# Measuring AI visibility in 2026: a practical guide
Measuring AI visibility: Google impressions, observed citations, sessions and conversions. The methods and limitations of Search Console and GA4.
Author: Louis Choquet · Founder, Pharelia
Published on 2026-07-19 · Updated on 2026-10-04
Source: https://pharelia.com/en/resources/mesurer-visibilite-ia
## The essentials in 60 seconds
Measuring AI visibility means tracking where and how your brand is cited in ChatGPT, Perplexity and Google's AI answers, then analyzing measurable traffic and conversions. Search Console covers AI Overviews and AI Mode impressions in its AI report; GA4 tracks attributable sessions and conversions. Observed answers and logs supplement this data without proving a citation or sale on their own.
Since Google confirmed the arrival of AI Overviews in France, the same question has come up every week on the accounts we manage: "how do we measure whether AI cites us?" This guide answers it with an operational method, the metrics that matter, counting pitfalls and the tools we actually use. For context on the concepts, see our article on [Google AI Overviews in France](/en/resources/apercu-ia-google-france).
## Why measure AI visibility now?
Google announced the [launch of AI Overviews and AI Mode in France on July 22, 2026](https://blog.google/intl/fr-fr/nouveautes-produits/explorez-obtenez-des-reponses/recherche-ia-apercus-mode/). AI Overviews appears when Google considers a generated answer useful for the query; it does not replace all conventional results. The aim now is to measure changes in your queries and pages, using a stable scope and a documented method.
The agreement includes three commitments: control over AI features (opt-in/opt-out), transparency through separate reporting of AI impressions, and remuneration under neighboring rights for around 450 publishers. Separate reporting does not yet exist in public tools: until then, you need to build the measurement yourself.
Keep a dated record: observed citations by engine, attributable sessions and validated conversions. If you have data from before the launch, compare it using an identical scope; otherwise, start monitoring today without claiming to reconstruct a before-and-after comparison. A change is not enough to attribute an effect to AI Overviews. This is what we define during a [conversation about your project](/en/contact).
- **July 22, 2026** : Launch of AI Overviews and AI Mode in France (https://blog.google/intl/fr-fr/nouveautes-produits/explorez-obtenez-des-reponses/recherche-ia-apercus-mode/)
- **~450** : Publishers covered by neighboring rights remuneration (Abondance (2026))
- **~69 %** : Estimated zero-click queries (Ahrefs (2026))
## What exactly should you measure?
Three groups of signals: presence (are you cited?), citation share (in what proportion of answers compared with competitors?) and downstream impact (AI referral traffic and conversions). Presence without impact provides false reassurance; impact without presence cannot be managed. We measure both ends of the chain.
One finding changes the method: citations barely overlap between engines. According to [Higoodie](https://higoodie.com/blog/ai-search-traffic-report-2026/), ChatGPT citations overlap with only around 12% of the Google SERP, and only ~11% of domains are cited by both ChatGPT and Perplexity. Measuring a single engine gives a misleading picture.
- Presence: track the queries in which your brand is cited, engine by engine.
- Citation share: compare your citation frequency with that of competitors.
- Impact: isolate AI traffic in GA4 and connect it to conversions.
- Never generalize from a single engine: citations overlap little.
## Which tools can track AI citations?
Search Console provides a [generative AI performance report](https://support.google.com/webmasters/answer/16984139?hl=en) for AI Overviews and AI Mode impressions. Google reports a global rollout as of August 31, 2026, with access limitations depending on the property and impression volume. This data is also included in the Web report: do not add it to that report's total. For ChatGPT and Perplexity, supplement this with a stable query panel, recorded answers and attributable sessions in GA4. Crawler logs indicate access, not citations.
An AI visibility measurement stack — combining sources is essential.
| Signal to measure | Data source | Limit to know |
| --- | --- | --- |
| Citations in answers | Scraping tracked queries (ChatGPT, Perplexity) | Nondeterministic answers requiring sampling |
| AI crawler visits | Server logs | Says nothing about actual citations |
| AI referral traffic | GA4: source/medium data and the AI Assistant channel | A lost source cannot be reconstructed from Direct |
| AI Overviews and AI Mode impressions | Search Console: generative AI performance report | Check availability; impressions already included in the Web report |
Compare identical periods, countries, devices and aggregations. The chart is aggregated by property, or by URL when a URL filter is applied; the page table may therefore have a different total. [Google's counting rules](https://support.google.com/webmasters/answer/7042828?hl=en) do not establish automatic double-counting of AI and conventional impressions. A CTR decline calls for an analysis of traffic and queries, not a blanket adjustment. Our [SEO audit checklist](/en/resources/audit-seo-checklist) supplements the technical checks.
In GA4, check the [AI Assistant channel and source/medium data](https://support.google.com/analytics/answer/9756891?hl=en) for identifiable sessions. Traffic with no identifiable source may appear as [Direct](https://support.google.com/analytics/answer/15258820?hl=en) — a custom channel group does not reconstruct a lost source. Do not therefore reattribute all Direct traffic to AI.
## Which metrics should you prioritize?
Two metrics dominate: AI citation share (your citation frequency across target queries) and the conversion rate of AI traffic. The second is surprising: according to Seer Interactive, AI Search traffic converts at around 14.2%, compared with ~2.8% for conventional organic traffic. A smaller volume can therefore have a substantial impact.
For traffic, start by measuring the distribution of sources. ChatGPT accounts for most AI referral traffic (~87%, Conductor 2026), while Perplexity ranks third (~7.7%, Statcounter May 2026). Prioritizing measurement efforts on ChatGPT and Perplexity covers the vast majority of the usable signal today.
### Conversion rate: AI vs organic traffic
AI traffic converts substantially better despite a lower volume.
| Period | Conversion rate (%) |
| --- | --- |
| AI Search traffic | 14.2 |
| Conventional organic traffic | 2.8 |
Source: Seer Interactive (2026)
Watch for clicks lost earlier in the journey. Ahrefs measures a CTR decline of 15 to 30% on informational queries where AI Overviews is deployed — an effect documented in [its study](https://ahrefs.com/blog/ai-overviews-reduce-clicks/). This is why it matters to track conversion, not just session volume.
## Why AI citations are so hard to track
Because engines do not draw on the same sources and change quickly. According to Ahrefs (15,000 queries), ~80% of LLM citations do not rank in Google's top 100. AI visibility is therefore not an automatic consequence of Google rankings: it needs to be measured separately.
The target also moves over time. The share of AI Overviews citations coming from Google's top 10 fell from around 76% in mid-2025 to ~38% in early 2026 (Higoodie). Your good ranking yesterday therefore no longer guarantees an AI citation. A one-off check is not enough: measurement needs to recur.
### Share of AI Overviews citations coming from Google's top 10
The relationship between Google rankings and AI citations is weakening rapidly.
| Period | Share from the top 10 (%) |
| --- | --- |
| Mid-2025 | 76 |
| Early 2026 | 38 |
Source: Higoodie (2026)
An observed tactical point: Reddit is cited disproportionately by LLMs because its content can be retrieved in full. For our clients, a structured presence on fetchable platforms is becoming a GEO distribution lever in its own right, to be measured like other surfaces.
The implication for the method: we do not measure AI visibility once a quarter; we sample it continuously. A query panel tracked every week captures these shifts, whereas a one-off audit misses them. Our services include this recurring monitoring rather than a fixed snapshot that ages within a few weeks.
## Do you need an llms.txt file to be measured and cited?
Not for Google. Google's official AI guide (May 15, 2026) is explicit: Google Search does not use llms.txt — either to help or to penalize. The file may serve some other crawlers, but no evidence establishes an effect on Google. Implementing it "for measurement" has no impact on Google.
The same [Google guide](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) debunks other myths: there is no requirement to split content into chunks, no need to write specifically for AI, and structured data is not required for generative AI (it remains useful for conventional rich results). The official position, reported by [Search Engine Journal](https://www.searchenginejournal.com/googles-new-ai-search-guide-calls-aeo-and-geo-still-seo/575026/): optimizing for generative AI means optimizing for search — in other words, SEO.
The implication for measurement: there is no point tracking the effect of llms.txt on Google; there is none. Focus the dashboard on actual citations and traffic. To clarify the scope of each discipline, see our note on [AEO vs SEO vs GEO](/en/resources/aeo-vs-seo-vs-geo).
## Does content quality still matter for citations?
Yes, more than ever. Google does not penalize AI as a production method: Danny Sullivan summarizes the position — "we judge quality, not the production method". But low-value mass content receives the lowest rating from evaluators. What you measure as visibility rests on a foundation of quality.
The Quality Rater Guidelines (September 2025 version, ~182 pages, ~16,000 evaluators) define three spam categories: Expired Domain Abuse, Site Reputation Abuse and Scaled Content Abuse, plus a section on low-effort, low-originality content. This is the framework applied by [Google's quality raters](https://developers.google.com/search/docs/fundamentals/creating-helpful-content).
The algorithmic context reinforces this: the May 2026 core update finished on June 2, 2026 (~12 days), and the [June 2026 spam update](https://searchengineland.com/google-releases-june-2026-spam-update-481002) launched on June 24, 2026, globally and in all languages. Measuring AI visibility without monitoring these updates means confusing an algorithm change with a content effect.
### Let's discuss your visibility
Let's talk about your website and priorities.
[Let's discuss your project](https://pharelia.com/en/contact)
## How to build an AI visibility dashboard
In four steps, without a miracle tool. Start with a panel of target queries, sample AI answers at fixed intervals, correct attribution in GA4, then connect citations and conversions. Consistency matters more than exhaustive coverage: 30 queries tracked every week are better than a huge audit once a quarter.
1. Define 20 to 50 truly strategic target queries, not the long tail.
2. Sample ChatGPT and Perplexity answers every week (answers vary; measure a frequency, not a fixed state).
3. Check GA4: the AI Assistant channel, source/medium data and conversions. Use custom channels if needed, without reattributing Direct sessions whose source is unknown.
4. Connect citations and conversions: this is the only metric that justifies the investment to management.
To maintain the production cadence that supports these citations, we use an editorial framework documented in our resource on the [content engine and cadence](/en/resources/moteur-de-contenu-cadence). Measurement and production form a single loop: publish, measure citations, adjust, publish again. A dashboard that leads to no content decision is useless: every weekly measurement should result in a concrete editorial action, otherwise you are measuring for its own sake.
## Sources
- [Official AI optimization guide — Google Search Central](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide)
- [AEO and GEO still SEO — Search Engine Journal](https://www.searchenginejournal.com/googles-new-ai-search-guide-calls-aeo-and-geo-still-seo/575026/)
- [AI Overviews arriving in France, summer 2026 — Abondance](https://www.abondance.com/20260630-2528492-google-ai-overviews-arrivee-france-ete.html)
- [AI Overviews reduce clicks — Ahrefs](https://ahrefs.com/blog/ai-overviews-reduce-clicks/)
- [AI Search Traffic Report 2026 — Higoodie](https://higoodie.com/blog/ai-search-traffic-report-2026/)
- [June 2026 spam update — Search Engine Land](https://searchengineland.com/google-releases-june-2026-spam-update-481002)
## Frequently asked questions
### What does measuring AI visibility mean?
It means tracking where your brand is cited in answers from ChatGPT, Perplexity and Google's AI Overviews, then connecting those citations to traffic and conversions. Since no tool covers everything, we combine answer scraping, server logs, GA4 and Search Console to obtain a reliable picture.
### When did AI Overviews become available in France?
Google announced the launch of AI Overviews and AI Mode in France on July 22, 2026. AI Overviews does not appear for every search: Google shows it when it provides a useful answer to the query.
### Does an llms.txt file improve my visibility on Google?
No. Google's official AI guide (May 2026) states that Google Search does not use llms.txt, either as a benefit or a penalty. The file may serve other crawlers, but there is no evidence of an effect on Google. There is therefore no need to track it in a Google visibility dashboard.
### Can you rely on Search Console for AI visibility?
Yes, within its scope. The generative AI performance report tracks AI Overviews and AI Mode impressions; check its availability and data volume on your property. Its impressions are already included in the Web report. Compare identical aggregations and analyze clicks and CTR in the Web report, then sessions and conversions in GA4. The AI report does not measure ChatGPT or Perplexity citations.
### Why is a good Google ranking not enough to be cited by AI?
Because AI engines do not draw their sources from Google's top results. According to Ahrefs, around 80% of LLM citations do not rank in the top 100. And the share of AI Overviews citations coming from Google's top 10 fell from around 76% in mid-2025 to 38% in early 2026 (Higoodie). AI visibility is measured separately.
### Is AI traffic worth measuring despite its low volume?
Yes, because it converts better. According to Seer Interactive, AI Search traffic converts at around 14.2%, compared with around 2.8% for conventional organic traffic. A smaller volume can therefore generate more conversions. This is why we always connect citations and conversions, not just the number of sessions.