AI visibility: How to check what AI assistants say about your SME

Which services does an AI assistant attribute to your business? Are the contact details and service area correct? And is your business mentioned when someone looks for a suitable provider? Targeted queries can help you check the picture these answers present. What matters is not just whether your business appears, but whether the statements are accurate and based on traceable sources.

The “KMU Digital Pulse” study by localsearch and HSLU shows how widespread this blind spot is: according to the results published on 24 September 2026, 58 per cent of Swiss SMEs have never checked how they are represented in generative AI searches. Only 21 per cent use AI tools in marketing.

These figures describe different things. Using AI in marketing does not mean knowing how your business is represented in AI answers. Conversely, you can check that representation without using AI in your day-to-day marketing work.

AI use and AI visibility are different management questions

When you use AI to draft a newsletter, you are working on your own content. The business commissions the work, reviews the draft and decides whether to publish it. When a business is described in a generative search, however, it is the subject of an answer produced elsewhere. The business does not have sole control over which information is combined or which providers are mentioned.

This distinction matters in everyday work: you can remove an incorrect sentence from a newsletter before sending it. An incorrect statement by an AI assistant must first be discovered. You may then be able to correct the underlying information. The answer itself is not under your editorial control.

The two study figures must therefore not be combined into a single rate. The available information shows neither how much the groups overlap nor whether businesses that have not checked are actually misrepresented or lose work as a result.

The management question is therefore not just: “Should we use AI?” Equally important is: “Which publicly available statements about our business are accurate, and which of them do AI assistants pick up?”

What makes generative search different

A conventional web search mainly provides results for the person searching to choose from. They open a website, compare offerings and assess the source. Generative search, by contrast, produces a connected answer. It can combine information from different origins and use it to make a selection or an assessment.

When searching for a provider, for example, a search result may lead to a business’s services page. An AI answer may already describe which jobs that business is supposedly suited to. Part of the assessment thus shifts from the reader to the system.

The two approaches have different strengths: conventional search makes it easier to check original information directly. Generative search makes it easier to bring scattered information together and allows follow-up questions. However, fluent wording can conceal the fact that a statement is poorly supported. Even a displayed source link does not automatically substantiate every sentence beside it.

The original source remains essential for current, authoritative information. A summary answer can be useful for initial orientation. Which approach fits depends on whether you are exploring options or need a reliable basis for commissioning work.

Which content can serve as a source

Not every AI assistant works the same way for every query. An answer may be based on information learned by the model, freshly retrieved web content or a combination of the two. Without web access, it is not a reliable test of what is on your website today.

For web-based searches, potential sources include publicly accessible company websites, business directories, company profiles, media reports and review platforms. Which sources were actually used can only be checked to the extent that the system discloses its sources and search activity.

Particularly relevant to your business is content that clearly describes what you offer: services, locations, service area, contact options and clearly explained limitations. A precise services page helps establish what your business is suited to. A generic advertising line does little to answer whether you can take on a specific job.

Public availability does not, however, guarantee either inclusion or accurate representation. Technical accessibility, search integration and the system’s selection also play a role. Confidential quotations, internal reports and unpublished minutes are not freely available web sources. They do not belong in public test queries either.

Use case: A quote request does not match your offering

Suppose your sales team receives an enquiry for a service your business does not offer. The prospective customer refers to an AI assistant. The person responsible first clarifies the requested work and directly corrects the expectation. That addresses the customer’s question, but the cause remains unresolved.

A factual query without customer data is suitable for checking: “Which services does this company offer, and which sources support that?” Add the company name in the assistant you are using. Then check whether the service in question is mentioned again and where the statement comes from.

If it appears on an old page of your own, the correction is your responsibility. If it comes from a directory profile, that profile needs updating. If there is no evidence, the statement remains unsupported. Asking for sources can help, but it is no substitute for checking them.

The deciding factor is the origin of the error. Outdated original information calls for content work. An incorrect summary despite accurate sources calls for documentation and further testing. Additional advertising copy does not automatically solve either problem.

Use case: A services page is being revised

Website revisions often focus on wording. For AI visibility, the task starts earlier: can someone outside the business tell from the page what work you actually take on?

Check the text with the people responsible for sales and delivery. Are the scope of services, service area and exclusions correct? Can current offerings be distinguished from former services? Then check your own pages and important external profiles for contradictions.

There are two approaches here. Promotional copy emphasises benefits and positioning. It can generate interest and give a brand a distinctive profile. A factual service description answers specific selection questions and makes it easier to check whether a job is a good fit. The two serve different purposes; neither replaces the other.

The page’s function is what matters. If it is intended to guide a specific enquiry, it needs reliable service information. If it is intended to attract attention, positioning can be more prominent. Verifiable information must still remain easy to find.

After publication, run another query with web search enabled. A lack of change does not, however, prove that the revision had no effect. Retrieval, updating and answer generation can only be understood to a limited extent from outside the system.

Use case: The marketing meeting needs a basis for decisions

The statement “We do not appear in AI” is not enough to support a decision. Without the exact question, it is unclear what was tested. A search for the company name tests something different from an open search for suitable providers.

For a check that others can follow, record the application used, the web search setting, the date and time, the exact query, the answer and the sources displayed. Check how your business is described, which services it is associated with and whether it is mentioned in a relevant provider search.

Then classify each statement that stands out: correct, outdated, incorrect or unverifiable. This produces specific tasks rather than a blanket “AI optimisation” project. An incorrect contact detail goes to whoever maintains the profiles. An unclear service description goes to the person responsible for that content. An unsupported recommendation remains an observation.

A single test does not produce a stable ranking. Answers can vary by application, question, time and conversation context. Repeating the same questions makes differences visible, but measures neither overall visibility nor lost revenue.

The first step for this week

Choose a publicly described service that matters to your business. Ask a commonly used AI assistant with web search to explain whether your company is suited to providing it and which sources support its answer. Then check the statements against your current information.

Record the result and assign any contradiction you find to a responsible person. If everything is correct, document that too. The first improvement is not a higher ranking, but replacing an unchecked assumption with a finding that can be traced and verified.