Search Console generative AI report: CMA metrics due 3 Dec 2026, baseline now

Summary. The UK Competition and Markets Authority (CMA) publisher conduct requirement, imposed on Google on 3 June 2026, comes into force on 3 December 2026. From then Google must give publishers clear and detailed engagement metrics for content used in AI Overviews and AI Mode. The CMA's interpretative notes expect impressions (including where a link sits in the answer), click-throughs with a way to identify them, and click-through rate, kept separate from organic results and delivered through a platform such as Search Console. Today the Search Console generative AI performance report, open to all websites since 31 August 2026, shows impressions only, by page, country, device and date. AI feature clicks are counted, but inside the Web totals of the Performance report, where they cannot be separated. Directory-level controls are due on 3 December 2026, page-level controls on 3 March 2027, and Google's first compliance report on 3 January 2027. The duties cover AI answers shown to people in the UK. Export the report monthly from November, pair it with Performance report clicks and GA4 engagement, and keep 60 to 90 days of server logs, so December's metrics have a baseline.

About this article. For SEO leads, publishers and analytics teams with UK search traffic, including Indian sites with UK readers. We checked dates and obligations against the CMA's notices, interpretative notes, summaries and press release, and product claims against Google's Search Console, Search Central, crawling and Analytics documentation, on 1 November 2026. This is general information, not legal advice, and no vendor paid for inclusion; Google changes these reports often, so re-check a field before relying on it. Researched with AI assistance and reviewed by our editorial team.

What the CMA ordered, and the dates that matter

The CMA designated Google as having strategic market status (SMS) in general search services on 10 October 2025, under the Digital Markets, Competition and Consumers Act 2024. On 3 June 2026 it imposed the publisher conduct requirement (the publisher CR), its first conduct requirement on Google, which stays in force until 10 October 2030 unless the CMA varies or revokes it.

Google must offer controls to keep content out of the training and grounding of its AI services (paragraph 3), with grounding controls for AI Overviews and AI Mode at directory level (4.a) and page level (4.b). It must explain how content is used (paragraph 6), give clear and detailed engagement metrics (paragraph 7), credit sources with clear links (paragraph 8), and not penalise publishers who use the controls (paragraph 9).

Paragraph 2 applies paragraphs 3 to 8 to AI outputs shown to people or businesses located in the UK, and Google can rely on the location the device reports. A publisher is anyone who makes content available on the web, so on our reading an Indian or US site with UK readers is covered for that UK audience.

The dates below come from the CMA notice, reporting notice and case page, and Google's announcement.

The CMA's August 2026 summary is plain: Google has until 3 December 2026 to comply, plus three more months for the page-level control. The press release expected important parts of the controls well before the nine-month deadline, and Google began testing them the same day. Nothing changed on the CMA side in the last week: the publisher CR page was last updated on 4 August 2026 and the case page on 23 September 2026.

What data publishers should expect from 3 December 2026

Paragraph 7 is one sentence; the detail is in the interpretative notes. Google may take a different approach if it can show the CMA that it complies, so read the table as the regulator's expectation, not a product specification. It compares the CMA notes and summary with Search Console on 1 November 2026.

Click data is the big change: today you can count how often AI answers show your links, not how often anyone follows them. The CMA summary also promises referral data so publishers can judge traffic quality, but Google has not said how it will mark those clicks, so do not build tagging around a guess. Google's launch posts said only that more metrics would come over time, and whether they arrive worldwide or only for UK traffic is Google's call.

What the generative AI report gives you today

Google's help page for the report is the authoritative description. For a screen-by-screen tour, see our guide to the Search Console AI performance report.

• Coverage. AI Overviews and AI Mode in Google Search. Discover's AI features have their own report, and Search Labs experiments are excluded.

• Metric. Impressions only: how many times links to your site were shown in a generative AI feature.

• Dimensions. Pages (the final URL after redirects, mostly the canonical), Countries, Dates (Pacific Time) and Devices. There is no Queries tab.

• Search type. Web text-based, or Web multimodal, which covers Lens, Circle to Search, image uploads and Chrome image search. The multimodal filter arrived on 24 September 2026.

• Aggregation and limits. The chart counts by property and the Pages table by page, so totals can differ. The Performance report's usual limits apply, including the 1,000-row table cap and its time-period limit, and values the report shows as a tilde or a dash export as zeros.

• Access. The help page documents an export button and says nothing about Search Analytics API or BigQuery bulk export access. Plan a manual monthly export.

Google's AI features documentation confirms that AI Overviews and AI Mode traffic sits in the Performance report under the Web search type, and that a page must be indexed and snippet-eligible to appear as a supporting link. The counting rules add three details: a click on an external link in either feature counts as a click, an AI Overview takes one position shared by all its links, and an AI Mode follow-up is a new query. Your AI clicks are already in your totals; you just cannot see which ones they are.

The Search generative AI control sits under Settings, with three values: include (the default), exclude, or inherit from the parent property (the default when a parent exists). It covers AI Overviews, AI Mode and generative AI features in Discover. A URL-prefix property such as https://example.com/blog/ can override its parent, a working form of directory-level control today. Google says exclusion generally takes a few days, the control is not a ranking or inclusion signal for the rest of Search, it does not override Merchant Center or Google Ads, and it does not affect AI training, which Google-Extended handles. Our explainer on the opt-out control covers the trade-offs.

Outputs shown in the UK versus elsewhere: who gets what

CMA rights attach to where the AI answer is shown, not where your server sits. Google has so far taken its UK-tested features global, but nothing obliges it to. Regional splits are not hypothetical: our piece on site reputation abuse rules in the EEA shows EU rules already shaping Search policy. The table draws on the CMA notice, the fair ranking summary and Google's control and crawler pages.

If you serve several markets, filter every export by Country and report the United Kingdom separately. That slice is where the new metrics are legally due first.

How to build your 2026 baseline in four layers

A baseline is only useful if you can pull the same numbers again after 3 December. Each layer answers one question.

Layer 1: export the generative AI report every month

1. On the first working day of each month, export the previous full month by Pages, Countries, Devices and Dates, with the export date in each file name.

1. Repeat the Pages export filtered to the United Kingdom, then to your other main markets (for an Indian business, usually India and the United States).

1. If your content is visual, repeat with the Web multimodal search type.

1. Tag every URL with its directory (/blog/, /guides/, /products/). The first new controls work per directory, so group the baseline that way.

Start with October and November 2026, plus any earlier months the report shows. Do not count on the history staying available: the Performance report's time-period limit applies here too.

Layer 2: pull Performance report clicks for the same pages

For the same months, URLs, countries and search type, export clicks, impressions, CTR and position from the Performance report. Then calculate per page and per directory:

AI impression share = AI feature impressions ÷ Web impressions.

This works because the generative AI report draws on the Web search type of the Performance report. Treat it as directional, and use the Pages table on both sides, since chart and table aggregate differently. Export top queries per directory too; the Performance report is your only clue to which questions trigger AI answers on your pages.

Layer 3: record GA4 engagement for AI-exposed pages

GA4 cannot isolate AI Overview or AI Mode visits. They arrive as google / organic like classic results, and Google documents no marker that separates them, which is why the CMA asks for a way to identify those clicks. So compare instead:

• Split landing pages into high and low AI impression share using Layer 2.

• For each group, record organic sessions, engagement rate, average engagement time and key events per session.

• Freeze key event definitions until the end of March 2027.

Separately, add a custom channel for AI assistant referrals (ChatGPT, Gemini, Copilot, Claude, Perplexity). GA4's documentation includes an AI assistants example, allows two custom channel groups on standard properties (five on 360) with up to 50 channels each, and applies them retroactively. Place the channel above Referral, because traffic lands in the first channel it matches. Google's sample regex is loose: branches such as .*gpt.* and .*gemini.* catch any source containing those letters. A tighter rule for the Source dimension:

            ^(.+\.)?(chatgpt\.com|openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai)$
          

Check the matched sources in Traffic acquisition before trusting it. Custom channel groups are not in the GA4 BigQuery export schema, so rebuild the rule there if you report from BigQuery.

Layer 4: keep raw server logs

Keep 60 to 90 days of raw access logs and record, per directory:

• Verified Googlebot crawl frequency. Google's crawler documentation lists user agents, IP ranges and the reverse DNS patterns for its common crawlers (hostnames under googlebot.com, now including geo.googlebot.com), and warns that user agents can be spoofed, so verify.

• What logs cannot show. Google-Extended has no separate HTTP user agent; it is a robots.txt control token, and crawling happens under the normal Google user agents. Logs also cannot tell AI Overview clicks from classic organic clicks.

• A GA4 cross-check. Search Console's help notes that tools such as Google Analytics only track users who have JavaScript enabled, so log counts give you a second number.

Logs hold IP addresses. India's DPDP Act 2023 treats any data about an identifiable individual as personal data, and UK data protection law takes a similar view, so set a retention period, document the purpose and limit access; our DPDP engineering playbook covers the controls.

Worked example: turning impressions into a baseline

These numbers are illustrative, not client data: a UK-facing B2B site with 120 indexed pages, using October 2026 exports filtered to the United Kingdom.

Group A's share is 40,000 ÷ 100,000 = 40% and its CTR 1,500 ÷ 100,000 = 1.5%. Group B's CTR is 3,750 ÷ 150,000 = 2.5%. Site-wide, 48,000 ÷ 300,000 = 16% of impressions came with an AI answer, and CTR is 6,000 ÷ 300,000 = 2.0%.

Do not conclude yet that AI answers cost group A a point of CTR; guides attract informational queries that click less anyway. The baseline is a fixed reference: 1,500 clicks, an unknown part of them from AI answers.

Now suppose December brings the same impression volumes and the new metrics show 240 AI clicks for group A. Its AI CTR is 240 ÷ 40,000 = 0.6%. The other 1,260 clicks came from about 60,000 non-AI impressions, so 1,260 ÷ 60,000 = 2.1%. The split is approximate, because one search can show a page in the AI answer and as a classic result, but that gap is the decision-grade number. If AI CTR stays far below classic CTR for two or three months while AI share rises, test a directory-level exclusion on part of group A and watch total clicks and GA4 key events. If AI clicks convert as well as classic organic, stay included.

Exclusion has a limit. Google's control page warns that content from other sites still appears in those features and may look similar to yours. You leave the answer; the answer stays.

Decision table: what to do now, by site type

The Merchant Center and training rows rest on Google's control page.

The fair ranking requirement, briefly

On 17 June 2026 the CMA imposed a fair ranking conduct requirement, and its summary gives Google until 17 December 2026 to comply. Organic results, including AI features, must be ranked on objective, non-discriminatory criteria, and Google must not consider whether a publisher advertises with it, has opted out of AI use, or has exercised legal or contractual rights against it. Material ranking changes need at least 30 business days' notice where publishers could act to limit the impact, and at least 15 business days' notice for changes that may cause significant ranking shifts and that Google can explain. Google must also offer ways to give feedback on changes and to request reconsideration of manual actions.

Log every ranking-change notice after 17 December beside your baseline, so a ranking shift is not mistaken for an AI feature effect. If a drop lines up with a core update instead, our core update recovery playbook covers the diagnosis.

New this week: a shared AI training signal

On 30 October 2026 Google said it plans to support a shared AI training control in robots.txt, based on the IETF AIPREF working group, and its crawler documentation now points to early 2027 and user-agent: __usage_ai-training. If a site opts out with that shared token and does not name Google-Extended, Google plans to treat it as a Google-Extended disallow; Google-Extended keeps working alongside it. This governs training only. Whether your pages appear in AI Overviews or AI Mode is still set by the Search generative AI control and snippet controls, so log any training-related robots.txt change separately from an AI feature exclusion.

Implementation checklist

1. By 30 November 2026: set up a Domain property plus a URL-prefix property for each directory you might treat differently.

1. By 30 November 2026: export October and November using the Layer 1 cuts and the Layer 2 pairing.

1. By 30 November 2026: create the GA4 AI assistants channel, place it above Referral and freeze key events.

1. By 30 November 2026: set 60 to 90 days of log retention and record each property's current control value, with the date.

1. 3 December 2026: add a Search Console annotation, check the report for new metrics and read Google's updated explanatory pages.

1. 17 December 2026: start logging fair ranking change notices.

1. January 2027: compare December with the baseline, and look for the public version of Google's first compliance report, which the CMA expects to require.

1. 3 March 2027: review the page-level control for pages that need different treatment from their directory.

1. Early 2027: when Google supports the AIPREF signal, review robots.txt training preferences alongside Google-Extended.

A note for India and other non-UK markets

Indian sites already have the exclusion control and the impressions report, but no legal guarantee of December's click metrics; we found no Indian rule equivalent to the CMA's publisher requirement as of 1 November 2026. Publishers with real UK readership, such as diaspora news sites or B2B exporters, should watch the UK country filter first. Everyone else should still build the baseline: if Google extends click data globally, as it did impressions, the comparison needs numbers from before the change.

How eCorpIT can help

eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our data and analytics services team can set up the monthly Search Console exports, GA4 channel groups, log pipelines and baseline dashboard this article describes before 3 December. Talk to us at /contact-us/.

Last updated: 1 November 2026.

References

1. Publisher Conduct Requirement Notice, 3 June 2026 — Competition and Markets Authority

1. Interpretative notes in relation to the Publisher Conduct Requirement, 3 June 2026 — Competition and Markets Authority

1. Generative AI performance report (Search) — Search Console Help

1. What are impressions, position, and clicks? — Search Console Help

1. Publisher CR compliance reporting notice, 3 June 2026 — Competition and Markets Authority

1. Google's general search and search advertising services (case page) — Competition and Markets Authority

1. New opportunities, control and insights for website owners, 3 June 2026 (updated 31 August 2026) — Google

1. The publisher conduct requirement (summary), August 2026 — Competition and Markets Authority

1. CMA secures fairer deal for publishers and improves Google search services in UK, 3 June 2026 — Competition and Markets Authority

1. Google search publisher conduct requirement — GOV.UK

1. Introducing Search Generative AI performance reports in Search Console, 3 June 2026 — Google Search Central Blog

1. Announcing web multimodal Search performance reporting in Search Console, 24 September 2026 — Google Search Central Blog

1. AI features and your website — Google Search Central

1. Search generative AI control — Search Console Help

1. Fair ranking conduct requirement (summary), August 2026 — Competition and Markets Authority

1. Google's common crawlers — Google Crawling Infrastructure

1. Custom channel groups — Google Analytics Help

1. About Search Console data — Search Console Help

1. The Digital Personal Data Protection Act, 2023 (Gazette of India, 11 August 2023) — Ministry of Electronics and Information Technology

1. Google search fair ranking conduct requirement — GOV.UK

1. Supporting a shared AI training control for websites, 30 October 2026 — Google Search Central Blog

1. Crawling documentation changelog, 30 October 2026 entry — Google Crawling Infrastructure