A publisher can appear in an AI answer without receiving a click. It can be cited without being read, retrieved without being cited, and relied on without leaving any visible signal at all.
The click is no longer a reliable receipt: Pew Research Center found in 2025 that Google users click result links less often when an AI summary appears, and rarely click the sources inside it. That left rights holders measuring AI exposure with screenshots and referral counts. Since 31 August 2026, Google Search Console reports AI Overview and AI Mode impressions for every website. It is the first first-party visibility signal publishers can hold, and it makes a measurement framework possible — provided each metric is read for what it proves, and nothing more.
In short
Publisher visibility in AI answers should be measured on five separate layers: citation visibility, accuracy, catalogue coverage, post-answer demand and assisted outcomes. Each answers a different question and stops at a different limit.
Google's new generative AI reports supply the Google layer of the first one: impressions by page, country and date, but no clicks and no queries. Other assistants still have to be measured on a fixed panel of questions the publisher runs and archives itself.
For a rights holder, the scorecard has a specific job. It informs opt-out decisions, licensing priorities and correction requests. It does not prove that content was used, and it should never be presented as if it did.
What Google's new report shows, and what it doesn't
Google began rolling out generative AI performance reports in Search Console in June 2026 and extended them to all websites on 31 August. They cover AI Overviews, AI Mode and generative AI features in Discover, with page, country and date dimensions, plus device for the Search report.
The limits matter as much as the data. The reports count impressions only: no clicks, and no query dimension. Sites with few AI impressions may not see a report at all. An impression shows that a page appeared in an AI feature; it does not show what the answer did with it.
The same release added a property-level setting, the Search generative AI control, which removes a site's content from those three surfaces. Google says it does not affect AI training, which Google-Extended governs, and the UK CMA has set March 2027 as the deadline for page-level controls. For a rights holder, the order of operations follows: export the impression baseline before deciding on an opt-out, and keep a dated record of the control state alongside the AI Search records publishers must hold themselves.
A five-layer scorecard for rights holders
Layer | Question | How to measure it | What it informs | Limit |
|---|---|---|---|---|
1. Citation visibility | Where, and how persistently, does the publisher appear? | Search Console AI impressions; citation rate and persistence on a fixed query panel per assistant | Attribution performance; opt-out baseline | Says nothing about retrieval or reliance |
2. Accuracy | Is the publisher represented faithfully? | Expert review of a disclosed sample of cited answers | Integrity of the work; correction requests | Misses errors in answers that do not name the source |
3. Coverage | Which rights-bearing assets surface? | Weighted share of priority titles and topics that appear in panel answers | Licensing priorities; which assets carry AI demand | Valid only for a stable, documented panel |
4. Post-answer demand | Does exposure resurface as interest? | Branded, title and author queries in Search Console; direct visits to frequently cited pages | Value that survives the lost click | Correlation, not causation |
5. Assisted outcomes | Does visibility contribute to revenue? | Subscriptions, registrations and licensing enquiries on paths that include a tagged AI referral | The commercial return of AI exposure | Untagged visits fall into direct traffic |
Report the five layers separately. A composite score can sit on an executive dashboard, but only if its weights are published: otherwise a rise in low-value citations can hide a fall in accuracy on the titles that matter.
Citation visibility: measure persistence, not appearances
Outside Google, citation visibility has to be measured on a panel: a fixed list of questions, run on a schedule across the assistants that matter, with the date, model, country and exact prompt archived for every answer. Screenshots taken when a stakeholder asks are anecdotes.
Two measures matter more than a raw count. Citation rate is the share of panel answers that name or link the publisher. Persistence is the share of observation dates on which the same source stays present for the same prompt — because AI citations rotate even when the page has not changed.
Accuracy: the layer most publishers skip
A prominent citation is only an asset if the answer represents the source correctly. In an October 2025 study coordinated by the EBU and led by the BBC, 22 public service media organisations reviewed more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity. 45% contained at least one significant issue, and 31% had serious sourcing problems.
Review a disclosed sample against five tests: does the cited page support the claim, is the meaning preserved, is the current version cited rather than a corrected one, are publisher, author and title identified correctly, and is the excerpt consistent with the permitted use. Publish the sample size, period and reviewers with any accuracy rate. When an assistant attaches a distorted claim to a publisher's name, the citation becomes a liability, and the record of it supports a correction request or a contract clause.
Coverage: weight the catalogue by rights value
Generic share of voice is the wrong measure for a rights holder, because not every question carries the same value. Build the panel around the catalogue: questions where the publisher owns reference material, questions tied to paid products, and questions touching assets it has licensed, reserved or is negotiating.
Tier each question before collecting results, then report coverage as a weighted rate. Keep the core panel stable for at least a quarter; a panel that changes when results deteriorate produces an attractive chart and an unreliable measure.
Demand and assisted outcomes: label inference as inference
An AI answer can create awareness without a referral. Search Console's branded query analysis, title and author searches, and direct visits to frequently cited pages can capture some of it. News cycles, campaigns and launches move the same signals, so annotate comparison periods and state competing explanations.
For outcomes, start from what can be observed. ChatGPT Search appends utm_source=chatgpt.com to citation links, which lets analytics attribute some visits; links copied without the tag still land in direct traffic. Report the number and value of conversion paths that include a tagged AI referral, with the lookback window used, and call them assisted rather than caused.
Where the dashboard stops
All five layers measure the visible edge of AI use. None shows which content an assistant retrieved but did not cite — the gap explained in why retrieved is not cited — and none can be reconciled with a licensing term.
Usage reporting is moving, but it is not there yet. COUNTER's adopted guidance on AI usage metrics separates passages an AI could access from passages selected for the final answer, and states that AI reporting is not subject to its formal audit. Its phase-two work, described in a September 2026 update, maps five telemetry events — retrieval, grounding, citation, presentation and engagement — but was still awaiting approval at that date. A vocabulary is not a record, which is why publishers still need to test who audits the usage number.
A citation is user-facing attribution. A licensing position needs attribution as a countable, rights-linked use. Keep the two datasets side by side, and never let one stand in for the other.
The objection: "A GEO dashboard is a marketing tool"
For a brand, it often is. For a rights holder, the same data drives three decisions marketing teams never face: whether to opt out of an AI surface, which assets to prioritise in licensing, and when to escalate a misattribution. The objection holds only if the dashboard is built for share of voice rather than for those decisions.
A 90-day baseline
First, define the rights-bearing assets in scope and a tiered panel of questions, then fix the assistants, countries and cadence.
Second, export Search Console's AI impressions and branded queries, run the panel weekly, and review an accuracy sample monthly.
Third, report the five layers separately, and send rights or reporting gaps to the licensing team rather than treating them as SEO defects. Where attribution enters a contract, it can be priced as a rate modifier on each licensed use.
Visibility metrics tell a publisher how it appears. AI usage evidence tells it what was used — and that is the layer Citations Logic is built for.
Book an AI visibility and evidence review
Read next
Why access agreements and visibility alone cannot carry a licensing position is set out in proof of AI content usage.
What happens to catalogues read by machines rather than people is covered in machine readership and the reading crisis.
Frequently asked questions
How can a publisher measure visibility in AI answers without clicks?
Use five separate layers: citation visibility, accuracy, catalogue coverage, post-answer demand and assisted outcomes. Search Console's generative AI reports provide Google impressions; other assistants require a fixed, archived panel of questions.
What does the Search Console generative AI report include?
Impressions from AI Overviews, AI Mode and generative AI features in Discover, by page, country and date, with device for the Search report. It does not include clicks or queries.
Can visibility metrics prove that content was used or licensed?
No. They show observable appearances and downstream demand. Proof of use requires a record connecting the content object, the right and the usage event.
Sources
Google Search Central — Introducing Search Generative AI performance reports in Search Console, June 2026 (updated 31 August 2026)
Search Generative AI performance reports
Search Engine Journal — Google Search Console AI Reports Rolled Out Worldwide
Search Console AI reports rolled out worldwide
European Broadcasting Union — AI's systemic distortion of news is consistent across languages and territories, 22 October 2025
EBU and BBC study on news integrity in AI assistants
COUNTER Metrics — Best Practice on Generative and Agentic AI Usage Metrics, updated 30 June 2026
Best Practice on Generative and Agentic AI Usage Metrics
COUNTER Metrics — AI update, 18 September 2026
COUNTER AI update, September 2026
Clickport — Why ChatGPT Traffic Shows as Direct in GA4, 2026
Why ChatGPT traffic shows as direct
Pew Research Center — Google users are less likely to click on links when an AI summary appears in the results, 22 July 2025
Pew Research Center on AI summaries and clicks
This article is operational analysis for publishing and rights teams, not legal advice.