AI Search Optimisation
A growing share of your customers now ask an AI assistant before they search. The assistant answers, cites three or four sources, and the buyer clicks one. If you are not among them, you were not in the consideration set.
A rendering check of what a non-JavaScript crawler receives from every important template · An entity audit across Companies House, LinkedIn, Google Business Profile and the directories you appear in · Organisation, LocalBusiness, Service, FAQPage and Article structured data, validated against the visible page · A substantial About page stating plainly what the company is, what it does, where it is and since when
On this page
- How do answer engines actually choose sources?
- What does a page need in order to be cited?
- What makes a page quotable by an answer engine?
- What do we actually do?
- Did we have this problem on our own website?
- What can nobody promise you?
- Can anyone guarantee you a citation in an AI answer?
- Does this replace normal SEO?
- What drives the cost of AI search optimisation?
How do answer engines actually choose sources?
They retrieve. When you ask an assistant a question about the current world, it usually does not answer from memory: it runs a search, retrieves a handful of pages, reads them, and writes an answer citing what it used. That process is called retrieval, and everything that follows is about being one of the pages it retrieves and can quote.
There is a lot of noise in this area, so it is worth saying plainly that none of the mechanism is exotic. Most of it is good SEO practice applied with a machine reader in mind, which is why we normally deliver it inside an ongoing search engagement rather than as a separate product.
The point
Most of it is good SEO practice applied with a machine reader in mind, which is why we normally deliver it inside an ongoing search engagement rather than as a separate product.
What does a page need in order to be cited?
Four conditions, and missing any one of them makes the rest irrelevant. The page has to be retrievable by the search the assistant actually runs, readable by a machine that may not execute JavaScript, written so the answer can be extracted directly, and attached to a business the model is confident it can identify.
- Your page has to be retrievable. It must be indexed, and it must be findable by the search the assistant runs — which is often not the phrasing the user typed.
- Your page has to be readable by a machine that may not run JavaScript. This is the one that quietly disqualifies a great many modern websites, and it is usually a rendering and crawlability fix rather than a content one.
- Your content has to answer the question directly. Assistants extract answers. A page that states the answer in the first two sentences under a clear heading is far easier to quote than one that builds to a conclusion over six paragraphs.
- Your business has to be a recognisable entity. The model needs to be confident about who you are, where you are and what you do, from consistent signals across multiple sources.
What makes a page quotable by an answer engine?
A page gets quoted when a machine can lift a clean, checkable answer out of it without reading the whole thing. That means a question as the heading, the answer immediately underneath, real structure rather than a description of one, and specifics only you could have written. Every item below is something you can apply yourself.
- Make the heading the question. Write it as the sentence a buyer would type or say — “What does a bathroom fitting job usually involve?” rather than “Our process”. The heading is what a retrieval system matches against, and a category label matches nothing at all.
- Answer it in the first forty to sixty words underneath. No run-up, no “there are several factors to consider”. State the answer, then argue it. A reader who saw only that block should already have what they came for, which is also what stops a visitor leaving before they enquire.
- Build the actual table rather than describing one. A paragraph saying costs vary by size, complexity and platform gives a machine nothing to lift. Three columns and six rows gives it something it can quote intact and attribute to you, and gives a reader something they can scan.
- Put in the specifics only you have. What you found on the last job of this kind, what a supplier actually quoted, what went wrong and what it took to put right. A model can paraphrase generic advice from a thousand sources, so first-hand detail is the part it cannot get elsewhere — which is why the work you can show is worth writing up properly.
- Show the date, and let it be honest. A visible last-reviewed date helps a machine judge whether the page is current. Backdating it, or refreshing it without changing anything, is how a source gets treated as unreliable once its facts stop matching what is on the rest of the site.
- Keep the facts agreeing with each other. If your service page says one thing and your business listings and profiles say another, a model hedges instead of citing. The same company name, the same address, the same description of what you do, everywhere it appears.
- Write sentences that survive being lifted out. Assume any sentence may appear on its own with your name attached and nothing around it. “We are passionate about excellence” says nothing in that position; “we do not publish prices because scope decides them” says something a person can act on.
- Make the structured data match what a human sees. FAQ markup containing questions that are not on the page, or a star rating nothing displays, is worse than no markup at all. Mark up only what is visible, and validate it before you ship it.
- Get the text into the HTML rather than assembling it afterwards. Many answer-engine crawlers do not run JavaScript, so anything drawn in the browser after page load simply is not there for them. Where text has to be moved into the HTML response, that is a build change rather than a writing one.
- Name who wrote it. A page attributed to a real person with a real role is easier for a model to treat as a source than an anonymous one, and it is considerably harder to fake.
What do we actually do?
Seven workstreams, in roughly the order they matter: rendering, entity, content structure, structured data, factual writing, presence in the sources models read, and measurement. The first one blocks everything else, because a crawler that receives an empty shell will never reach any of the six that follow it.
- Making your site machine-readable. The foundation, and the most common blocker. If content is drawn by JavaScript after page load, a crawler that does not render sees an empty shell. Google renders; many AI crawlers do not. We check what is actually in the HTML response and fix it where it is missing — and where the fix lives in the application rather than the content, the same team handles the build work itself.
- Entity establishment. Getting a machine to be certain who you are. One consistent business name and legal entity everywhere. Organisation schema with founding date, founder, address and links to your genuine profiles. A substantial About page that states plainly what the company is, what it does, where it is and since when. Consistent details across Companies House, LinkedIn, Google Business Profile and every directory — the same consistency that map pack and citation work depends on. Assistants triangulate, and inconsistency makes them hedge; a hedged model does not cite you.
- Question-shaped content architecture. Restructuring pages around the questions buyers actually ask, with the answer stated immediately under each heading. This is simultaneously better for human readers, which is why it is worth doing regardless.
- Structured data. FAQPage, Organization, LocalBusiness, Service, Article with a named author and dates. Marking up only what is visible on the page, validated properly.
- Citable, factual writing. Assistants favour specific, checkable content. Figures, timeframes, price ranges, named technologies, dated information. “We are passionate about delivering excellence” is unquotable. “A custom CRM in the UK typically costs between X and Y, and takes Z weeks” is exactly what gets extracted. Our own cost and comparison guides are written that way for the same reason.
- Presence in the sources models read. Assistants draw on more than your website. Directories, review platforms, chamber listings, industry publications, forums and community sites. Being consistently described across those sources is part of being a confident entity.
- Measurement. Segmenting referral traffic from AI assistants in GA4, tracking branded search volume as a proxy for AI-driven discovery, and periodically testing the questions that matter to you by asking the assistants directly and recording whether you appear.
Did we have this problem on our own website?
Yes, and we fixed it. In September 2026 we audited our own website. Every one of the 47 URLs in our sitemap served byte-identical HTML to a crawler that does not run JavaScript: the same title, the same meta description, no h1 at all, fifteen words of body text, and no structured data on any page.
An agency selling answer engine optimisation, with a website that answer engines could barely read.
We are telling you this for three reasons. It is true. It is a common failure that almost nobody checks for, so it may well apply to you. And an agency that will not admit its own mistakes is not going to give you an honest assessment of yours. We fixed it, and we will show you the before and after when we speak.
What can nobody promise you?
No agency can guarantee that your page will be cited by ChatGPT, Gemini, Perplexity or Google's AI Overviews. There is no submission process, no ranking to buy, no relationship to leverage. Retrieval varies by question phrasing, by model, by version, and between two people asking the same thing minutes apart.
What can be done is real and worth doing: make the site readable, make the entity unambiguous, structure the content so answers are extractable, and be consistently described wherever models look. That measurably improves your chances. It is not a guarantee, and we will not dress it up as one.
We would also encourage scepticism about the size of the opportunity. AI referral traffic is currently small for most businesses compared with ordinary search. It is growing, and the work overlaps almost entirely with good SEO practice, which is why we think it is worth doing now. But if someone tells you AI search has replaced Google for your customers, ask them for the data.
The point
AI referral traffic is currently small for most businesses compared with ordinary search.
Can anyone guarantee you a citation in an AI answer?
No. There is no submission form, no ranking to buy and no way to control retrieval. If an agency promises you a citation in ChatGPT or Google's AI Overviews, ask them to explain the mechanism by which they would deliver it.
Does this replace normal SEO?
No, and be wary of anyone who says it does. Almost everything on this page is an intensification of things that were already good practice: technical health, entity clarity, structured data, clear writing, authority. The work overlaps so heavily that we generally deliver it inside an SEO engagement rather than as a separate product.
Where it genuinely differs is emphasis. Answer engines care more about extractability and entity confidence than traditional ranking did, and they are far less forgiving of a site they cannot read without executing JavaScript. If you are deciding where to start, the honest sequence is the technical foundations first, then the entity, then the content — the same order a full SEO programme would follow anyway.
The point
Almost everything on this page is an intensification of things that were already good practice: technical health, entity clarity, structured data, clear writing, authority.
What drives the cost of AI search optimisation?
Most of it is decided before anyone writes a line of schema: how many templates have to be checked, whether content is drawn by JavaScript and has to be moved into the HTML response, how inconsistently your business is described across the web, and how many pages need restructuring around real questions. Implementation is scoped after the review.
| Service | What it is |
|---|---|
| Visibility review | Rendering and crawlability assessment, entity audit across the web, structured data review, content extractability analysis, and a baseline test of how assistants currently answer questions in your market. |
| Implementation | Rendering fixes, entity work, schema implementation, content restructuring. Scoped after the review. |
| Within a retainer | Delivered as part of the Growth SEO plan and above, which is how most clients take it. |
How We Work
Rendering and retrievability check
What a crawler that does not execute JavaScript actually receives, and whether your pages are indexed and findable by the search an assistant would run — which is often not the phrasing the user typed.
Entity audit
How consistently your business is described across Companies House, LinkedIn, Google Business Profile, review platforms and directories. Assistants triangulate, and inconsistency makes them hedge.
Baseline assistant test
We ask each major assistant the questions that matter in your market and record whether you appear and how you are described, before any work starts.
Fix the foundations
Rendering fixes so content exists in the HTML response, Organisation and LocalBusiness schema, a substantial About page, and identical details everywhere.
Frequently Asked Questions
What is answer engine optimisation and how is it different from SEO?
Answer engine optimisation is making your business easy for an AI assistant to find, read, understand and cite when it answers a question. It differs from ordinary SEO mainly in emphasis: the goal is being quoted rather than ranked, which puts more weight on machine-readability, direct answers and entity clarity. The underlying work overlaps heavily.
What is the difference between AEO and GEO?
Very little in practice, and the terms are used interchangeably. Answer Engine Optimisation usually refers to being cited in AI answers generally; Generative Engine Optimisation is sometimes used for generative systems specifically. We treat them as one discipline on one page, because splitting them across two thin pages helps nobody.
How do I get my business to appear in ChatGPT answers?
Be findable, readable, quotable and unambiguous. Make sure your content exists in the HTML rather than only after JavaScript runs, state answers directly under question-shaped headings, add Organization and FAQ structured data, keep your business details identical everywhere, and be present in the directories and review sites models read. There is no submission form.
Can you guarantee my site will be cited by an AI system?
No, and nobody can. There is no ranking to buy and no way to control retrieval, which varies by phrasing, model and version. We can materially improve your chances and we can measure whether it worked. We cannot promise placement, and an agency that does is telling you something untrue.
Does AI search optimisation replace normal SEO?
No. It is largely the same work with different emphasis, and we usually deliver it inside an SEO engagement. If AI search matters to you, the honest sequence is: fix the technical foundations, establish the entity, restructure the content — all of which improve your ordinary search performance at the same time.
How do I measure traffic coming from AI search?
Referrals from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com can be segmented in GA4. Branded search in Search Console is a useful proxy for people who discover you through an assistant and then look you up. We also ask the assistants the questions that matter in your market and record whether you appear.
Why does JavaScript rendering matter for AI crawlers?
Because many of them do not execute JavaScript. They request the page, read whatever HTML comes back, and move on. If your content is drawn in the browser, what they read is an empty shell. Google is more capable here, which is why a site can rank acceptably and still be invisible to assistants.
Is it too early to bother with this?
It would be if the work were separate. Because it is largely the same work that improves ordinary search — technical health, structured data, clear answers, entity consistency — the sensible position is to do it now and benefit in both places. What we would not advise is paying a premium for it as a bolt-on.
Let's Build Something That Matters.
Tell us what you're trying to achieve and we'll help you find the right solution.