Generative Engine Optimization: how to get cited in AI Overviews and ChatGPT

Two years ago, ranking number one meant you won the click. Today a large share of searches end inside an answer — Google's AI Overviews, ChatGPT, Perplexity, Gemini — where the person reads a synthesised paragraph and never scrolls to the ten blue links. Being position one and being named inside the answer are now two different battles, and most teams are only fighting the first one.

We have spent the last year watching which pages get pulled into AI answers and which get ignored. The pattern is consistent enough to act on, and it has a name: Generative Engine Optimization, or GEO — optimising not for a ranking slot, but for being quoted, cited, or named by a language model when it writes an answer. This is a field guide to the moves that actually earn those citations, written from what we see work, not from theory.

What Generative Engine Optimization actually is

GEO is the practice of making your content easy for a language model to retrieve, trust, and quote when it composes an answer. It is not a replacement for SEO and it is not a trick. It is a shift in the unit you optimise: from the page that ranks to the passage that gets cited. An AI Overview is not ranking ten pages and showing you the best one. It is retrieving a handful of candidate passages, deciding which ones answer the sub-question cleanly, and stitching them into prose — attaching a citation to the source of each claim it keeps. Everything about GEO follows from that mechanic.

How AI engines decide who to cite

Three behaviours matter more than anything else we have tested. First, engines cite passages, not whole pages. A 3,000-word guide gets cited for the one paragraph that answered a narrow question crisply, and the rest is dead weight as far as the model is concerned. Self-contained, extractable answers win. Second, engines reward specificity. “Build high-quality backlinks” never gets cited because there is nothing to attribute; “in our audits, profiles above roughly 30% exact-match anchors tend to underperform” does, because the model composing an answer reaches for a concrete, sourceable claim. Third, engines trust entities they already recognise. If your brand is mentioned across the web in the right context, the model is far likelier to name you — which means GEO is downstream of being talked about, and therefore downstream of links and digital PR, not separate from them.

The seven on-page changes that earn citations

These are the edits we make to a page when the goal is AI-answer inclusion. None of them is exotic; the discipline is in doing all of them consistently.

1. Answer the question in the first two sentences of every section. Lead with the claim, then support it. Models lift the lead sentence far more often than a buried conclusion. 2. Write quotable units. One-sentence statements of fact, ideally with a number, that stand on their own without the surrounding paragraph. 3. Add real FAQ schema. Use the actual questions people type into the engines, answered in two or three sentences. 4. Make your entities unambiguous. A consistent brand name, an About page, Organization schema and sameAs links to your profiles all help the model disambiguate you before it decides to cite you. 5. Cite your own sources. Pages that reference primary data and name where it came from read as more trustworthy, and get treated as a citable source in turn. 6. Keep one idea per URL. Sprawling “ultimate guides” dilute the passage-level relevance retrieval depends on. 7. Earn mentions off-site. You cannot optimise your way into an answer if the wider web never mentions you.

Why schema is the highest-leverage lever

Of everything on that list, structured FAQ markup is the change we have seen move inclusion fastest. The reason is simple: an FAQ block is already shaped the way an answer engine wants its input — a clear question paired with a short, self-contained answer. When you mark that up with FAQPage schema, you are handing the model a pre-extracted passage with an explicit question attached. Pair it with Article and BreadcrumbList schema so the engine understands what the page is and where it sits, and you have removed most of the friction between your content and a citation. This page uses exactly that stack, which is the honest way to write about it.

Why GEO is link building's second act, not its replacement

Here is the part most “GEO” takes get wrong. It is not a replacement for links; it is the same trust signals, measured at the answer layer. The brands getting named in AI answers are, overwhelmingly, the ones the model already recognises as authorities — and that recognition still comes from relevant links and real mentions earned across the web. GEO is how you present your authority so it is easy to quote. Links and digital PR are how you earn the right to be quoted at all. Teams that treat the two as separate disciplines tend to produce beautifully structured pages that no model has any reason to trust. The sequence that works is: earn genuine authority, then structure it for extraction.

How to actually measure it

You cannot improve what you do not watch. The practical loop is: pick the ten questions a buyer would ask an AI about your category, ask them across ChatGPT and Gemini on a schedule, and record whether your brand is named, who is named instead, and which source the model cited. That last column is the gold. If a competitor keeps getting cited for a question you should own, the fix is usually obvious once you read their cited passage: they answered it more cleanly, or they are simply mentioned in more of the places the model trusts. This is the exact loop we built into OptiLink's AI-visibility tracking, because doing it by hand across engines every week does not scale.

Where to start this week

If you do nothing else, do three things. Pick your five highest-intent pages and rewrite the opening sentence of each section to lead with the answer. Add genuine FAQ schema to those pages using real questions, answered in two or three sentences each. And run an AI-visibility check to see which questions you already get named for and which you do not, so you are investing in the gaps that matter rather than the ones that feel urgent. GEO rewards the same thing good SEO always has — being genuinely useful and genuinely trusted — it just measures it one layer closer to the user.

AI Overviews, ChatGPT and Perplexity are not the same target

It helps to stop treating “AI” as one destination. Google's AI Overviews sit on top of the classic index, so they lean heavily on pages that already rank and already carry conventional authority signals — which is why strong SEO is still the price of entry there. ChatGPT with browsing, and Perplexity, behave more like live research assistants: they run fresh retrieval, weigh how directly a passage answers the prompt, and are more willing to cite a smaller site if its answer is the cleanest one available. Gemini sits somewhere between the two, tied closely to Google's understanding of entities. The practical takeaway is that you win AI Overviews mostly by being an authority that also happens to be well-structured, and you win the chat engines mostly by being the clearest, most self-contained answer to a specific question. A page built for both — authoritative and extractable — is the only durable bet, because the engines keep converging on the same preference for trustworthy, directly-useful passages.

A worked example: from ignored to cited

Take a page that ranked on page two for “how often does Google crawl new backlinks” and never appeared in a single AI answer. Nothing was wrong with it exactly; it was just a wall of hedged prose. We changed three things. We rewrote the opening of each section to state the answer first — “Google typically re-crawls and re-evaluates a new backlink within a few days to a few weeks, depending on the linking page's own crawl frequency” — a single sentence a model can lift whole. We added an FAQ block answering the four follow-up questions people actually ask, each in two sentences. And we added one concrete data point from our own logs instead of a vague “it varies.” Within a few weeks the page started surfacing as a cited source in chat-engine answers for that query and its close variants. We did not build a single new link to it in that window; we simply made the authority it already had easy to quote. That is the whole game in miniature.

The GEO mistakes we see most often

The first mistake is burying the answer. Writers trained on classic SEO warm up for three sentences before saying anything a model can use, and the model moves on. The second is faking specificity — inventing statistics to sound quotable. Engines increasingly cross-check claims, and worse, your readers do; a fabricated number is a trust fault that outlives any short-term gain, and we never recommend it. The third is schema theatre: marking up FAQ blocks whose answers are keyword-stuffed filler rather than genuine two-sentence answers. Schema only helps when the content underneath is real. The fourth, and most expensive, is treating GEO as a content-only project. If your brand is absent from the places a model already trusts, no amount of on-page structure will conjure a citation out of thin air. You have to earn the mentions as well as format the page.

Your two-week GEO sprint

Here is a plan you can actually run. In the first few days, list the fifteen questions a buyer in your category would ask an AI, and check today where you stand on each — named, not named, or named-but-not-cited. Pick the five questions you are closest to owning. Over the next week, rebuild the pages that target those questions: answer-first openings, genuine FAQ schema, one real data point each, Organization and Article schema in place, and a clear internal link to your most authoritative related page so the engine can trace your expertise. In the second week, turn to off-page: pitch one piece of original data to publications in your niche, and make sure your brand is described consistently everywhere it already appears. Then re-run the same fifteen questions and watch the named-and-cited column. The gains compound, because every page you fix and every mention you earn makes the next citation easier for the model to justify.

None of this is a hack, and that is the point. The engines are converging on rewarding exactly what they claim to reward: content that is genuinely useful, genuinely trustworthy, and easy to extract an answer from. GEO is not a loophole around that standard — it is the discipline of meeting it on purpose, one passage at a time, and then proving you meet it by measuring whether the machines that now sit between you and your customer actually say your name.

Who should own GEO inside a company

One reason generative engine optimization stalls is that nobody is clearly responsible for it. It falls between teams: SEO treats it as a content problem, content treats it as a technical problem, and PR treats it as somebody else's problem entirely. In practice GEO draws on all three, which is why it works best with a single owner who can pull each lever — someone who can commission the first-hand content that earns citations, enforce the structure that makes it extractable, and coordinate the off-site mentions that build the authority engines look for. The owner does not need to do all the work, but they need to be accountable for the one metric that matters: whether the engines name you for the questions your buyers ask. Without that accountability, GEO becomes everyone's background task and nobody's result, and the compounding never starts.

The refresh cadence that keeps you cited

Being cited once is not the same as staying cited. Engines favour current sources, and a page that was authoritative eighteen months ago can quietly fall out of the answer set as the topic moves and fresher, more specific sources appear. The fix is a deliberate refresh cadence on the pages that matter most: revisit your highest-value pages on a schedule, update the figures, add what you have learned since, and re-state the claims as cleanly as the first time. This is not churn for its own sake — it is maintenance of the signals that keep you in the answer. A small set of genuinely maintained pages will outperform a large set of pages published once and abandoned, because the maintained ones keep earning the freshness and accuracy that the engines reward.

Brand mentions matter, not just links

Classic SEO trained everyone to think in links, but the AI engines read something broader: mentions. Being named in a roundup, quoted in an article, discussed in a forum, or referenced in a comparison all contribute to the model's sense that you are a recognised entity on a topic, whether or not those mentions carry a clickable link. This widens where authority can come from. It means an unlinked mention in a respected publication can help your citation odds, and it means the off-site work worth doing is not narrowly about link acquisition but about becoming genuinely talked-about in the places your topic is discussed. The practical shift is from chasing links to earning presence — being the brand that comes up when people in your field discuss the problem you solve.

Where GEO is heading

The specific tactics in this guide will shift as the engines evolve, but the direction of travel is stable enough to plan around. Answers will keep getting better at synthesising multiple sources, which rewards being genuinely distinctive rather than redundantly comprehensive. Attribution will matter more, not less, as the engines face pressure to show their sources, which makes being a nameable, credible entity increasingly valuable. And the premium on first-hand experience will grow as generic, model-generated content floods every topic, leaving real expertise as the scarce and citable thing. None of that requires predicting the next model release. It just requires building the durable assets — genuine expertise, clear structure, recognised authority — that every version of these engines has rewarded and shows every sign of rewarding more.

Why this beats chasing algorithm updates

A final framing worth internalising: generative engine optimization done this way is largely immune to the anxiety that dominates traditional SEO, the constant scramble to react to the latest algorithm update. The reason is that none of the durable work here depends on a quirk of ranking logic that a future update might erase. Genuine expertise, clearly-structured content, a well-understood entity, and earned authority are rewarded by every version of these engines because they are what the engines are fundamentally trying to surface. When you build on those, an update is far more likely to help you than hurt you, because updates generally move the engines toward rewarding genuine quality, not away from it. That is the quiet advantage of optimising for substance rather than for tricks: you stop playing a game you can only lose eventually, and start building assets that each successive update tends to validate rather than threaten.

Frequently asked questions

Does Generative Engine Optimization replace SEO?

No. GEO is a layer on top of SEO. The same authority, link and E-E-A-T signals decide whether an AI engine trusts you enough to cite you; GEO just structures that authority so it is easy to quote.

How quickly does GEO work?

Inclusion in AI answers can move within a few weeks of restructuring a page for extractable, answer-first content with FAQ schema. Building the off-site authority that earns citations still takes months, like classic SEO.

What is the single highest-impact GEO change?

Adding genuine FAQ schema — real questions answered in two or three self-contained sentences — on pages that already have topical authority. It hands the engine a pre-extracted, attributable passage.

Do AI Overviews and ChatGPT reward the same things?

Partly. AI Overviews lean on pages that already rank and carry conventional authority, while ChatGPT and Perplexity run fresh retrieval and favour the clearest self-contained answer. A page that is both authoritative and extractable wins in both.

How do I measure whether AI engines cite my brand?

Pick the key questions a buyer would ask an AI in your category, ask them across ChatGPT and Gemini on a schedule, and record whether you are named and who is cited instead. OptiLink automates this as an AI-visibility check.