Search Console’s AI Report Is Live: An AI SEO Reset

Two years of guessing about visibility inside AI answers ended quietly, inside a help document. Google confirmed on 31 August 2026 that its generative AI performance report, together with the control that lets a site withdraw from Search’s generative AI features, had finished rolling out to every website worldwide. The wording in Search Console Help is flat and final: as of 31 August 2026, Google says, the insights are live for all websites.

For a Hong Kong marketing manager who has spent the year fielding the same question from a director who reads headlines, there is finally a number to put in the deck. Whether it is the number anyone wanted is a separate matter. AI SEO, meaning search engine optimisation (SEO) aimed at the answer surfaces instead of the ten blue links, now has a native measurement baseline where it previously had third-party trackers and inference.

What Google Switched On

Two things arrived together. The first is the generative AI performance report in Search Console, which counts impressions for links to your site inside Google’s generative AI features. Google names those features precisely as AI Overviews and AI Mode. The data breaks down by page, country and date, with a device split available on the Search version of the report.

The second is the Search generative AI control, a property-level toggle sitting in Search Console settings. Google’s own description leaves little room for interpretation. Sites that opt out, the documentation states, “will not receive traffic or impressions from our generative AI features”. Google adds that the control “isn’t used as a ranking or inclusion signal affecting other parts of Search”, and that it has no bearing on model training, which is still governed separately through Google-Extended.

Both features had been visible to a slice of properties since June 2026, when Google first announced the reporting and controls for website owners. The end of August is simply the point at which the partial rollout stopped being partial.

Impressions, and Nothing Else

Here is the part that will disappoint anyone hoping for a traffic report. The generative AI performance report carries impressions. It does not carry clicks, click-through rate, average position or query data. An impression means a link to your site was shown to someone inside a generative AI feature, and Google aggregates by property, so two results from the same site appearing in one answer are counted once.

The report also folds AI Overviews and AI Mode into a single figure with no way to pull them apart, which limits diagnosis more than the headline suggests. A brand appearing constantly in AI Overviews for informational queries and never in AI Mode for commercial ones sees one line that looks perfectly healthy. That gap is familiar ground for answer engine optimisation (AEO), the practice of earning a place inside AI-generated answers. The clearest AEO wins for Hong Kong retailers have come from planning for impressions that hold steady while clicks fall away, and the new report confirms the first half of that picture while leaving the second half unmeasured.

The Toggle That Looks Tempting

Publishers frustrated by AI answers will be drawn straight to the opt-out control, and it deserves a slower read than it is likely to get. Switching it off removes a site from AI Overviews and AI Mode entirely, including the links inside them, as Search Engine Land reported on the day of the global rollout. For a business whose category is increasingly answered instead of listed, that is a large door to close on a hunch.

Six Things to Do With the Report This Month

The data is thin, and thin data used properly still beats rich data used late. A sensible first pass takes under an hour.

• Export the current figures before anything else changes, so the baseline is dated to the rollout and not to the day somebody finally noticed the report existed.
• Set generative AI impressions against total Search impressions for the same period, then track that ratio monthly instead of watching the raw number bounce around.
• Sort by page. The pages Google pulls into AI answers are rarely the pages the marketing team assumes, and the distance between the two is the actual brief.
• Split by country. A Hong Kong property serving local traffic, mainland visitors and Singapore buyers will show very different patterns under one domain.
• Open the settings page and confirm the generative AI control reads the way you expect, particularly on properties inherited from a previous agency or developer.
• Note which pages draw no impressions at all, then check whether they are thin, gated, or simply never structured to answer a question anybody asks.

None of that produces a strategy on its own. It produces the first honest picture of which parts of a site the machine is willing to quote, which is the only sound starting point for AI SEO planning in the months ahead.

Where the Report Cannot Follow You

Our team is based in Singapore and we work across Singapore, Hong Kong and mainland China, which makes one limitation of this report obvious in a way it may not be to a purely local reader. Search Console measures Google surfaces. For a Hong Kong brand whose growth plan runs north, that covers a shrinking share of the ground.

Baidu reported in its second quarter 2026 results that Baidu App’s monthly active users reached 644 million in June 2026. None of that audience appears anywhere in Search Console. A Hong Kong company can post a clean, improving generative AI impressions line while being absent from the answer surfaces its mainland customers actually use, and nothing in the Google report will hint at the gap.

The reverse trap catches Singapore businesses moving into Hong Kong. Google’s position in Hong Kong search is strong enough that the report is genuinely representative there, but the queries behind it arrive in Traditional Chinese and in English for the same commercial intent, with Cantonese phrasing shaping both. A country split will show Hong Kong performing. It will not show that a large share of the underlying demand was expressed in a language the content never addressed. Measuring one market with another market’s instrument is the quiet way regional campaigns go wrong.

Final Thoughts

A measurement tool is not a strategy, and the first month of data from any new report tends to be more interesting than it is useful. Give it a quarter. The ratio of generative impressions to total impressions, tracked patiently, will say more about your position in answer surfaces than any single month’s figure.

If you would like a second pair of eyes on what your report is showing, or on the parts of the region it cannot see, we are glad to talk it through. Our SEO and content marketing work across Singapore, Hong Kong and mainland China tends to begin exactly here, with somebody’s export on the table and a question about what it means.