SEO
Measure AI visibility : Search Console, citations and sources
Definition
Measuring AI visibility means tracking signals rather than one fixed ranking.
AI visibility describes how a website, brand or page appears in answers generated or assisted by artificial intelligence : a citation with a link, a displayed source, a brand mention, a summarized page, a reused idea, an impression in a Google AI feature or an indirect presence inside an answer.
Unlike classic SEO, this visibility cannot be reduced to a stable position. AI answers vary by tool, query wording, language, location, context, model, interface and sometimes personalization.
The point is not to promise an absolute AI ranking. It is to build a serious method : observe, compare, document, interpret and improve the content that can become reliable sources in new search environments.
AI visibility is not measured like a stable rank. It is measured by a website’s ability to be understood, reused, cited or selected as a source.
Approach
Observe what is cited, understood, reused or ignored.
At Edikka, AI visibility measurement is not treated as a promise of ranking inside assistants. It is handled as a management discipline : tracking citations, sources, mentions, available impressions, cited competitors and answer quality.
This approach separates what is truly measurable, what is only observable, what remains partial and what must be interpreted carefully. The goal is not to attribute everything. It is to understand whether the website is progressively becoming a clear, reliable and usable source.
Citations
02Sources
03Mentions
04Impressions
Positioning
The question is not how to be cited by AI, but how to track that visibility.
AI visibility measurement should not be confused with optimizing a website to become more citable. Making content clear, reliable, structured and citable is an editorial and technical project. Here, the subject is more precise : how to observe and manage signs of presence in AI environments.
This method complements the definition pillar What is GEO ? : the pillar explains how content can be selected or cited, while this page shows how to observe those signals over time.
This article does not repeat the analysis of AI Overviews and AI Mode either. It focuses on the tracking method : test queries, citations, sources, competitors, mentions, available impressions, interpretation and monthly improvement.
Optimize
Improve content, proof, structure, indexing and clarity.
Observe
Test queries, record citations, compare sources and track competitors.
Measure
Cross-check Google impressions, citations, mentions, visible pages and conversions.
Manage
Turn observations into editorial, SEO or technical actions.
Issue
Why AI visibility is not measured like classic SEO.
Classic SEO relies on relatively familiar indicators : average position, impressions, clicks, CTR, indexed pages, queries and conversions. These signals remain essential, but they are no longer enough to understand a website’s visibility inside AI-generated or AI-assisted answers.
In an AI answer, content can be cited with a link, used as a source, mentioned without a link, reformulated without visible attribution, or absent despite strong classic organic visibility. AI search does not replace traditional results with a new simple ranking. It synthesizes, selects, reformulates and contextualizes.
SEO position
Shows a place in classic results, but does not guarantee a citation inside an AI answer.
AI citation
Shows that a page is displayed as a source, but does not guarantee a visit or a conversion.
Mention
Can strengthen brand awareness, but remains difficult to attribute when it is not linked.
AI impression
Can indicate presence in Google Search, without proving traffic or a decision on its own.
Measurable signals
What can actually be measured today.
There is not yet a single metric capable of measuring an entire website’s AI visibility. The right method combines several signals : some are directly measurable, others are only observable, and some remain partial.
| Tracked signal | What it indicates | Reliability | How to observe it | Main limitation |
|---|---|---|---|---|
| Citation with link | A page or domain appears as a visible source. | High | Regular tests in ChatGPT, Perplexity or Google. | Results vary by query and tool. |
| Displayed source | The website is presented in a source block or panel. | High for a point-in-time observation | Manual or semi-automated source logging. | Does not cover the full AI ecosystem. |
| Unlinked brand mention | The brand is named in an answer. | Medium | Observe answers across a query corpus. | Difficult attribution and no direct click. |
| Page reused or summarized | Information from the website seems used or reformulated. | Medium to low | Compare the AI answer with the source content. | The origin can remain uncertain. |
| Cited competitor | Another domain appears as a recurring source. | High if the source is displayed | Track cited domains by query. | Does not automatically explain why the competitor is chosen. |
| Answer quality | The answer correctly understands the subject or expertise. | Medium | Qualitative scoring grid. | Partly human assessment. |
| Source fidelity | The answer respects the content that is actually published. | Medium | Compare answer, cited page and context. | Assistants may simplify or omit nuance. |
| Trigger query | The wording that triggers a citation or mention. | Medium | Stable corpus of test queries. | A wording variation can change the answer. |
| Google AI impression | A URL appears in some Google Search AI features. | High if the report is available | Dedicated Search Console reports, depending on availability. | Partial, progressive and limited to Google. |
| Organic conversions | Search traffic produces business value. | High if tracking is clean | Analytics, CRM, forms, calls or quotes. | Indirect attribution when AI exposure does not generate a click. |
No single indicator is enough. Measurement becomes useful when signals are cross-checked, followed over time and connected to concrete editorial decisions.
ChatGPT
Track citations and sources without a dedicated analytics dashboard.
ChatGPT can display sources when web search is used. These citations are useful for observing whether a page or domain appears as a source in certain answers.
However, website owners do not have a complete analytics dashboard that automatically tracks every ChatGPT citation for their domain. The method must therefore remain regular, documented and comparative.
Observation method
Build a stable corpus of test queries.
- Create a list of 20 to 50 strategic queries.
- Separate queries by intent : informational, commercial, comparative, local or expertise.
- Test the same queries at regular intervals.
- Record whether the brand is mentioned.
- Record whether the domain is cited as a source.
- Identify the cited page when a link is displayed.
- Log cited competitors.
- Assess whether the answer correctly understands the expertise.
- Note whether an idea from the website seems reused without a link.
- Compare results month after month.
One test is never enough. The exact query wording, test timing, web search access and interface used can change the results.
Perplexity
The clearest terrain for tracking AI sources.
Perplexity puts sources at the center of the experience. Today it is one of the clearest environments for observing how an AI answer uses web pages, domains and citations.
This does not mean Perplexity represents the whole AI ecosystem. Its results have their own logic. But for a tracking process, it is a useful lab for sourced visibility.
Record displayed sources for each strategic query.
Count how often the domain appears across a stable corpus.
Identify which URLs are cited and whether they match priority pages.
Spot competing domains that regularly return as sources.
Perplexity can also be tracked semi-automatically through its tools and APIs. But the goal should not be to turn the analysis into a heavy technical project. The goal is a clear reading : which sources appear, for which intents, against which competitors and with what evolution.
Since June 3, 2026, Google has made part of AI visibility observable at last.
The new fact is not only the existence of AI Overviews or AI Mode. The strategic fact is Google’s June 3, 2026 announcement of dedicated reports for generative feature performance in Search Console. For the first time, Google isolates part of the impressions tied to its AI experiences instead of leaving them blended only into classic Search performance.
This opening does not turn Search Console into a complete GEO dashboard. It provides a proprietary signal on Google Search, useful for tracking exposure, but insufficient for measuring all of a website’s AI visibility in ChatGPT, Perplexity or other assistants.
At launch, the report mainly answers one question : which URLs appear in certain Google AI experiences, at what impression volume, in which countries, on which devices and on which dates.
It does not yet answer the questions that matter directly for business performance : which click should be attributed to an AI answer, which CTR should be compared, which average position should be read, which precise query should be analyzed and which conversion should be attached to that exposure.
June 3, 2026
AccessProgressive rollout, first in the United Kingdom, then more broadly
Google AI impressions, the pages involved, countries, devices and appearance dates.
No dedicated reading of clicks, CTR, average position or queries at launch.
AI Overviews, AI Mode and generative Google features, not ChatGPT or Perplexity.
Google provides a first native signal, but the method still has to cross-check several sources.
Use Search Console as a Google AI thermometer.
Impressions indicate that a URL was shown in a generative Google experience. They help identify visible pages, but do not yet support a conclusion about traffic or value.
Do not confuse AI impression, citation and click.
A Google impression, a displayed source in Perplexity, a ChatGPT citation and an organic visit do not tell the same story. Merging them creates a flattering but fragile metric.
Connect Google impressions with manual observations.
The right reading compares Google AI impressions with citations observed in ChatGPT, Perplexity, cited competitors, visible pillar pages and organic conversions.
France
For French websites, the strategic window opens before the data is fully available.
The initial rollout in the United Kingdom is not incidental. It sits in a context of regulatory pressure from the Competition and Markets Authority (CMA), which is asking Google for more control, transparency, metrics and attribution around content used in AI features.
For a French company, this creates a preparation window. There is no need to wait until reports are available everywhere to start tracking strategic queries, pages that should become citable, competitors already being reused and Google signals that will gradually arrive in France and elsewhere.
The real advantage is not having the report before everyone else. It is having a reference framework ready when Google AI impressions become readable in France.
France should therefore be read as a market to prepare : define queries, identify decisive content, document competing sources and strengthen the pages that can be reused in AI answers.
When Google data becomes more widely accessible, companies that are already structured will be able to compare trends. Others will only discover that they have no reliable starting point.
Waiting for a perfect report
PrepareA corpus, pillar pages and a competitive benchmark
VPN
The possible role of a VPN : international monitoring, not official proof.
A VPN can help observe certain markets where AI experiences are more visible. It can be useful for understanding the types of queries that trigger AI answers, cited sources, reused content formats or visible competitors in other countries.
But a VPN test does not necessarily represent what a French user will see. It does not perfectly simulate language, location, Google account, personalization, history, device, datacenter or ongoing tests.
A VPN can help anticipate a trend. It cannot establish a definitive truth about a website’s AI visibility in France.
A VPN should remain a punctual, manual and documented monitoring tool. It should not be used to scrape Google or automate queries at scale.
Edikka Method
Build an AI visibility dashboard in 8 steps.
Useful measurement must be stable, comparable and actionable. The Edikka method builds a query corpus, observes the same environments every month, then turns signals into improvement actions.
Corpus
Define 20 to 50 strategic queries.
The corpus must reflect the topics that actually matter to the company : expertise areas, services, customer questions, comparisons, local queries, decision content and trust-building intents.
Intents
Classify queries by intent.
Queries should be grouped by intent : informational, comparative, commercial, local, expertise or decision. This helps identify where the website is visible and where it remains absent.
Tests
Test ChatGPT, Perplexity and Google.
The same queries should be observed across several environments. The goal is not to obtain one single truth, but to compare sources, answers, competitors and visible pages.
Logging
Record citations, sources, mentions, competitors and visible pages.
Each test should be documented with the exact query, tool, date, cited sources, visible pages, cited competitors and answer quality.
Cross-check
Cross-check with Search Console when data is available.
Google AI impressions should be cross-checked with classic impressions, clicks, organic pages, long-tail queries, pillar pages and conversions.
Diagnosis
Identify pages that are absent, misunderstood or challenged by competitors.
A competitor cited several times often reveals an editorial angle to analyze. An absent page may lack clarity, proof, structure, internal links or authority.
Improvement
Improve content based on observed signals.
Actions may involve clarity, headings, definitions, examples, proof, internal linking, pillar pages, structured data and indexing.
Routine
Repeat the tracking every month.
AI visibility must be tracked over time. A monthly evolution is more useful than an isolated test because it reveals trends, recurring competitors and content to reinforce.
Dashboard
Columns to track in an AI visibility dashboard.
A simple dashboard is enough to start. The key is to keep the same queries, the same tools, the same criteria and the same tracking frequency.
| Query | Intent | Tool tested | Site cited | Page cited | Brand mentioned | Cited competitors | Presence type | Answer quality | Action to take | Monthly evolution |
|---|---|---|---|---|---|---|---|---|---|---|
| Example strategic query | Comparative | Perplexity | Yes / No | Cited URL | Yes / No | Competing domains | Citation, mention, source or absence | Good, partial or weak | Enrich, clarify, connect or create | Stable, rising or falling |
Interpretation
How to interpret results without drawing the wrong conclusion.
AI visibility must be interpreted carefully. A one-off absence does not mean failure. A citation does not guarantee traffic. A Google AI impression does not guarantee a conversion. An unlinked mention can have value, but remains difficult to attribute.
What matters is the trend : do the same pages return more often ? Are cited competitors changing ? Do answers understand the expertise better ? Are strategic contents appearing more often in the tested environments ?
Compare, contextualize, improve.
An isolated test is worth less than a trend observed over several months.
Each result depends on the tool, query, country, language and moment.
Answer quality matters as much as simple presence in a source.
Each observation should lead to an editorial, SEO or technical action.
Improvement
How to improve your chances of being cited or reused by AI systems.
The goal is not to look for a special or guaranteed optimization. The same foundations remain decisive : useful content, clear structure, indexing, semantic HTML, proof, internal links, coherent visible data and genuinely maintained pages.
Create pillar pages, clear headings, explicit subheadings, definitions and readable sections.
Add examples, methods, use cases, verifiable information and proof of expertise.
Strengthen internal links between pillar pages, satellite articles, FAQs and service pages.
Update strategic content, check indexing and align structured data.
Vague, overly marketing-oriented or poorly verifiable copy is harder to use. The most citable content is often content that answers a question clearly, explains its limits and provides precise context.
Limits
The limits of AI measurement.
AI measurement remains imperfect. Answers vary between tools, interfaces, models, countries, languages, accounts, devices and test dates. No tool currently provides a universal and complete view of a website’s AI visibility.
Conclusions should therefore not be rushed. A one-off citation can be encouraging, but it is not enough to prove dominance. A one-off absence can point to a problem, but it does not prove total invisibility.
- Answers can vary depending on the exact query wording.
- Tools do not always cite the same sources.
- Location, language, account and history can influence the experience.
- Google reports remain progressive and do not cover the whole AI ecosystem.
- A citation, a mention, an impression and a click do not measure the same thing.
- Attributing a conversion to AI exposure without a click remains difficult.
Reliable AI measurement accepts uncertainty : it is meant to read trends, not to produce absolute proof from an isolated screenshot.
Conclusion
AI visibility is built, observed and managed over time.
AI visibility is not won with a trick. It is built with reliable, structured, understandable, verifiable content that is tracked over time.
Google’s June 3, 2026 announcement changes the level of visibility : Google AI impressions are becoming progressively observable, but they do not replace ChatGPT citations, Perplexity sources or analysis of visible competitors.
The real question is therefore no longer only whether the website ranks first on Google. It becomes whether the website is clear, useful and reliable enough to be understood, cited or reused in the new search environments.
Measuring AI visibility means building a regular, comparable and actionable observation process : tracking citations, sources, mentions, Google impressions and cited competitors in order to progressively improve content.
AI visibility is not measured like a ranking. It is managed like influence.
In AI engines, a brand can be cited, summarized, reused, ignored or replaced by a competitor. The issue is no longer only to track a position, but to understand whether the website is becoming a source clear, reliable and useful enough to be selected.
At Edikka, we treat AI measurement as strategic observation: tracking citations, sources, mentions, reused pages, visible competitors and answer quality. What matters is not absolute certainty, but an actionable trend that helps strengthen content.
Track what is cited, not only what is ranked
A citation, a displayed source, a brand mention or a summarized page are different signals. Tracking them separately shows where the website gains authority and where it remains absent.
Observe the competitors AI systems prefer
The sources cited instead of you are often more instructive than your absence. They reveal content judged clearer, better structured, more credible or easier to use.
Turn observation into editorial improvement
Measuring AI visibility has value only if every record leads to an action: clarify a page, strengthen proof, add an FAQ, improve internal linking or structure an answer.
AI visibility is not proven with an isolated screenshot. It is managed through records over time: observed sources, cited competitors, reused pages and concrete editorial decisions.
Go further on this topic
Additional answers to clarify the key points covered in this article.
