AEO and GEO: getting found in AI search.

What changed in B2B buyer research

A growing share of B2B research now starts as a question to ChatGPT, Claude or Perplexity ('best B2B web design agencies in Leeds', 'Duda vs Webflow for an agency site') rather than a Google query. The engine answers with a synthesis and a handful of citations. If you are not in the citation set, you were never considered.

The economics differ from SEO: volumes are lower but the engine pre-qualifies and pre-sells. Buyers arriving from AI answers convert at noticeably higher rates because they arrive shortlisted.

The structural change is that the shortlist now happens before you know the buyer exists. In search, a buyer saw ten results and chose; you could win from position six with a better page. In an AI answer, three names are given and the rest are not mentioned. There is no position six to improve from, which is why this is a visibility problem rather than a ranking one.

What is the difference between AEO, GEO and SEO?

SEO optimises for a ranked list of links. AEO — answer engine optimisation — optimises for being the source an engine quotes when it answers a question directly. GEO, generative engine optimisation, is the same discipline named for generative assistants specifically. In practice the terms are used interchangeably and the work overlaps heavily.

They are not competing strategies. The pages that earn AI citations are, almost without exception, pages that were already good for search: clear structure, real information, credible sourcing. What changes is the emphasis — away from keyword coverage and towards factual precision, entity naming and self-contained answers.

Is AI search actually sending traffic yet?

In B2B it is sending less traffic than search and better traffic. The honest position is that volumes are still small in most sectors and growing, and that anyone quoting confident industry-wide percentages is guessing, because the engines do not publish referral data consistently and much of it arrives unattributed.

The reason to act now is not the current volume, it is that the work is the same work you should be doing anyway. Answer-first pages, consistent facts and real FAQs improve conversion for human readers regardless of whether an engine ever quotes them. That makes this an unusually low-risk bet: the downside case is that you have a clearer website.

What AI engines actually cite

Engines quote pages they can parse into confident statements. In practice that means: the first sentence under a heading answers the heading, facts (prices, locations, client names, metrics) are stated plainly and consistently across the site, FAQs are written as real questions with self-contained answers, and the site declares itself via schema and llms.txt.

Marketing prose is the enemy. 'We deliver transformative digital experiences' gives an engine nothing to quote. 'B2B websites from £4,500, built in Leeds, with case studies at +128% organic sessions' gives it everything.

The test to apply to any paragraph you write is whether it survives being lifted out of the page. If a sentence only makes sense after reading the two above it, it will not be quoted, because a quote arrives alone. Self-contained is the operative property, and it is why the FAQ format performs so well: the format forces it.

• Answer-first paragraphs under every H2
• Question-shaped H3s with 2 to 4 sentence answers
• Named entities: clients, prices, places, platforms, metrics
• Consistent facts sitewide (one price, one address, one claim)
• FAQPage, Service, Organization and Article schema
• llms.txt manifest at the site root

Why does factual consistency matter so much?

Because an engine synthesising an answer is weighing sources against each other, and a contradiction is a reason to drop you. If your homepage says projects start at £5,000, your pricing page says £4,500 and a directory listing says £3,000, none of those numbers is safe to quote and the engine will reach for a competitor who states one figure everywhere.

This applies beyond price. Your company description, your location, your founding date, your service names and your headline client results should be identical wherever they appear, including on third-party profiles you do not control but can edit. Consistency is unglamorous and it is most of the work.

Does schema markup help with AI citations?

It helps by removing ambiguity rather than by ranking you. Organization schema states plainly who you are, FAQPage marks a question as a question, and Article schema establishes authorship and dates. None of it makes a weak page citable, but all of it makes a strong page easier to parse confidently.

Treat schema as the machine-readable version of facts already visible on the page. Schema that describes content the page does not contain is a liability, both with search engines and with the assistants reading them.

What is llms.txt and do I need one?

llms.txt is a plain-Markdown file at your site root that states what the site is and links to its most important pages with a one-line description each. It is a proposed convention rather than a ratified standard, and support across engines is inconsistent.

It costs an afternoon and cannot hurt, which is the whole argument for it. We ship one on every site we build, including this one, on the basis that a small file that may become important is a better bet than retrofitting it later across a hundred pages.

How to measure it

Define the prompts your buyers plausibly ask, run them across the major engines on a schedule, and record which brands get mentioned and which URLs get cited. Track share of answer by topic month over month, the way you track rankings. This measurement is the core of our AI visibility service, from £850 a month, reported alongside the actions taken.

Two numbers are worth separating, because they move independently. Mention rate is how often your brand appears in an answer at all. Citation rate is how often one of your URLs is given as a source. A brand can be well known enough to be mentioned from the model's training and still never be cited, which means no traffic and no control over how you are described.

Expect variance and do not over-read a single run. Ask the same question twice and you can get different answers, different sources and a different shortlist, because these systems are probabilistic and personalised. This is why the discipline is scheduled sampling across many prompts rather than one dramatic screenshot of a bad answer, which is the form most of these conversations start in.

What do I do if an AI describes my business wrongly?

Fix the source rather than arguing with the output. Incorrect descriptions almost always trace to stale material the engine can reach: an old about page, an outdated directory listing, a press release from a previous positioning, a Companies House record, a review profile nobody has updated in four years.

Then state the correct version plainly and repeatedly on your own site, in the simple factual register these systems quote well. Corrections take time to propagate — weeks rather than days — and there is no submit-a-correction form, which is exactly why consistency is worth maintaining before it becomes a problem.

Where to start

Restructure your highest-value service pages first: answer-first openings, real FAQs, schema. Then fix consistency (your price, address and positioning should match everywhere). Then measure. Or start with our £1,500 Action Plan, which includes an AI search readiness review of your current site against this exact checklist.

If you want a single afternoon's work with the best return, write the ten questions your sales team answers most often and publish honest, self-contained answers to them. That one exercise produces citable content, improves conversion for human readers, and gives you the prompt list to measure against. Almost everything else in this article is refinement on top of it.

Do not block the crawlers you want to be cited by. It is worth checking robots.txt explicitly, because several platforms and plugins now block AI user agents by default, and a site can be invisible to assistants for a reason nobody in the business chose.