Search Console and AI Search: What You Can Measure

What Google announced

Google presented reports on June 3, 2026 devoted to the generative functions of Search and Discover. A note added to this announcement indicates a global rollout by August 31. Advertised dimensions include impressions, pages, countries, dates, and, for Search, devices. The official Google publication specifies this perimeter.

This data is already included in the overall performance: the dedicated report provides a specific reading, not an additional total to add up. This tracking concerns the Google experiences covered by the report. It does not automatically measure your mentions in all market assistants. For this broader question, check out our guide on business visibility in ChatGPT.

Start with the right property and period

Open the property corresponding to the site you want to analyze. Check the domain, URL variations, and your access rights before concluding that a report is missing. The interface and available data may evolve; rely on the titles actually displayed and the documentation linked from your account. Save the date of your first observation so that you can compare subsequent readings.

Choose a full period, then a comparison period of the same length. A two-day report can be dominated by an event, a reporting delay, or a single page. For a small business, monthly tracking is a practical starting point. Avoid simultaneously changing the period, the country and the pages studied: you will no longer know which factor explains the observed difference.

Read a print without giving it too much meaning

An impression is an exposure measure according to the reporting rules. It is not proof of careful reading, preference for your brand or commercial request. A page can be displayed in a context where the Internet user obtains his answer without clicking. Conversely, low volume may relate to an issue very close to your offering and deserve your attention.

Keep three separate columns: visibility measured by Google, behavior observed on the site and commercial result. If the report does not provide a metric, do not reconstruct it by dividing totals that do not cover the same scope. A click-through rate calculated from inconsistent data gives misleading accuracy. Document limits directly in your tracking table.

Identify pages that meet a real need

Classify the pages exposed by intention: understand a problem, compare solutions, prepare a project or contact the company. An advice page and a service page do not play the same role. See if each offers a coherent sequence. A reader preparing a redesign needs a framework and examples before filling out a form.

Then reread the important pages. Does the main answer appear quickly? Is the information dated where it matters? Are the examples verifiable? Are the links to the services relevant? The guide to a site invisible in Google remains useful if your difficulty primarily concerns exploration or indexing. A new report does not replace these technical bases.

Build a usable monthly table

Your table can contain the page, the need addressed, the period, the country observed, the impressions of the report and the actions performed on the content. Add visits and requests separately when your tools can measure them correctly. A comment column helps note events: campaign, title change, publication of a customer case or technical problem.

Fictitious example: a page on choosing a reservation system goes from 120 to 180 impressions over two comparable periods. This represents an increase of 50%, but does not demonstrate that the change in the title is the cause. If requests remain stable, examine the rest of the journey and the low volume before calling it success or failure. The numbers are used to guide an audit.

Improve one page at a time

Choose a page that receives visibility and corresponds to your activity. Correct a specific weakness: incomplete answer, poorly explained price, missing example or action difficult to find. Note the date and modification. Wait until you have a usable period before making more changes. This discipline does not allow perfect attribution, but it makes observations more interpretable.

Avoid recipes that promise automatic citation in AI responses. Markup, repetition of words, or a hidden block of text does not guarantee a selection. Work on real usefulness: original information, accurate presentation of the company and correct technical access. The article on Google and AI responses in France explains how to make your pages more understandable.

Linking tracking to contact processing

An increase in visibility loses its point if the form does not work or if requests remain unanswered. Test the appeals, links and mobile path of the chosen pages. Check that messages are reaching the right person. Add a simple business discovery question to the sales follow-up, without pretending that customer recall constitutes accurate attribution.

For a low traffic site, take several months together before drawing a firm conclusion. A single request can significantly change a percentage. Describe gross volumes in addition to variations: “two more requests” is often more meaningful than “100% growth”. If visits increase without useful contacts, examine blocking of a site that does not generate quotes.

Reporting errors to avoid

Do not combine impressions from the dedicated report with the overall total as if they were two separate audiences. Don't present Google data as a measure of the entire AI market. Also, don’t turn an exposed page into proof that your business is consistently recommended. Each indicator must keep its name and scope in the report.

Prepare a short conclusion for each review: what the data shows, what it does not allow us to affirm and the action chosen. This way of working helps the manager decide without getting lost in screenshots. The first month's good result may simply be identifying three pages to improve and one measurement issue to fix.

Frequently asked questions

Does the report measure mentions in ChatGPT?

No. It covers the Google features indicated in its documentation. Other motors and assistants require separate observations.

Can we add these impressions to the overall report?

No, Google says this data is already included in overall performance. Adding them together would create double counting.

Does an increase in impressions mean more customers?

Not by her alone. Compare visits, contacts and useful records separately, with consistent volumes and periods.

What to do if no data appears?

Check ownership, access, filters and current documentation. The absence of visible data is not enough to prove a penalty or site failure.