AI Search Optimisation: How to Get Cited by AI Answers

A growing share of your buyers no longer see a results page. They ask ChatGPT, Perplexity, Claude or Google's AI Overviews a question, read a synthesised answer, and only click through to the one or two sources the answer chose to cite. If your firm is not one of those sources, you were not outranked. You were simply never mentioned.

AI search optimisation is the work of making your content the source AI assistants cite and recommend when they answer questions in your market. You will also see it called answer engine optimisation (AEO) or generative engine optimisation (GEO). The label matters less than the shift: the unit of competition is no longer a ranking position, it is a citation.

Is this actually worth doing? Our own numbers

We can offer first-party evidence rather than industry projections. On our own site, visitors referred by AI assistants are the most engaged traffic source we measure, spending well over a minute on site on average against a site-wide norm of under half that. More to the point, AI referrals have produced real enquiries, five figures of pipeline and paying clients for us, with no advertising spend behind them.

The volumes are small next to organic search. The intent is not. Someone who arrives from an AI assistant's recommendation has usually already read a synthesis of their options and chosen to look closer at you specifically. Treat it as a high-intent referral channel that is still cheap to compete in, because most of your competitors have not started.

How AI assistants choose their sources

Each engine works differently, but the practical pattern behind citations is consistent enough to act on.

They have to be able to read you. AI assistants gather material through their own crawlers. If your robots.txt blocks them (some sites do this by default, or by a security plugin's decision nobody reviewed), you have opted out of the channel. The first check in any AI search audit is simply: which AI crawlers can fetch your pages?

They quote pages that answer the question directly. An AI assembling an answer wants a clean, liftable statement: a price, a step list, a definition, a criterion. Pages that open with three paragraphs of scene-setting before the substance give the engine nothing to quote. Pages that state the answer plainly, then justify it, get lifted.

They favour specifics over adjectives. "Costs typically range from £8,000 to £45,000 depending on scope" is quotable. "Competitive pricing tailored to your needs" is invisible. Numbers, dates, named criteria and honest trade-offs are what a synthesised answer is made of.

They reward question-shaped structure. Assistants answer questions, so content organised as questions and answers maps directly onto what they produce. This is why FAQ sections and question-form headings keep earning citations even where classic search has stopped rewarding the markup.

They cross-check. Claims that appear consistently across your site, your directory listings and third-party sources are safer for an engine to repeat than claims that appear once. Inconsistency (an old address, a stale price, two different descriptions of what you do) reads as unreliability.

What to actually do, in order

1. Check crawler access. Confirm your robots.txt permits the AI crawlers you care about (the ChatGPT, Perplexity, Claude and Google AI crawlers at minimum) and that your pages render as readable HTML rather than requiring JavaScript to show the substance.
2. Pick the questions you should be the answer to. Not every query, the ones where your genuine expertise gives you something original to say: what things cost, how to choose, what goes wrong, how to fix it. AI engines are synthesising judgements, and undersupplied judgement questions are where a specialist firm can win against much larger competitors.
3. Rewrite key pages to lead with the answer. Definition or verdict first, reasoning after. Add the specifics you were nervous about publishing; they are precisely what gets cited.
4. Add question-and-answer structure. Real questions, phrased as your buyers phrase them, each with a self-contained answer that survives being quoted out of context.
5. Keep facts current and consistent everywhere. Assistants penalise contradiction more silently than search ever did. When a price or claim changes, change it everywhere, including the profiles you forgot you had.
6. Measure it. AI referral traffic identifies itself in analytics (referrers from the assistant domains), and your CRM should carry an AI-referral source so you can see enquiries and revenue, not just visits. That is how we know this channel converts: we track it to won deals, not to sessions.

What does not work

Keyword stuffing has no purchase here; assistants read meaning, not density. Publishing thin "what is X" content that restates what the vendor already says gives the engine no reason to cite you instead of the vendor. And the emerging convention of an llms.txt file (a plain-text index of your key pages for AI crawlers) is cheap housekeeping, not a lever: independent testing across thousands of domains has so far found no measurable citation benefit from publishing one, and no major AI platform says it uses the file to choose citations. Publish one by all means, but nobody should sell it to you as a ranking tactic.

The uncomfortable truth is that AI search optimisation is mostly the discipline of publishing genuinely useful, specific, current content, executed technically well. There is no trick, which is good news for firms with real expertise and bad news for everyone else.

Where this fits with your SEO

It is not a replacement. The same crawlable site, fast pages and coherent structure that classic search rewards are the foundation AI engines build on, and organic rankings still carry far more volume. The right framing is one content and engineering discipline serving two kinds of results page: ranked links and synthesised answers. If your site currently earns neither, the diagnosis is usually shared causes, which is why we assess them together in a website audit that scores both search health and AI visibility. For a broader look at how your site converts the traffic it gets, book a diagnostic.

Frequently asked questions

Is AI search optimisation different from SEO?

It shares the foundations (crawlability, structure, authority) but optimises for a different output: being cited inside a synthesised answer rather than ranked on a results page. In practice the biggest behavioural difference is content style: answer-first, specific and quotable beats comprehensive and slow to the point.

How do you measure AI search traffic?

Referrer domains from the assistants show up in analytics, and a source field in your CRM lets you follow AI-referred enquiries through to revenue. Volumes are typically small, so measure enquiries and deals rather than sessions; the channel's value is intent, not scale.

Does llms.txt work?

It is an inexpensive emerging convention: a plain-text file telling AI crawlers what your site covers and which pages matter. Publishing one is sensible housekeeping, but independent testing has so far found no measurable citation benefit from it, so treat it as a low-cost hedge rather than a lever.

Can a small firm compete with big brands in AI answers?

More easily than in classic search, currently. Assistants synthesise judgements, and they cite whoever published the most useful specific answer, which is often a specialist rather than the biggest brand. Undersupplied questions in your niche are winnable now in a way that head-term rankings are not.

John Kelleher is a Claude Certified Architect (Foundations and Professional) and leads SpotDev, a Claude Registered Partner and OpenAI Select Partner.