What Businesses Should Measure When Customers Discover Them Through AI Search

A marketing manager asks an AI assistant to recommend suppliers in the company’s category. Several competitors appear, but the company does not. It is an uncomfortable result, particularly when the business has invested in its website and already attracts visitors from Google.

That single answer is a reason to investigate, but it is a poor basis for a new marketing strategy. Businesses need a repeatable way to observe AI search, understand what the results mean and connect those observations to customer activity. Otherwise, a changing set of answers can become another dashboard that attracts attention without improving decisions.

Define the customer questions worth tracking

Start with the questions customers ask while choosing a product or service. Include category research, specific use cases, comparisons, implementation concerns and buying constraints. A software company, for example, may care about recommendations for a particular team size, integration requirement or operating market.

Keep branded and unbranded questions separate. “What does this company offer?” tests whether information about an existing brand is represented accurately. “Which suppliers support this requirement?” examines whether the brand enters a buyer’s consideration set. Both matter, but they answer different business questions.

Create a manageable initial list and record why each question belongs there. Sales calls, support requests and customer interviews are useful inputs. Avoid filling the list with variations designed mainly to make the company appear more often; that produces a flattering sample rather than useful evidence.

Distinguish a mention from a citation

A mention means the brand appears in an answer. A citation means a source is linked or referenced in support of some part of that answer. A business can receive one without the other, and neither automatically means the assistant recommends it to the user.

Record the context around each appearance. Was the brand recommended for a suitable use case, listed as an alternative or mentioned in a warning? Does the answer accurately describe the offer? If a source is cited, identify whether it is the company’s website, a comparison article or another publication.

These details make the findings actionable. An incorrect feature description suggests a different investigation from absence on a relevant comparison question. A citation to an outdated third-party page may require checking the information published there before changing the company’s own website.

Treat repeated checks as a sample

AI answers can vary, so a single screenshot should not stand in for a reliable trend. Keep the wording, market and checking method consistent where possible. Record the date, the platform and the settings available to the person running the check.

Run the same questions on a repeatable schedule and keep the full responses when practical. If the question set changes, document the change rather than silently comparing the new results with the old ones. A useful report should make it possible to see whether the measurement process changed alongside the result.

Manual checks can establish a baseline before a business buys software. As the scope grows, CompareCrest’s guide to AI visibility tools can help teams compare monitoring options. Evaluate platform coverage, the questions included, geographic controls and access to the underlying responses before choosing a subscription.

Connect visibility to referral visits

The next question is whether people reach the website after using an AI service. Review referral sources in the analytics system and group identifiable AI referrals consistently. Check the reporting setup before assuming every unfamiliar source represents a new channel.

A referral visit shows an observable website interaction, but it does not reveal every influence on the buyer. Someone may see a brand in an answer and return later through a search or a direct visit. Other journeys may lose source information before the visitor arrives.

Use that limitation to interpret the numbers carefully. Report identifiable AI referrals as the visits the business can observe. Avoid presenting them as a complete count of everyone influenced by AI search, or treating the absence of recorded referrals as proof that no discovery took place.

Measure what visitors do after arrival

A visit becomes more commercially informative when the team can see the next action. Relevant events might include a demonstration request, a product trial, a quotation enquiry or a purchase. The appropriate outcome depends on the business model and the length of the buying process.

Compare landing pages and lead quality alongside visit totals. A small group of relevant visitors reaching a detailed service page may deserve more attention than a larger group reading unrelated content. Review whether the enquiries match the target customer and whether sales accepts them as suitable opportunities.

Ask new customers how they discovered the business, using an optional open question where appropriate. Their answers can add context to analytics without being treated as a perfect attribution system. Keep the original responses, because “I saw you online” and a named AI assistant provide different levels of evidence.

Audit the information behind inaccurate answers

When an answer is wrong, first check the information available about the business. Product names, pricing descriptions, supported markets and service categories may differ across old pages and third-party profiles. An outdated comparison can remain visible long after the company updates its homepage.

Create a short issue log containing the question, the inaccurate statement, the sources shown and the correct information. Then identify the likely places requiring an update. This creates a reviewable task rather than a vague instruction to “improve AI visibility.”

If the business needs support turning these findings into a programme of work, GrowthFollo’s GEO and AI search service is one option to assess. Ask for priorities tied to documented gaps, clear responsibilities and a measurement method. Progress should be evaluated through repeated observations and relevant business outcomes, without promises of a guaranteed recommendation.

Keep the technical investigation specific

Technical checks should begin with the pages customers and search systems need to access. Confirm that important information is available, that intended public pages are not accidentally excluded and that navigation connects relevant content. Investigate actual access failures before commissioning a large rewrite.

For Google’s AI search features, Google Search Central’s guidance says established SEO practices remain relevant and that special AI markup is not required. A technical proposal should therefore explain the particular problem being addressed rather than sell an unspecified “AI schema” addition.

Other platforms have their own access mechanisms and documentation. Do not assume that a crawler used for model training serves the same purpose as a crawler used for search or user-requested retrieval. Review the applicable controls before changing access settings, and verify the result for the platform being investigated.

Build a monthly report that supports decisions

Keep the report focused on a few connected measures: coverage of relevant questions, accuracy of brand descriptions, identifiable referrals and suitable customer actions. Include the sample size and any changes to the checking method. Put unusual movements in context rather than treating every fluctuation as a success or failure.

End each review with specific work: correct an outdated explanation, improve a useful comparison, investigate a blocked page or interview customers about an emerging question. Assign an owner and review the outcome later. This gives AI search measurement a practical role in marketing planning and helps the business learn which changes deserve continued investment.