

A model can only recommend you if it can cleanly extract a passage that answers the question. This is the part designers and founders get wrong most often: we write to impress, in clever, winding prose, and a model can't lift a single clean sentence out of it. Clarity means saying the answer plainly, near the top, before the throat-clearing. If someone asks "what's the best approach to X," the strongest content on the web states the answer in one direct passage and then supports it. Comprehensive guides and honest "best X for Y" formats win here because they mirror the exact shape of the response the AI is trying to build. You're not writing an essay. You're pre-writing the model's answer for it.

2. Credibility — the new word-of-mouth

An engine won't stake its reputation on a brand only that brand talks about. It looks for corroboration: the same name showing up, consistently, across sources it already trusts. Reviews, mentions, being covered in an industry publication, getting cited by a site with real authority — these are the signals that turn "a company that exists" into "a company worth recommending." And it's weighted: a mention on a high-authority site is worth far more than ten on places nobody trusts. This is the slow lever. You can't fake it in a weekend, which is exactly why it works — and why the founders who start early build a moat the latecomers can't sprint past.

3. Machine-readable structure — the lever nobody is pulling

This is the one I get genuinely excited about, because it's fully in your control and almost everyone skips it. Models parse the web far more easily when you hand them clean signals: schema markup that spells out who you are and what you do, consistent entity data so "Elysium Designs" means the same thing everywhere, organized pages instead of a content junk drawer, and an llms.txt file — a simple text file that tells AI crawlers, in plain language, what your site is about and what matters on it.

I implemented llms.txt on my own site well before most designers had heard the term, and I'll tell you plainly: it is one of the highest-leverage, lowest-effort moves available right now, precisely because your competitors don't know it exists yet. Clarity and credibility are contested fields. Machine-readable structure is an open one. That's rare, and it won't stay open forever.

The founder's GEO audit: 30 minutes, brutal honesty

Before you optimize anything, you need to know where you actually stand — not where you assume you stand. Here's the audit I run, and you can do the first pass yourself tonight.

Step 1 — Write the questions your buyer actually asks

List 10 to 15 real questions your ideal customer would type into an AI. Not keywords — questions. "Best design studio for an AI startup." "Who should I hire to build an MVP?" "What's the best branding agency for a fintech seed round?" Think like your buyer at the exact moment they're looking for someone like you.

Step 2 — Run every one of them, for real

Open ChatGPT, Perplexity, and Google's AI results, and actually ask each question. Don't guess the output. Read it. For each one, note three things: do you show up, which competitors do, and — this is the gold — what got cited as the source. That citation list is not trivia. It's your roadmap. It's the exact set of pages and sites the engine already trusts on your topic.

Step 3 — Read the citations as instructions

Where a competitor got named, go see why. Was it a comparison article? A directory? A publication that covered them? Each cited source is a door, and now you know which doors the engine is walking through. You're not guessing at strategy anymore. The machine just told you what it rewards.

The citation list under an AI answer is the closest thing to a leaked exam. Most founders never think to read it.

Turning the audit into a plan

Once you can see the gap, the work gets almost mechanical. Here's the order I'd actually do it in — fastest-moving levers first.

• Write the definitive answer to each key question. One clear, comprehensive, honestly-argued page per question that matters. Plain language, answer up top. This is what gets lifted into the response.
• Ship the technical layer. Add schema markup, publish an llms.txt file, and make your entity data consistent everywhere your name appears. This is the open field — get there before your competitors do.
• Build credibility deliberately. Pursue the mentions, reviews, and third-party coverage that showed up in your citation audit. Aim for higher-authority sources; a few strong ones beat a pile of weak ones.
• Re-run the audit monthly. GEO isn't a launch, it's a loop. Ask the same questions, watch your presence change, and follow the new citations.

Notice how much of that first list you control outright. Clarity and structure don't require permission, budget approval, or anyone else's help. You can start tonight. The credibility layer takes longer — but it compounds, which is the good kind of slow.

The mistakes I see founders make with GEO

Because the field is new, most of the errors are honest ones — people applying old instincts to a game with new rules. These are the ones I watch cost founders the most, in rough order of how often I see them.

Treating it like SEO with new keywords

The most common mistake is assuming GEO is just SEO with a fresh coat of paint — stuff more keywords in, chase more backlinks, done. It isn't. Keyword density does very little when a model is trying to extract a clean, correct answer. What moves the needle is a passage that states the truth clearly and is corroborated elsewhere. You're optimizing for comprehension and trust, not for a ranking algorithm counting term frequency.

Writing for cleverness instead of extraction

Founders and marketers love a clever opening. But if a model has to wade through three paragraphs of atmosphere to find your actual point, it will quietly grab a competitor who just said the thing. Lead with the answer. Be the site that states, in one clean sentence, exactly what the buyer asked — then earn the rest.

Skipping the technical layer because it's invisible

Schema and llms.txt don't show up on your homepage, so they feel optional. They aren't. They're the difference between a model confidently understanding who you are and a model guessing. This is the cheapest, fastest edge available, and it's invisible precisely to the people who'd otherwise compete with you for it. Do the boring, unglamorous work here and you win a race most people don't know they're running.

Checking once and calling it done

GEO is not a launch you ship and forget. The engines change, your competitors publish, the citations shift. Founders who run the audit once and never again slowly slide out of the answers without noticing. The ones who treat it as a monthly loop keep compounding. This is a habit, not a project.

How long until it works?

The honest answer, because founders always ask and deserve a real one: it depends on which engine and which lever. Open-world engines that read the live web — Perplexity, Google AI Overviews — update in weeks, not months. Publish a genuinely clear answer, structure it well, and you can watch yourself appear surprisingly fast. Across the board, most brands see measurable movement in their AI citations in roughly four to eight weeks. The credibility signals that depend on other people covering you move slower, over months. But the parts you fully own — content clarity, schema, llms.txt — start working almost immediately, which is why I always start there.

This same thinking runs through everything I write about building a company AI and humans both take seriously — whether that's designing AI products people actually trust or knowing when a startup's brand is ready to scale. Discoverability, trust, and clarity are the same muscle pointed at different targets.

Why the window is open right now

Here's the part I want founders to actually feel, not just read. The field is wide open. Most of your competitors don't know GEO exists — they're still buying backlinks and tweaking meta titles for a search behavior that's quietly draining away. The engines that recommend you are new, the best practices are barely written, and the technical lever that moves the needle most is the one almost nobody is pulling. That combination — high impact, low competition, short timeline — does not come along often. It's the same asymmetry early SEO had in 2005, except this time the results talk.

This is the new word-of-mouth. When someone asks a machine "who should I trust with this," you want to be the name it says without hesitating. That doesn't happen by accident, and it definitely doesn't happen to the people waiting to see if it's real. It happens to the ones who structured their story so cleanly that the machine had no choice but to repeat it.

---

The brands AI recommends aren't the loudest or the biggest. They're the clearest — the ones who made themselves easy to trust, easy to quote, and impossible to leave out of the answer.