AI assistants are becoming the first step in the buying journey. An assistant can only recommend a business its market has already learned to describe, which makes this a positioning problem before it is a technical one.
Key judgement
You cannot optimise your way into an answer the market never learned to give.At a glance
Salesforce reported on 28 July 2026 that purchase journeys starting in AI chat grew 200% year on year.
Between August 2025 and May 2026, Salesforce measured traditional search down 15% as a discovery route and brand-owned properties down 7%, while AI assistants and other new channels rose 38%.
An assistant summarises what is already written and repeated about a business, so a consistent claim matters more than page-level tags.
Generative engine optimisation extends a clear position. It cannot manufacture one.
What has changed in how customers find businesses?
The first step in a buying journey is moving from a search box to a conversation. Salesforce reported on 28 July 2026 that purchase journeys beginning in AI chat grew 200% year on year, drawing on behavioural analysis of 1.5 billion shoppers across 37 countries between the first quarter of 2024 and the first quarter of 2026, alongside a survey of 3,450 commerce professionals in 20 countries.
The same Salesforce analysis measured the direction of travel between August 2025 and May 2026: traditional search fell 15% as a discovery route and brand-owned properties fell 7%, while newer channels including AI assistants, social platform AI and delivery apps grew 38%. Discovery is not disappearing. It is moving to places a business does not own and cannot directly edit.
That last point is the strategically important one. When a customer searched, a business could improve the page they landed on. When a customer asks an assistant, the business is not in the room. What arrives is a summary, assembled from sources the business did not write, delivered with no opportunity to clarify.
Can you optimise your way into an AI answer?
Not on its own. The reflex when discovery shifts is to treat it as a technical problem and buy a technical fix, and an industry has grown quickly around generative engine optimisation to sell one. Schema markup, answer-first formatting and clean semantic HTML all genuinely help an engine retrieve and quote a page, and this site is built to that standard. They decide how well a claim travels. They do not decide whether there is a claim worth carrying.
An assistant asked to recommend a brand strategist, a bookkeeper or a bakery does not rank pages. It produces a description of each candidate and a reason to prefer one. If the market describes a business in four different ways, the assistant either picks one at random, blends them into something vague, or omits the business in favour of one it can characterise confidently. No amount of markup resolves an ambiguity that exists in the market rather than in the code.
Where does an AI assistant get what it says about your business?
An assistant answers from two places: what it absorbed during training, and what it retrieves at the moment of asking. Neither is your website alone. Training data is drawn from a very wide crawl of the public web, and retrieval pulls whichever passages the model judges most relevant and trustworthy, which will often be a directory, a publication, a marketplace listing or someone else's article rather than your own homepage.
This is why consistency outranks polish. A business described the same way across its website, its LinkedIn profile, its listings, its press mentions and its clients' own words gives a model a stable, corroborated pattern to repeat. A business described differently in each place gives the model a choice, and models resolve choices by favouring whichever version is most frequently and most confidently stated elsewhere. That version may not be the one the founder would choose.
What makes a business retrievable by an AI assistant?
Three conditions make a business retrievable, and they are strategic rather than technical. First, one claim: a single sentence stating who the business is for and why it is the right choice. Second, repetition: that claim stated the same way everywhere the business appears. Third, evidence: named credentials, specific results or verifiable facts attached to the claim so a cautious model has something concrete to cite.
The third condition is the one most often missed. Language models are trained to hedge, and an unattributed claim is easy to drop. “We deliver outstanding results” gives a model nothing it can safely repeat. “CIM-accredited brand strategist working with founder-led UK businesses of one to ten people” gives it a specific, checkable statement it can attribute. Specificity is not a stylistic preference here. It is what makes a passage safe enough to quote.
This is the same discipline that makes a position work with human buyers, which is the point. Nothing about AI discovery requires a separate strategy. It raises the cost of not having one, because a vague position that a patient human might have puzzled out is simply skipped by a machine choosing between candidates.
Define the one claim your market can repeat.
Brand Positioning establishes a defensible market position, a positioning statement and the messaging pillars that carry it consistently wherever the business appears.
Explore Brand Positioning →How is this different from search engine optimisation?
Search engine optimisation earns a ranked link and is measured in clicks. Visibility in an AI answer earns a citation or a recommendation and is measured in whether the business is named, described correctly and preferred. The two disciplines share techniques but reward different things, and a business can be strong at one while invisible in the other.
| Search-engine visibility | AI-answer visibility | |
|---|---|---|
| Unit of success | A ranked link, then a click. | A citation, a correct description, a recommendation. |
| What is optimised | The page. | The claim, wherever it is repeated. |
| What decides the outcome | An index and a ranking algorithm. | A model synthesising retrieved passages. |
| What is rewarded | Relevance and authority signals on your own site. | Consistency and corroboration across many sources. |
| Where it fails | The page is thin, slow or poorly matched. | The market describes the business in several different ways. |
| Who controls it | Largely the business. | Partly the business, mostly everyone else. |
Both still matter. The practical error is sequencing: treating AI visibility as a technical project to be started after the positioning work has been indefinitely postponed.
What should a founder do this quarter?
Five steps, in order, and the first four cost nothing but judgement. The sequence matters more than the speed: each step is wasted if the one before it has not been done, which is why the technical work sits last rather than first.
- Ask the assistants what they already say. Put “what does [your business] do, and who is it for?” to ChatGPT, Claude, Gemini and Perplexity. Record the answers verbatim. This is your baseline, and it is usually the moment the problem becomes obvious.
- Write the one sentence. Who the business is for, what it does, and why it is the right choice. One sentence, not a paragraph. If it cannot be said in one sentence, that is the finding.
- Make it consistent where you control it. Website, LinkedIn, directories, proposals, email signature, listings. Same claim, same words. Variation reads as creativity to a person and as uncertainty to a model.
- Attach evidence to the claim. Credentials, named results, specific scope, real figures. Give a cautious model something concrete enough to repeat.
- Then do the technical work. Structured data, answer-first page structure, an
llms.txtfile, clean semantic markup. Genuinely useful, and now carrying something worth retrieving.
Re-run the first step a quarter later. It is the closest thing to a measurement this discipline currently offers, and it costs ten minutes.
What this evidence does not show
Two honest limits, because the evidence here is newer and thinner than the confidence around it. Salesforce's figures are drawn from commerce and retail behaviour, so a UK service business should read them as direction rather than as a forecast of its own numbers. The shift is real and the magnitude is not transferable.
The second limit concerns the argument rather than the data. WARC published a discussion on 28 July 2026, featuring Alex Brownsell, Dr Karen Nelson-Field, Carl Wåreus and Scott Reijinders, holding that brand equity drives visibility in AI recommendations because “equity can't be built without memory, and memory can't be built without attention”. That is a persuasive argument and it matches what is observable, but it is a recorded discussion rather than published research, and no quantified study yet establishes the link. It should be weighed as expert reasoning, not cited as proof. Anyone selling certainty about AI visibility in 2026 is selling ahead of the evidence.
Start with the claim
Make your business easy to describe, and easy to recommend.
The £200 Focus Consultation answers one defined brand or marketing question in writing within two working days, with baseline competitor context, so you can test the thinking before commissioning a full engagement.
What this means for your business
- Establish what assistants currently say about the business before investing in any technical AI-visibility work.
- Treat inconsistency in how the business is described as the primary problem, not a cosmetic one.
- Attach specific, checkable evidence to the central claim so a cautious model can safely repeat it.
- Sequence the work: position first, consistency second, technical implementation last.