How to measure changes in your AI visibility as an expert
Keep a useful baseline, compare the same questions and interpret changes in your appearances without overstating what they prove.
On this page
If you change your website, publish an article, or clarify your services, it is tempting to run a new AI search immediately and treat the result as a verdict. A single answer is too narrow for that. It may differ because the question changed, the locale changed, the engine changed, or the answer simply varied.
Treat each scan as a dated snapshot of what specified engines returned for saved questions. Compare only snapshots with compatible scope.
This records appearances in answers, not your ability or professional quality.
Build a baseline you can repeat
Start with questions a prospective client might ask without knowing your name. Save the exact wording, along with the locale, language, date, and engines checked. Name-recognition questions have a purpose, but keep them out of a discovery baseline because they supply the identity you want to observe.
For a fictional example, Mateo Ruiz is an independent operations consultant who helps subscription businesses improve customer onboarding. Mateo saves these questions for US English checks:
- Which independent consultants help subscription businesses improve customer onboarding?
- Who can review a SaaS onboarding process that has a high drop-off after signup?
- Which experts advise software teams on onboarding handoffs between sales and customer success?
These are illustrative questions, not evidence of real prompt demand. Mateo records the date and the four engines in scope: ChatGPT, Gemini, Claude, and Perplexity. New Gistful scans use that four-engine scope. A historical scan with fewer engines should remain labelled with its original scope rather than being treated as equivalent.
The question selection guide explains how to choose a set you can stand behind. The buyer-question worksheet gives you a fill-in record for the exact wording and scope. Save it before you see the result.
Separate an observed absence from a missing answer
Each returned answer can be reviewed for a confirmed appearance. If Mateo does not appear in a completed answer, he can record “not observed in this answer.” If an engine fails, blocks the request, or produces no usable answer, he should record “unmeasured.”
Those labels keep your denominator honest. A failed check is not proof that an engine omitted you.
Here is a fictional baseline record for Mateo. “Appearance” means that a completed answer contained a verified reference to the same Mateo Ruiz. It does not mean the engine endorsed his work.
| Engine | Completed answers | Verified appearances | Unmeasured checks | Notes |
|---|---|---|---|---|
| ChatGPT | 3 | 1 | 0 | One answer linked to a relevant profile |
| Gemini | 2 | 0 | 1 | One request did not return a usable answer |
| Claude | 3 | 1 | 0 | One appearance was a brief candidate mention |
| Perplexity | 3 | 0 | 0 | No verified appearance in these answers |
Mateo can calculate 2 verified appearances across 11 completed answers, or 18.2% for this snapshot. He should not calculate 2 out of 12 by treating the unmeasured Gemini request as a negative. Retain the underlying answers, since percentages alone hide identity mistakes and changes in wording.
Gistful’s results and Gist Score documentation describes its own evidence and scoring rules. Use the product’s definitions when reading a product score, rather than substituting a home-made percentage.
Keep the scope compatible when you rescan
Before a later comparison, check the question text, locale, language, engine set, profile scope and collection methodology. A new question about local businesses is not comparable to an old question about US software companies. Neither is a new four-engine scan comparable to an older scan that checked only one engine.
Google Search illustrates why context belongs in your notes: its results can vary with time, location, language, device, recent searches, and personalization. Why search results differ That makes scope part of the evidence.
If your services have changed, retain the earlier scan and its question text before editing your saved questions. Label the new scan as a new baseline. This preserves what you actually measured at each point.
| Comparison question | What to do |
|---|---|
| Same questions, locale, language, profile, methodology and engines? | Compare completed-answer evidence carefully |
| One engine was unavailable in either scan? | Compare the compatible evidence only, and retain the gap |
| Questions or locale changed? | Start a new baseline |
| You updated your site between scans? | Note the date and change, without claiming it caused the result |
Read the answers as well as the totals
Suppose Mateo’s later scan has four verified appearances across 12 completed answers. The raw rate is 33.3%, but the earlier scan had only 11 completed answers. First inspect the 11 matching question-and-engine pairs. Record the newly available answer separately. Otherwise, the headline percentage mixes changed answers with changed availability. It is worth opening each answer to check whether the name is truly his, whether the wording is a recommendation or a neutral mention, and whether displayed citations point to relevant material.
Do not treat a higher number as proof that your edit caused the change. Public information, retrieval, query interpretation, and the returned answers can all change between snapshots. Do not treat a lower number as evidence that your expertise declined. The measurement records a narrow set of answer outcomes.
The citation versus recommendation guide shows why those labels should remain separate. If another expert appears where you do not, read the evidence without guessing at hidden reasons.
A repeatable rescan checklist
- Use the saved discovery questions without editing the wording.
- Preserve the same locale, language, profile, methodology and engine scope.
- Save completed answers and visible sources, not just a score.
- Verify that each apparent name is actually you.
- Mark failed or missing answers as unmeasured.
- Record public changes you made between scans as context.
- Start a new baseline when scope changes.
An explicit rescan gives you a new dated record of how selected AI answers surfaced your expertise. Keep the scope and evidence alongside the result, and the comparison stays useful even when the answer changes.