How to get your brand cited in AI search

What is AI search visibility, and what are GEO and AEO?

You will see this work called generative engine optimization (GEO) or answer engine optimization (AEO). The two terms describe the same goal and are used more or less interchangeably. I use the plainer phrase, AI search visibility, because the outcome is what matters: when a buyer asks an AI which tool to choose, are you in the answer or not.

The distinction from search rankings is not academic. A ranked list gives ten options and lets the buyer choose. An AI answer usually gives three or four, already filtered, often with a reason attached to each. Being absent from that shortlist is not the same as ranking eleventh. It is closer to not existing.

This matters more than it used to because the answer increasingly ends the search rather than starting it.

Read those numbers carefully, because they cut both ways. Clicks fall, which is the part everyone quotes. But the mention itself is now doing the persuading, whether or not anyone clicks. If your brand is named and described accurately inside the answer, you have influenced the shortlist even when you get no traffic at all.

How is AI search different from SEO?

For a while the two overlapped enough that good SEO carried you into AI answers by default. That link has weakened sharply.

Even allowing for that methodology caveat, the direction is unambiguous. The majority of citations now come from pages that are not sitting in the top ten for the question being asked. Ranking is one input among several, not the gate.

The mechanism behind that shift is worth understanding, because it explains most of what follows. When you ask an AI engine a question, it does not usually run your question as a single search. It decomposes it into a set of narrower sub-questions, retrieves for each, and assembles an answer from what comes back. The industry calls this query fan-out.

So the question the engine actually searched is rarely the question the buyer typed. You are not competing for one keyword. You are competing across a spray of related sub-questions you never see, any one of which can be where your competitor gets picked up.

Three practical consequences follow:

• Breadth of coverage across a topic beats depth on a single page. A brand discussed across many adjacent angles gets caught by more sub-queries.
• Keyword position is a poor progress metric. You can gain citations while rankings sit still, and lose them while rankings hold.
• Pages with little or no traditional search visibility can still be cited heavily, which is why source selection looks strange to anyone reading it through an SEO lens.

How do AI engines decide which brands to cite?

It helps to separate those layers, because they respond to completely different kinds of work and on completely different timescales.

The model's prior knowledge

Some of what an engine says about your category is baked into the model itself, absorbed during training. This is the slowest layer to influence and the one nobody can edit directly. It shifts only as models retrain and as the wider written record about your category changes. Anyone promising to change what a model inherently believes about you on a short timeline does not understand this layer.

The retrieval layer

Most modern answers also fetch live sources at the moment of asking, then write the answer from those. This layer is far more responsive, and it is where most legitimate work pays off. If the sources retrieved for the sub-questions in your category mention you, you can appear in the answer within days of those sources changing.

Corroboration

This is the part most people underestimate. Engines are built to avoid stating things that only one source claims, particularly promotional things a brand says about itself. A claim that appears in one place, on your own domain, is weak evidence. The same claim, phrased consistently, appearing across several independent sources, becomes something the engine will repeat.

That is the whole game in one sentence: you are not optimising a page, you are building agreement.

Why do ChatGPT, Perplexity and Gemini name different brands?

This is measurable, and the differences are not subtle.

Look at what those numbers imply. The single largest source for one engine contributes almost nothing to another. A brand with a thorough, well-cited encyclopedia presence and no community footprint will show up in one engine and vanish in the next. Reverse that profile and you get the mirror image.

This is why a single visibility score is misleading. There is no one AI search. There are several, they disagree, and they need to be measured separately. It is also why "we did well in ChatGPT" tells you very little about the other two thirds of the market.

One more thing worth internalising: the most-cited sources on that list are not owned by the brands being named. They are independent platforms. Which leads directly to the least comfortable part of this work, further down.

What does a brand that gets cited actually look like?

These are the properties I look for when I audit a brand. They are principles rather than tactics, and they hold across all three engines even as the specifics change underneath.

• Entity clarity. The engine can state what you are, who you serve, and which category you belong to, without hedging. Brands that describe themselves in invented category language tend to be summarised vaguely or skipped, because the engine cannot place them against the question.
• Consistent description. The same core claims, phrased compatibly, wherever you appear. Contradictions between your own site, your profiles and third-party write-ups give an engine reason to distrust all of them.
• Independent corroboration. Claims about you exist somewhere other than your own domain, written by people who do not work for you.
• Presence where the category is discussed. Not everywhere. Specifically where the engines serving your buyers are demonstrably drawing from, which differs by engine and by category.
• Checkable specifics. Concrete, verifiable statements outperform superlatives. "Integrates with X, Y and Z" is quotable. "The leading platform" is not, and engines increasingly strip that kind of language out.
• Recency. Some engines weight fresh sources heavily. A category presence that was accurate three years ago and has not been touched since decays quietly.

Notice that only two of those six are things you can do on your own website. That ratio is the single most important thing to understand before budgeting for this work.

What can you actually do on your own site?

The on-site layer is the part of this work that most closely resembles familiar SEO practice, and the part you can start on today without help. In broad terms it means answering the real questions your buyers ask, directly and in your own words, rather than burying the answer under preamble. It means stating what you are in language your category actually uses. It means clean, accurate structured data so machines can parse who you are and what you offer without guessing. It means keeping your factual claims current and consistent.

Structured data deserves a specific note, because it is oversold. Marking up your pages helps engines parse and attribute you correctly. It does not buy you a citation. I have seen immaculately marked-up sites with no AI presence whatsoever, because nothing outside the domain corroborated anything they said.

The honest ceiling: your own site is one source among many, and the engines deliberately weight independent agreement above self-description. Getting the on-site layer right moves you from "unquotable" to "quotable". It does not move you to "quoted".

Why does most of the work happen off your own site?

This is the part of the work that gets glossed over in most articles on this topic, usually because it is the part that is genuinely hard.

The principle is straightforward: you are trying to become part of the record that engines consult when someone asks about your category. The execution is not straightforward. It means earning presence on independent sources, which by definition cannot be bought or shortcut without the kind of manipulation that gets brands filtered out later. It means doing that across several distinct source types, because the engines have different diets. It means the sources have to be the ones actually feeding answers in your specific category, which is an empirical question, not a guess.

I am deliberately not publishing which sources I target, in what sequence, or how I prioritise them. That mapping is category-specific, it changes as the engines change, and working it out is most of what clients pay me for. What I will say plainly is this: the brands that win here treat it as an ongoing operating discipline, not a project with an end date.

It is also where the return is. The traffic that does arrive from AI search behaves differently from search traffic, because the visitor has usually already been briefed on their options before they land.

How long does it take to get cited in AI search?

I would rather lose a deal at this paragraph than at month four, so here is the unvarnished version.

Three things set the pace, and none of them can be rushed:

• Source refresh. Changes to your own site can be picked up quickly. Changes to independent sources have to be made, then indexed, then retrieved, then chosen. Each step adds delay.
• Consensus takes repetition. A single new mention rarely shifts an answer. Agreement across sources is what moves it, and agreement accumulates slowly by nature.
• The engines keep moving. The Ahrefs figures above dropped from 76% to 38% in about seven months, partly because Google changed the model behind AI Overviews. Work tuned to last year's behaviour degrades without attention.

That last point is why this is not a project you finish. Visibility that is not maintained decays, both because your sources go stale and because the engines revise how they select them. Anyone quoting you a fixed timeline to a fixed outcome is either inexperienced or not being straight with you.

What you should expect instead is a measurable trend line: a starting score, monthly re-measurement, and a number that moves in the right direction with clear attribution for why. If it stops moving, the plan changes.

How do you measure AI search visibility?

This is the most common place I see people fool themselves. Someone asks ChatGPT once, sees their brand, and concludes they are visible. Someone else asks once, does not see it, and concludes the work failed. Both conclusions are unsupported, because these systems are probabilistic. The same question asked ten times produces a distribution, not an answer.

Useful measurement therefore has a few non-negotiable properties:

• Volume. Enough runs per question that you are reading a rate, not an anecdote.
• Per-engine separation. Scores for ChatGPT, Perplexity and Gemini reported separately, never blended, given how differently they source.
• Real buying questions. The questions your buyers actually ask, not the ones that flatter you. This is a discipline problem more than a technical one.
• Share, not position. The metric is what proportion of answers you appear in, and how you are described, not where you sit in a list.
• A stable baseline. Measured the same way every month, or the trend is meaningless.

Describing yourself as named, cited or absent is also worth separating. Being mentioned in passing is not the same as being cited as the source, and the second is considerably more valuable.

Should you do this yourself or hire someone?

Everything in this guide is knowledge, not a secret. A capable in-house marketer who reads widely can work out the principles. What is harder to sustain is the operating rhythm underneath them: measuring properly every month across three engines, keeping a map of which sources actually feed your category, doing the off-site work consistently for two or three quarters before it compounds, and noticing when an engine changes behaviour and quietly undoes some of your gains.

That is a real ongoing job. Most in-house teams start it, get a promising first month, then lose it to whatever is on fire in quarter three. The work does not fail because it was misunderstood. It fails because it was not maintained.

So the honest decision rule: if you have someone who can own this every month and will not be pulled off it, do it in-house. If you do not, hire it out, whether that is me or somebody else. What does not work is doing it for six weeks and concluding it does not work.

Common questions

Is generative engine optimization the same as answer engine optimization?

Effectively yes. GEO and AEO are competing labels for the same goal, getting a brand named inside AI-generated answers. The distinction people sometimes draw, that AEO covers direct-answer formats and GEO covers generated prose, has not held up in practice. Do not spend time on the taxonomy.

Does blocking AI crawlers protect my content?

It stops you being cited. That is the trade. If your business depends on being recommended to buyers, blocking the crawlers that decide who gets recommended works directly against you. If your business is publishing content that AI answers would substitute for, the calculation is genuinely different. Decide deliberately rather than by leaving a default in place.

Do I still need traditional SEO?

Yes. Search is still where a large share of buyers start, and a well-structured, well-linked site remains part of what makes you retrievable. Treat AI visibility as an additional surface with different mechanics, not a replacement programme.

Can I pay to appear in AI answers?

Not in the organic citation set, and attempts to fake corroboration tend to get filtered once detected. Paid placements in AI products exist and are expanding, but they are advertising, and they are separate from the citations discussed here.

What if the AI says something wrong about my brand?

This is more common than most brands realise, and it is a visibility problem rather than a separate one. Incorrect descriptions usually trace back to stale or contradictory sources the engine is drawing on. The fix is the same discipline: find what is feeding the error and correct the record at the source, which takes the same months as everything else here.