Before buying an AI visibility tool, ask for this export

Before buying an AI visibility tool, ask the seller to export ten of the answers behind its score. You want the question, the date, the assistant, the answer and the page it cited. If the export cannot explain why the score moved, the dashboard is a weak basis for deciding what to publish next.

That does not make monitoring useless. A tool can save a marketer time, reveal an incorrect description of a business or show which sources appear for a useful set of buying questions. The purchase becomes harder to justify when its only output is a number nobody can connect to a decision.

This is a buying checklist based on documentation checked October 4, 2026. We have not benchmarked or ranked individual vendors.

First, establish what the score counts

Ask the vendor to finish this sentence: “Your visibility score is the share of ______ in which ______ happened.”

The first blank might be a fixed set of tracked questions, repeated runs or a larger estimated question universe. The second might be a brand mention, a link to your website or a favorable recommendation. Those are different measurements.

Google's guidance on third-party SEO tools says such tools do not have access to its internal ranking data and cannot guarantee performance. Treat a vendor's score as its own measurement, then inspect the method.

Here is a hypothetical example. Last month, your business appeared in 8 of 20 tracked answers: 40%. This month, the tool adds 20 harder questions and your business appears in 10 of 40: 25%. The percentage fell while the number of observed appearances rose. Neither number alone establishes that customers are less likely to find you.

For a fair comparison, preserve a fixed question set and show newly added questions separately.

Request a record you can actually challenge

Send the vendor this checklist before starting a paid trial:

Field in the export	What it lets you check
Exact question and question-set version	Whether the tool changed the comparison
Assistant, date and stated location	Which experience the observation represents
Answer text or a reproducible saved record	Whether the business was described accurately
Brand mention and cited URL in separate columns	Whether you were named, linked, both or neither
Run count and failed runs	Whether missing checks changed the denominator
Explanation of the score	Whether the headline number matches the evidence

Privacy or licensing rules may limit what a vendor can retain. Ask what substitute evidence it provides. A limitation explained before purchase is more useful than an impressive percentage with no audit trail.

Choose questions from actual sales calls, support requests and customer language. A local repair company needs coverage of its service and area. Tracking broad national “best software” questions would tell it little about its own buyers.

Keep visibility, visits and sales in different columns

A saved answer mentioning your business is evidence of that answer. It is not a visitor count.

Google Analytics now defines an AI Assistant channel for recognized assistant referrals. Google's own AI Overviews and AI Mode remain under Organic Search. The distinction comes from Google's current channel definitions, so do not expect an assistant-referral total to capture every AI-influenced discovery.

Keep three records:

1. Observed answers: what a monitoring tool or manual check found.
2. Recorded visits: what your website measurement recorded.
3. Business outcomes: inquiries, purchases or subscriptions you can verify.

Someone may see an answer and later visit directly. That creates an attribution gap. It does not justify assigning all unexplained sales to AI search. Our AI traffic measurement guide covers the reporting setup.

Fix a concrete problem before buying more coverage

If the export shows an old price, find the cited page and correct the underlying information. If your important service page cannot be found in Search, investigate its indexing and content before expanding a monitoring subscription.

Google's AI-features documentation says eligibility still rests on ordinary Search requirements. It does not require a special AI file or special structured data. That is useful protection against buying a technical fix whose claimed requirement does not exist.

Our judgment: a small team with a handful of important customer questions can begin with a dated manual record. Pay for a tool when checking those questions becomes enough work to justify automation, or when its evidence reliably changes an important content or reputation decision.

Set the renewal rule now

Before the trial, name the person who will read the report and the work they can authorize. Record the current monthly effort, the questions covered and any known factual errors.

At renewal, ask: Did we save measurable checking time? Did we find and fix a meaningful problem? Did the export remain comparable? Can we explain any business improvement without claiming causation the data cannot establish?

If the answer is only “the score went up,” keep investigating before renewing. A useful report should help someone do something better.

Get the next owner decision brief through The Memo.