Skip to content

David Shapiro / Reviewed November 14, 2026

Measure what the answer changes.

An answer can mention your brand, link to your site, or describe your product incorrectly. Those observations call for different actions.

Use the platform reports that now exist

Google announced dedicated Generative AI performance reports in June 2026. It states that the worldwide rollout finished on August 31, 2026.

The Search report shows impressions within supported generative AI features. Available dimensions include pages, countries, devices, and dates.

A missing report can reflect insufficient impressions. Check the site’s inclusion control before concluding that the report is unavailable.

Bing’s AI Performance view reports citation activity across supported AI experiences. Its documentation distinguishes citations from ranking, authority, and placement in an answer.

Give each measure a clear job

Platform impressions help describe exposure. Sampled answers can show how a system represents the brand. Referral visits describe observable website traffic.

Qualified enquiries and accepted opportunities help describe commercial response. They require an agreed definition and suitable sales data.

A citation is not a customer. A customer can also encounter a brand without following a citation. Keep these observations separate.

  • Visibility: Did the brand or a page appear?
  • Representation: Was the important information correct?
  • Action: Did a relevant person visit or make contact?
  • Commercial quality: Did the enquiry become a useful opportunity?

Repeat the questions

AI answers can change across sessions, questions, locations, and platforms. One screenshot gives you one observation.

SparkToro and Gumshoe research examined repeated brand recommendations. It found substantial variation and supported using repeated observations when assessing visibility.

A useful panel starts with real buying questions. Record the prompt, platform, date, location setting, answer, and cited URLs.

Use several phrasings where they represent the same decision. Keep missing responses and collection failures separate from valid answers without a mention.

Make the sample visible

Our starting engagement plans twelve questions across two selected platforms. Three repeats and two measurement periods produce 144 planned observations.

This is a defined observation panel. It is not a representative sample of every customer search.

Report how many valid answers were collected. Show the count behind each percentage and describe platform changes or missing observations.

A rise from one mention to two can look large as a percentage. The small number still gives limited evidence.

Choose a test that matches the claim

A content change followed by more citations does not establish that the change caused the increase. Other pages, competitors, and models may also change.

Use comparable unchanged pages or questions when practical. Define the comparison, outcome, and decision rule before publishing.

For a conversion test, measure qualified actions per eligible visit. For an accuracy problem, score the selected claims as correct, wrong, or missing.

If a test lacks enough observations, report the uncertainty. A useful business review can recommend further measurement instead of declaring a result.

Turn the report into a decision

End with the next action the evidence supports. A report should help decide what to continue, change, investigate, or stop.

An inaccurate product description may justify an information review. Strong exposure with weak enquiries may justify a buying-path review.

Neither finding automatically supports more content. Match the action to the problem that was observed.

Try this with your team.

  1. Choose one commercially important question.
  2. Record the answer and source links on two relevant platforms.
  3. Repeat the check under the same conditions.
  4. Separate what appeared, what was correct, and what customers actually did.
Sources and evidence

Reviewed November 14, 2026. Platform guidance can change. Original hypotheses remain separate from the sources below.

Work with David Shapiro

See where your brand stands.

Five buyer questions. Three AI products. Your brand and two competitors. A free manual review with three priorities.

For established brands considering engagements from $5,000/month.

Have a broader project? Tell me about it →
Get your free snapshot