Generative engine optimisation (GEO): a founder's field guide

SEO optimises for a list of links. GEO optimises for a paragraph that has already decided. When someone asks ChatGPT or Perplexity which tool to use, the engine reads a handful of sources, synthesises an answer, and names two or three products. You are either in that paragraph or you are not. There is no second page to climb to.

Generative engine optimisation is the practice of being one of the things an AI engine pulls into that answer. It is not a rebrand of SEO and it is not magic. It is a specific, learnable craft with a small number of levers. This is a field guide for a solo founder who builds well, has no agency budget, and wants to understand the mechanism before spending a single hour on it.

What generative engine optimisation actually is

Start with how the engine produces an answer, because the tactics fall out of the mechanism. When you ask a generative engine a question like "what is the best tool for X", one of two things happens, and increasingly both happen at once.

The first is retrieval. The engine runs a search behind the scenes, fetches a set of pages, reads them, and uses them as raw material for the answer. Perplexity does this openly and shows you the sources. ChatGPT and Gemini do it when the question needs fresh or specific information, and Google AI Overviews does it on the search results page itself. The pages it retrieves are the ones in play. Nothing else can be cited.

The second is parametric recall. The model has already read a large slice of the public web during training. Some facts about your product, or your category, or your competitors, are baked into its weights. When the engine answers from memory rather than from a live fetch, the question becomes whether the model learned anything about you at all, and whether what it learned was accurate.

GEO works on both paths. You want to be present in the sources an engine retrieves, and you want to be a known, correctly-described entity in what it already remembers. Everything below is a way of pushing on one of those two levers.

Why GEO is not SEO

The two overlap. Good SEO content is still the raw material GEO feeds on, and a page that ranks well is a page an engine is more likely to retrieve. But the optimisation targets are different, and three differences matter enough to change what you actually do.

The unit of success changed

In SEO the unit is a ranked URL. In GEO the unit is a claim. An engine does not cite your homepage. It cites a sentence, a comparison, a specific fact it can lift and attribute. If your pages are full of vague positioning and adjectives, there is nothing for the engine to extract. If they contain clear, checkable statements, there is. This single shift reorganises how you write.

You can win without the click

SEO success was a visit. GEO success is often a mention with no visit at all. The engine reads your page, names you in the answer, and the user never lands on your site. That feels like a loss if you only count sessions, but the mention is doing the work. The person asked which tool to use and heard your name in the reply. AI assistants increasingly answer questions without sending a click, so a chunk of your distribution now happens off your property, in a paragraph you do not control. GEO is about earning your place in that paragraph anyway.

The corpus is wider than your site

Classic SEO is mostly about pages you own. GEO depends heavily on pages you do not. Engines reach for sources they treat as trustworthy and specific: a Reddit thread where someone describes the exact problem and the tools they tried, a comparison article on a third-party site, a directory listing, a well-structured documentation page. Your own site is necessary but not sufficient. Half the work is being correctly described in places you do not own.

The GEO playbook, in priority order

Here is the part you can act on. The order matters. Each step is cheap for a solo founder and compounds with the ones above it. Do them top to bottom.

Make claims an engine can lift

This is the highest-leverage step and most founders skip it. Go through your core pages and find every place you have written a soft adjective where a hard claim belongs. "Powerful analytics" is not citable. "Tracks events without code and surfaces drop-off in a funnel within the first session" is. The second sentence can be lifted whole into an answer and attributed to you. The first cannot.

Practically, this means writing declarative, specific, self-contained sentences. State what your product does, who it is for, how it differs from the obvious alternative, and what it costs, in plain language, near the top of the relevant page. Each sentence should make sense if an engine quotes it on its own, with no surrounding context. Imagine the model has a pair of scissors and can keep only one sentence. Write that sentence.

A concrete test: take any paragraph on your site and ask whether it answers a question a user would actually type. If a paragraph answers "what does this do for a team of five" or "how is this different from the market leader", it is GEO-ready. If it only signals vibe, rewrite it.

Be a clear, consistent entity

Engines reason about the world in terms of entities: named things with stable properties. Your product is an entity. So is your company, your category, and each competitor. The engine's job, when it answers, is to retrieve or recall the right entities and describe them correctly. Your job is to make your entity easy to identify and hard to confuse.

That means using your product name consistently across every surface, pairing it with the same short category descriptor every time, and making sure the obvious external references agree. If your homepage calls you a "customer feedback platform", your directory listings, your company profiles, your social bios, and your documentation should not call you four other things. Consistency is what lets a model collapse all those mentions into one confident entity rather than a fuzzy cloud it hedges on.

A quick way to audit this: ask a few of the engines directly who you are and what you do. The answer tells you what the model currently believes. If it is wrong, blank, or confused, that is your entity problem stated out loud, and it tells you exactly which fact to go correct in the wider corpus.

Add structured data

Structured data is machine-readable markup you add to your pages that states facts explicitly rather than leaving them buried in prose. Schema.org markup for your organisation, your product, your FAQ entries, and your pricing gives a retrieving engine an unambiguous version of the facts it would otherwise have to infer from your layout.

For a founder this is a one-time job with lasting payoff. Add Organization markup so your name, logo, and category are explicit. Add Product or SoftwareApplication markup for the thing you sell. Turn your help and pricing questions into real FAQ markup, where each question and answer is a discrete, labelled pair. The point is not the markup for its own sake. The point is that an FAQ entry is already shaped exactly like the question-and-answer an engine wants to produce, so it is unusually easy to lift.

Answer the real questions in full

Generative engines are answering questions, so write the answers. Not keyword pages. Actual questions a user in your category types, answered completely on a page of their own. "How do I do X without a developer." "What is the difference between tool A and tool B." "Is there a free way to do Y." Each of these is a query someone is putting to an engine right now, and the page that answers it cleanly is the page that gets pulled into the reply.

Structure each answer so the engine can find the resolution fast. State the answer in the first two sentences, then explain. Bury nothing. Comparison questions are especially valuable because they are exactly what someone asks at the point of deciding, and a fair, specific comparison page is the kind of source engines like to cite. Write the honest comparison, including where the other tool is the better choice. Hedged, salesy comparisons read as untrustworthy to a model the same way they do to a person.

Earn mentions in the sources engines read

This is the slowest lever and the one with the highest ceiling. Engines lean on a recurring set of trusted, specific sources: community threads, comparison and review sites, directories, and the occasional piece of press. To be cited in an answer about your category, you often need to be present in those sources, not just on your own site.

For a solo founder the realistic version of this is unglamorous and works. Be genuinely useful in the communities where your users already discuss the problem, so that the threads an engine retrieves happen to mention you in context. Get listed in the directories and review sites for your category, accurately described. Earn the comparison-article mentions by being a real option worth comparing. None of this is link-buying or spam. It is being a known, correctly-described option in the places that already shape what the engines believe.

One caution. Do not manufacture fake mentions or astroturf community threads. Engines and the platforms that host these sources are both getting better at detecting it, and a model that learns your name from low-trust spam describes you with low confidence, which is worse than not being known. Slow, real presence beats fast, fake presence here.

Measuring whether GEO works

GEO is harder to measure than SEO because the win is often invisible. There is no ranking position for "named in a ChatGPT answer". You have to assemble the picture from a few angles.

Ask the engines your category's real questions, on a regular cadence, and record whether you are mentioned and how accurately. That is your direct signal. Watch referral traffic from the AI assistants in your analytics, which is small for most products today but worth a trend line. Keep an eye on Search Console for the question-shaped queries you are now answering, since the same content earns both retrieval into AI answers and traditional impressions. None of these is a single clean number. Together they tell you whether the engines are starting to know you.

The honest framing is that this is a trend you watch over months, not a dial you read on a Tuesday. The point of measuring is to catch the direction early, not to produce a precise figure that does not exist yet.

Where to start

If you do nothing else, do the first two steps. Rewrite your core pages so they contain claims an engine can lift, and make your product a clear, consistently-described entity across every surface. Those two are cheap, they sit entirely within a solo founder's control, and they are the foundation everything else builds on. Structured data, question pages, and earned mentions all compound on top of a clear entity making liftable claims. They do little on top of vague positioning.

The mindset shift is the real takeaway. Stop optimising for a ranked list and start optimising to be a fact an engine is confident enough to state. Write claims, not adjectives. Be one consistent entity, not a fuzzy cloud. Answer the actual questions in full. Show up, honestly, in the sources the engines already trust.

If you want a read on where you stand before you start, the free Growth Snapshot scores your discoverability across seven dimensions, including how the AI engines currently describe you. AfterLaunch then monitors those surfaces and drafts the work in your voice for you to approve. Either way, the playbook above is yours to run. The engines are already answering questions about your category. The only question is whether your name is in the answer.