Ask an AI service to recommend a local business and the answer may sound as though it has carefully evaluated the options. It can describe a company, explain why it might fit and present a confident list of suggestions.

The language is not the evidence. To understand what the answer can actually establish, I start with the public information behind it and the way the service is being used.

For a small business, that distinction is more useful than asking whether “AI knows the brand”. A name appearing in an answer, an accurate description and a justified recommendation are separate events.

A generated answer is not a business record

A business has facts that can be checked: its trading name, current services, location, contact details and the people responsible for it. Those facts may appear on the website, business listings, professional profiles or other public sources.

An AI answer can reproduce, combine or misstate that information. Its polished wording does not make it the authoritative record of the business.

When a service uses web search, the visible sources offer a starting point for checking the answer. OpenAI’s explanation of web search in ChatGPT describes how search and linked sources can form part of the experience. The presence of sources is helpful, but the reader still needs to inspect what they support.

I would not treat an answer without visible support as proof of the company’s current offer. I would look for the business’s own current information and other relevant evidence before acting on consequential details.

Availability of information is different from retrieval

A useful page can be publicly available without appearing in a particular response. A service may not retrieve it for that question, and access does not guarantee selection.

Google’s guidance for AI features explains that established search requirements and helpful content remain relevant. It does not prescribe a special file or markup that guarantees appearance in an AI experience.

Similarly, OpenAI documents different crawler purposes and controls. Allowing a relevant crawler is an access decision, not a contract that the brand will be recommended.

This matters when diagnosing an absence. “The brand did not appear in this answer” does not tell us whether the problem is an inaccessible page, insufficient relevant information, the wording of the question or something else. Those possibilities require separate checks. My practical checklist for making a website ready for AI search explains the site-level checks that support that diagnosis.

Public facts need to agree with the real business

Suppose an illustrative repair company serves customers by appointment but an old directory describes it as a walk-in shop. Its website has the correct arrangement, while a social profile still lists an earlier address.

The company has not created one clear public account. It has created several sources that disagree. Even before considering AI, a customer can be misled by choosing the wrong page.

The first improvement is to verify the facts with the business and correct important public errors. It is not to repeat the preferred description more aggressively in prompts.

For eligible Google Business Profiles, Google’s business representation rules provide platform-specific guidance. Other public pages still need their own accuracy review. Matching words across sites is useful only when the underlying information is correct.

A citation supports a particular claim, not the entire answer

When an answer cites a page, open it. Check whether the page supports the nearby statement and whether the information is current enough for the decision.

A source may establish that a business offers a service without establishing that it is the best option. It may describe a location without confirming delivery coverage. The recommendation can contain an additional inference that the source does not directly support.

I separate the answer into claims before judging it. Which statements concern identity? Which concern capabilities? Which compare the brand with alternatives? The evidence required becomes more demanding as the answer moves from description to preference.

This is also why a business should avoid celebrating every citation as a complete endorsement. The link may be useful, but its role needs to be understood.

Recommendation questions contain hidden conditions

“Recommend a photographer” leaves out the job, location, budget, style and timetable. A more specific question can produce a different set of relevant considerations.

That does not mean a business should keep changing the question until it receives a flattering answer. It means the audit should record the context that makes an answer useful or irrelevant.

A brand may be suitable for a small product shoot and unsuitable for a large event. A fair public description should make that scope clear rather than encouraging every system and reader to treat the offer as universal.

The practical work is to answer real buyer questions with accurate detail. A specialist service page, a clear comparison or an original example can help readers understand fit even when no AI outcome is guaranteed. That context is central to understanding how AI discovers and recommends a local business.

What I would check first

I would verify the business facts, inspect the important public pages and confirm that the relevant information is accessible. Then I would document a small set of questions tied to real customer decisions.

For each observed answer, I would record the exact question, date, service, relevant settings and visible sources. I would distinguish a mention, an accurate description, a recommendation and a cited source.

I would not collapse those observations into a universal ranking or use one favourable response as proof of market-wide visibility. The record should preserve uncertainty rather than edit it out.

This is the thinking behind the work I am developing through Sparlab: help businesses organise useful public information and assess what can be observed. A platform can support that work, but it cannot make a third-party answer behave like a guaranteed placement.

The useful question is narrower

Instead of asking whether AI knows your business, ask whether the public information needed for a particular customer decision is accurate, accessible and supportable.

Then ask what a specific observed answer got right, what it got wrong and which sources explain the result. Those questions lead to work a business can carry out and review.

A convincing paragraph is easy to overvalue. A clear record of facts, sources and limitations is a better foundation for both the customer and the person responsible for the brand.