Guide : AI Search Measurement and Reporting

TL;DR — AI search measurement is figuring out how beaucoup AI outils — Google AI Overviews, ChatGPT, Perplexity, Gemini, Copilot — are sending vous trafic and mentioning votre brand. The catch: la plupart of it doesn’t montrer up cleanly in votre analytics. A lot of AI visits land in votre “Direct” bucket with aucun étiquette, and Google’s Generative AI report now separates AI-feature visibility, but it fait pas expose clicks, CTR, or requêtes. So au lieu de un perfect number, vous stack a few partial views.

Evidence for this claim Google's Generative AI Performance report provides impressions for participating properties but does not expose clicks, CTR, or query dimensions. Scope: The June 2026 Search Console Generative AI report rollout; property availability can vary. Confidence: high · Verified: 2026-07-28 Google Search Central: Introducing Search Generative AI performance reports Search Console Help: Generative AI performance report Evidence for this claim Analytics attribution depends on available campaign and referrer information, so some sessions can be classified as direct when no usable source is available. Scope: Google Analytics attribution behavior; does not establish a universal percentage for AI referrals. Confidence: high · Verified: 2026-07-28 Google Analytics Help: Traffic-source dimensions

Pourquoi ce is hard

In old-school SEO, the chain was simple: someone searches, sees votre lien, clicks, and votre analytics records “organic search.” With AI search que chain garde breaking in three places:

1. The click disappears. AI Overviews souvent réponse the question on lune page, so the utilisateur jamais clicks via. Vous got exposure, but aucun visit to count.
2. The étiquette disappears. Quand someone fait click from an AI outil, nombreux of ceux outils don’t tell votre site où the visit came from. With aucun “referrer,” the visit obtient dumped into Direct — the même bucket as bookmarks and typed-in URLs.
3. The tooling isn’t là yet. Google seulement ajouté an “AI Assistant” channel to Analytics in May 2026, and même que misses a lot. Search Console montre AI Overview impressions now, but pas clicks.

Ce que vous pouvez en réalité mesurer

Don’t expect un dashboard. Think of it as a few différent views, chaque showing partie of the picture:

• Trafic — visits from AI outils (in GA4’s AI Assistant channel, or Ahrefs Web Analytics).
• Visibility — how souvent vous apparaître in AI réponses (impressions in Search Console; “citations” in Bing Webmaster Outils).
• Mentions — si AI outils nom votre brand quand personnes demander relevant questions.
• Conversions — si que AI trafic en réalité turns into customers (souvent it converts really bien).

The un chose to remember

The number in votre analytics is a floor, pas a ceiling. AI trafic pourrait regarder comme a tiny 0,25% of votre visits — but a big chunk of the réel number is hiding in Direct, and AI visitors tend to convert at beaucoup plus élevé rates. Pour Ahrefs, 0,5% of visitors drove 12,1% of signups in Ahrefs’ June 2025 internal study. So don’t dismiss AI search simplement parce que the visible slice semble petit.

Vouloir the complet framework — the five measurement layers, the exact outils, and the correlation données — switch to the Avancé tab.

TL;DR — AI search attribution remains incomplete: some visits arrive sans usable referrer information and GSC inclut AI-feature activity dans Web search reporting plutôt que exposing a separate AI filter. Parce que aucun unique outil sees it tout, vous stack layers — direct attribution, LLM visibility, share of voice, hallucination monitoring, and self-report (plus incrementality quand vous have the volume). Treat ce que vous pouvez mesurer as a floor. And remember the punchline: the signals que drive LLM visibility peut overlap with broader brand signals, so interpret AI-search metrics alongside brand health plutôt que as an isolated channel.

Evidence for this claim Google's Generative AI Performance report provides impressions for participating properties but does not expose clicks, CTR, or query dimensions. Scope: The June 2026 Search Console Generative AI report rollout; property availability can vary. Confidence: high · Verified: 2026-07-28 Google Search Central: Introducing Search Generative AI performance reports Search Console Help: Generative AI performance report Evidence for this claim Analytics attribution depends on available campaign and referrer information, so some sessions can be classified as direct when no usable source is available. Scope: Google Analytics attribution behavior; does not establish a universal percentage for AI referrals. Confidence: high · Verified: 2026-07-28 Google Analytics Help: Traffic-source dimensions

Pourquoi AI search is harder to mesurer que traditional search

Three choses break at une fois.

Attribution échec. La plupart AI-referred trafic arrives with aucun referrer signal, so it lands in Direct plutôt que as an AI channel. Estimates of how beaucoup AI trafic is invisible ce façon range from à propos de 35% to 70,6% (the élevé fin is vendor-sourced from Loamly’s 446 405-visit sample — utiliser the range as directional, pas gospel). Ce isn’t a Google conspiracy; it’s how referrers fonctionner. As I put it quand we testé ce at Ahrefs: “Websites have contrôler over ce que info ils send. Ils peut send the complet chemin, simplement the origin, or nothing — it’s up to les. We report whatever referrer we’re told to report. Si ils don’t send us un, alors it voudrait go in the ‘Direct’ bucket.”

Visibility sans clicks. AI Overviews generate impressions and brand exposure sans sending trafic. The old impressions → clicks → conversions funnel doesn’t hold quand the réponse is the destination.

Aucun standard tooling. Jusqu’à May 2026 là was aucun native GA4 channel pour AI trafic. GSC’s June 2026 Generative AI report separates impressions pour participating properties, but pas clicks, CTR, or requêtes. Bing ajouté AI performances metrics in February 2026 as a public preview. And the result of tout ce friction: seulement 16% of brands systematically track AI search performances (McKinsey, September 2025). La plupart of votre competitors are measuring nothing — qui signifie quelconque réel measurement system is an edge.

The floor-not-ceiling principle

Si vous seulement remember un framing, faire it ce un. Publié AI-traffic shares hover autour 0,25% of average site trafic (Ahrefs’ study of ~82 000 sites) — but that’s simplement the measurable slice. Two corrections inflate it:

1. Dark trafic. With 35–70% of AI visits referrer-less, the réel number is plausibly 2–3x ce que votre analytics montrer.
2. The conversion premium. AI trafic converts dramatically meilleur. Pour Ahrefs, 0,5% of visitors drove 12,1% of signups — a 23x premium, and ceux visitors browsed ~50% plus pages per session with a lower bounce rate. Industry-wide the premium is plus comme 4–4,4x (Semrush/Adobe), but the direction is consistent.

So the correct number to ouvrir a stakeholder report with is pas 0,25%. It’s the growth rate (9,7x in 12 months) or the conversion premium. Ceux reframe the stakes; the traffic-share number undersells les.

The five layers of measurement

Aucun unique outil sees the whole picture, so vous stack partial views. Ce is Paul DeMott’s 5-couche GEO framework (Moteur de recherche Land, May 2026), with my propre données folded into chaque couche. Chaque couche ci-dessous is aussi its propre deep-dive in ce cluster.

Couche 1 — Direct attribution (retrieved vs. mentioned vs. cited)

Ce is GA4 plus Ahrefs Web Analytics: who en réalité visited, from qui AI source.

• GA4’s AI Assistant channel (ajouté May 13, 2026) catches referred sessions from ChatGPT, Gemini, Claude, Copilot, Grok, and similaire. Utile — but it seulement sees sessions que arrive with a referrer. The 35–70% que don’t encore sit in Direct, aucun matter how vous configurer channels. It aussi excludes Google AI Overviews and AI Mode, qui apparaître as plain Organic Search, didn’t appliquer retroactively, and uses un of votre two custom channel-group slots.
• Ahrefs Web Analytics has the AI channel construit in plutôt que requiring custom setup, and updates closer to réel temps.
• The vocabulary matters ici: retrieved ≠ mentioned ≠ cited. A retrieval is votre content being récupéré; a mention is votre brand named in the réponse; a citation is votre URL lié as a source. Construire a report on the incorrect un and the whole chose misleads. Voir Retrieved vs. Mentioned vs. Cited in AI.

Couche 2 — LLM visibility (the impressions-and-citations couche)

How souvent vous apparaître in AI réponses, click or aucun click.

• GSC Gen AI Performances Reports (June 2026) montrer impressions seulement — aucun clicks, aucun CTR, aucun position — pour AI Overviews, AI Mode, and Découvrir AI fonctionnalités. Réel exposure signal, but vous pouvez’t turn it into trafic. (Plus ci-dessous and in the GSC cluster article.)
• Bing Webmaster Outils’ AI Performances report (Feb 2026 preview) is plus transparent: total citations, average cited pages, grounding requêtes (the phrases Copilot searched internally to trouver vous), and page-level citation activity.
• Reality vérifier: seulement 38% of pages cited in Google AI Overviews ranked in the traditional top 10 (bas from 76%) — LLM visibility n’est pas the même as votre blue- lien rankings.

Couche 3 — AI hallucination monitoring

A wrinkle que has aucun equivalent in traditional search: AI outils send trafic to pages que don’t exist and décrire votre brand inaccurately. In my Ahrefs AI trafic page-type analysis, I trouvé 3,6% of Ahrefs’ AI assistant trafic went to non-existent (hallucinated) URLs. Au-delà bad liens, vous vouloir to watch si models décrire votre product, pricing, and positioning correctement — a incorrect “fact” repeated à travers models is a measurable reputation problem. Voir AI Hallucination Monitoring.

Couche 4 — Self-report (the bridge analytics can’t construire)

Ajouter a “How did you first hear about us?” field with AI options to votre sales, contact, and post-conversion formulaires. Ce is the seulement couche que captures AI’s top-and-middle-of-funnel influence — the discovery que happened weeks avant a referrer-less Direct visit. DeMott reports ce surfaces double-digit AI attribution in some pipeline studies. It’s low-tech and it fonctionne precisely où the tracking fails. Voir AI Trafic Attribution (qui folds the dark-traffic problem and the self-report fix ensemble).

Couche 5 — Share of voice (and pourquoi it’s a trap on its propre)

Share of Voice is the percentage of relevant AI réponses in qui votre brand is mentioned or cited. Outils automate it at scale: Ahrefs Brand Radar (400M+ search-backed prompts à travers ChatGPT, Perplexity, Gemini, Copilot, AI Overviews, AI Mode, and Grok as of July 2026, up from 350M+ précédent ce année — it’s a growing, live-updating index, so treat the exact figure as directional; monthly refresh), Semrush AI Toolkit (100M+ prompts), Profound, BrightEdge, Scrunch. Vous pouvez aussi do it manually: run a fixed prompt définir à travers 3+ models monthly and tally mentions.

But heed DeMott’s warning: “Share of Voice is a vanity metric sans business connection.” SOV indique vous how souvent vous montrer up, pas si showing up drives awareness, trafic, or pipeline. Toujours pair it with Couche 1 (trafic) and Couche 4 (self-report). Voir AI Share of Voice (SoV).

Un plus honest couche au-delà ces five: incrementality testing. Difference-in-differences — a high-AI-visibility tester cohort vs. a contrôler — over 6–12 months is the seulement façon to prove AI search caused revenue plutôt que merely correlating with it. It’s the slowest and the la plupart rigorous. La plupart teams won’t obtenir ici pour a année; collect the baseline données now so vous pouvez.

Platform-specific reporting (the gotchas)

Recherche Google Console. The June 2026 Generative AI report provides a separate visibility view pour participating properties, but it is impressions-only: aucun clicks, CTR, or requête dimension. Ne faites pas infer visits from ceux impressions or splice les into the standard Web report as though the metrics were equivalent. The utile signal: rising AI Overview impressions alongside flat or falling clicks in the principal report is “The Great Decoupling” — measurable AI Overview cannibalization (Ahrefs saw the blog’s clicks/impressions correlation flip from +0,425 to −0,352, with AIOs tied to a 34,5% CTR reduction).

Bing Webmaster Outils. Ahead of Google on transparency. Its AI Performances report exposes citations separately from organic clicks, and the grounding requêtes are a genuinely unique window — ils tell vous ce que Copilot was en réalité trying to réponse quand it pulled votre page. Comparer votre most-cited Bing pages to votre top organic pages; gaps are opportunity.

GA4 & Ahrefs Web Analytics. Covered in Couche 1 — les deux carry the même dark-traffic limitation; Ahrefs is built-in and faster, GA4 is configurable and excludes AIO/AI Mode.

Ce que correlates with LLM visibility (Ahrefs’ 75 000-brand study)

In Ahrefs’ 75 000-brand study by Louise Linehan and Xibeijia Guan, qui I discussed in my Evolve 2025 talk, the signals ranked comme ce by correlation with AI Overview visibility:

• Branded web mentions — 0,664 (the strongest signal)
• Branded anchor text — 0,527
• Branded search volume — 0,392
• Domain Rating — 0,326 (the weakest of the four)

Lire que liste à nouveau. Three of the four are brand signals, pas classic technical SEO ones. The implication is the through-line of ce whole hub: measuring AI visibility is mostly measuring brand health, and the fonctionner que improves AI visibility (mentions, citations, authority) is the même fonctionner que improves traditional SEO.

Où to go suivant

Ce hub is the map. Chaque couche is its propre deep dive in the cluster:

• Retrieved vs. Mentioned vs. Cited in AI — the three visibility types vous doit pas conflate, and qui ones vous pouvez en réalité mesurer.
• LLM Visibility / AI Visibility — impressions, citations, and how to audit votre presence in AI réponses with outils comme Brand Radar.
• AI Hallucination Monitoring — tracking incorrect facts, bad pricing, and the hallucinated-URL trafic wrinkle.
• AI Trafic Attribution — the dark-traffic problem in complet, platform-by-platform referrer behavior, GA4 setup, and the self-report bridge.
• AI Share of Voice (SoV) — defining it, measuring it, the outils, and how to garder it from being a vanity metric.
• GA4 pour AI Trafic — the réel GA4 configuration: custom channel groupes, referrer regex, and Explorations pour segmenting AI-platform trafic.
• AI Robot d’exploration Log Analysis — reading raw server logs to mesurer AI-bot explorer activity, vérifier user-agents contre réel IPs, and spot crawl-vs-render problems.

AI summary

A condensed prendre on the Avancé version:

• The problem: AI search is growing ~9,7x année over année and converts up to 23x meilleur que organic, but it’s nearly invisible in analytics. 35–70% of AI visits arrive with aucun referrer and land in Direct; GSC’s separate Generative AI report has impressions but aucun clicks or requêtes. Seulement 16% of brands systematically track AI search (McKinsey, Sept 2025).
• Floor, pas ceiling: the ~0,25% AI-traffic share understates reality — dark trafic probable 2–3x’s it, and AI visitors convert disproportionately (Ahrefs: 0,5% of visitors → 12,1% of signups). Lead reports with growth/conversion, pas the trafic share.
• Stack five layers (DeMott’s framework): (1) Direct attribution (GA4 AI Assistant channel + Ahrefs Web Analytics; remarque GA4 misses AIO/AI Mode); (2) LLM visibility (GSC Gen AI impressions-only; Bing citations + grounding requêtes); (3) Hallucination monitoring (3,6% of Ahrefs AI trafic hit non-existent pages); (4) Self-report (formulaire/survey attribution — the seulement top-funnel signal); (5) Share of voice (Brand Radar et al. — but “vanity sans business connection” unless tied to pipeline). Ajouter incrementality quand vous have volume.
• Don’t conflate retrieved ≠ mentioned ≠ cited.
• Utiliser the complet funnel: accessible → indexé/eligible → retrieved → mentioned → cited → clicked → converted → revenue. Report the premier failed stage au lieu de promoting an upstream proxy into a downstream outcome.
• The Great Decoupling: rising AIO impressions + flat/falling clicks in GSC = cannibalization.
• The punchline: AI-visibility correlates la plupart with brand signals — web mentions (0,664), anchor text (0,527), branded search (0,392), DR (0,326). So measuring AI search is mostly measuring brand health.

Documentation officielle

Primary-source documentation from the moteur de recherches on AI measurement.

Google

• AI fonctionnalités and votre website — how AI Overviews/AI Mode trafic is (and isn’t) reported; “counted towards the overall data in Search Console.”
• Gen AI Performances Reports announcement (June 3, 2026) — the nouveau impressions-only reports pour AI Overviews, AI Mode, and Découvrir AI fonctionnalités.

GA4 (Google Analytics)

• What’s nouveau — AI Assistant channel (May 13, 2026) — the nouveau Par défaut Channel Groupe entry pour chatbot referrals.
• Par défaut Channel Groupes — how GA4 buckets trafic, notamment Direct.
• Custom Channel Groupes — building an AI-source regex pour pre-2026 coverage.

Bing / Microsoft

• Introducing AI Performances in Bing Webmaster Outils (Public Preview) (Feb 2026) — total citations, average cited pages, grounding requêtes, and page-level citation activity.

Quotes from the source

On-the-record statements from the platforms and from me (Patrick) on the attribution gap. Chaque lien jumps to the source.

Google / GA4 — the nouveau (partial) tooling

• “You can now identify how users are discovering your site through chatbots like ChatGPT, Gemini, and Claude via a new AI Assistant channel in your Default Channel Group reports.” — GA4 What’s Nouveau, May 13, 2026. Jump to quote
• Nouveau GSC Gen AI Performances Reports give “dedicated views of your impressions within generative AI features on Search, such as AI Overviews and AI Mode, as well as generative AI features in Discover” — impressions seulement; click données is pas inclus, and Google dit it’s rolling the reports out to a subset of sites premier. — Recherche Google Central, June 3, 2026. Jump to quote

Microsoft Bing — the valeur of clicks

• “As AI advances, the value of clicks is rising and expected to continue rising.” — Microsoft Bing. Jump to context

Patrick Stox (Ahrefs) — the attribution gap

• “Attribution is important, and we don’t have it yet. It makes me suspicious that there’s something Google doesn’t want us to know about this traffic.” — May 21, 2025. Jump to quote
• “There’s a cost to bots crawling your websites and there’s a social contract between search engines and website owners, where search engines add value by sending referral traffic to websites.” — May 26, 2025. Jump to quote
• “The first LLM to add more value to users by showing impressions and clicks to website owners will likely have a big advantage.” — May 26, 2025. Jump to quote
• “Websites have control over what info they send. They can send the full path, just the origin, or nothing — it’s up to them. We report whatever referrer we’re told to report. If they don’t send us one, then it would go in the ‘Direct’ bucket.” — May 26, 2025. Jump to quote

Ryan Law (Ahrefs, Director of Content) — on CTR novelty

• “These CTRs are probably the highest they’ll ever be as the novelty of this format wears off.” — June 2025. Jump to quote

Paul DeMott (Helium SEO) — on vanity metrics

• “Share of Voice is a vanity metric without business connection.” — Paul DeMott’s 5-Couche GEO Measurement Framework, Moteur de recherche Land, May 2026.

The five measurement layers — cheat sheet

The hub En un coup d’œil: chaque couche, ce que it measures, the outil(s) pour it, and ce que the données semble comme.

Three terms vous doit pas conflate

Platform reporting gotchas

• GSC: the Generative AI report separates impressions pour participating properties but exposes aucun clicks, CTR, or requête dimension.
• The Great Decoupling signal: AIO impressions up + clicks flat/bas = AI cannibalization.
• Bing: plus transparent que Google — citations are broken out, and grounding requêtes montrer ce que Copilot searched to trouver vous.
• GA4 AI Assistant channel: referrer-only, excludes AIO/AI Mode, pas retroactive, uses a custom channel-group slot.

Free-tools baseline (aucun paid platform nécessaire): GSC Gen AI reports (impressions) + GA4 AI Assistant channel (referred trafic) + Bing WMT AI Performances (citations) + a monthly manual prompt définir à travers ChatGPT/Perplexity/Gemini pour SOV.

Construire an AI-search measurement stack

• Define the business questions avant selecting metrics.
• Document qui AI referrals GA4 peut recognize and qui journeys remain unknown.
• Garder AI-feature trafic blended into organic reporting où the source platform ne fait pas expose a separate dimension.
• Freeze a representative prompt pool pour visibility tracking.
• Separate retrieval evidence, brand mentions, and lié citations.
• Track conversion and assisted outcomes, pas seulement visits or appearances.
• Ajouter first-party “how did you hear about us?” evidence où it fits the journey.
• Record model, platform, outil, prompt, date, market, and methodology changements.
• Reconcile dashboards to underlying events, logs, and captured réponses.
• Étiquette modeled or directional numbers differently from observed counts.

Examiner the report avant sharing it

• Fait every chart state its source and observation window?
• Are blind spots décrit au lieu de silently estimated?
• Are comparisons like-for-like à travers platforms and temps?
• Peut a stakeholder distinguish trafic, visibility, mention, and citation metrics?
• Fait the report connecter changements to a decision or suivant investigation?

Testez vos connaissances: AI search measurement

Ressources utiles

My writing on AI measurement

• AI Assistants Are Breaking Web Analytics and Hurting Leur Future — the foundational attribution-gap piece; platform-by-platform referrer testing.
• Google Made It So Vous pouvez’t Track Clicks From AI Mode (Partially Fixed) — historical context on the AI Mode referrer problem and its partial fix; the plus tard GSC Generative AI report adds impressions, pas click or requête reporting.
• Fait AI Search Trafic Convert Meilleur Que Traditional Search? Pour Ahrefs, Yes — the 23x conversion benchmark (0,5% of visitors = 12,1% of signups).
• 80% of Our AI Search Trafic Goes to Our Homepage, Product Pages, and Free Outils — page-type breakdown, plus the 3,6% hallucinated-page wrinkle.

My talks & slides

• GEO? AEO? LLMO? What’s With Tout Ce AI Stuff? — Ahrefs Evolve 2025 (video) — the 9,7x growth study and the AI-visibility correlation données.
• Evolve 2025 slides (SpeakerDeck).
• AI SEO webinar (July 30, 2025). My standing disclaimer s’applique: ce is my understanding of systems, pas a guarantee it’s 100% complet or accurate.

Supporting Ahrefs research

• AI Trafic Has Increased 9,7x in the Past Année (Linehan & Guan) — 81 947 sites.
• 63% of Websites Recevoir AI Trafic — a trackable floor, pas total AI influence.
• The Great Decoupling (Ryan Law) — impressions/clicks correlation flip; 34,5% CTR reduction.
• AI Visibility Audit (Gavoyannis & Law) — an 8-step Brand Radar audit; Mentions vs. Citations vs. Impressions.
• How to Choisir the Meilleur Prompts to Monitor Votre AI Search Visibility (Glen Allsopp) — 11 données sources pour a prompt définir; cadence and clustering.

From autour the industry

• Paul DeMott — the 5-Couche GEO Measurement Framework (Moteur de recherche Land, May 2026), the structural spine of ce hub.
• AI Trafic Attribution Crisis (Loamly, Nov 2025) — 446 405-visit sample showing 70,6% of AI trafic lands in Direct; source of the high-end dark-traffic estimate utilisé throughout ce hub.
• AI Referral Trafic Guide (Semrush, Sep 2025) — platform-by-platform breakdown of referral behavior from major AI outils; utile alongside Patrick’s propre referrer testing.
• ChatGPT Search Insights: 17-Month Analysis (Semrush, Apr 2026) — longitudinal regarder at ChatGPT’s share of AI-referred trafic and how the platform mix has shifted since launch.
• Gen AI Website Trafic Share (The Digital Bloom, Mar 2026) — January 2026 AI trafic share benchmarks à travers a grand site sample; bon pour contextualizing the 0,25% floor figure.
• SparkToro — Zero-Click Search Research — Rand Fishkin’s ongoing fonctionner on how few Recherche Googlees send a click; foundational context pour pourquoi AI Overview impressions sans clicks is an accelerating trend plutôt que an anomaly.

Stats worth citing

• AI trafic grew 9,7x over 12 months — my Evolve 2025 study of ~82 000 sites (and the 81 947-site Ahrefs analysis); average search trafic fell ~21% over the même period. Source
• 35–70,6% of AI visits land in Direct (aucun referrer) — range à travers studies; the élevé fin is Loamly’s 446 405-visit sample, so treat it as directional. Source
• 0,5% of Ahrefs visitors drove 12,1% of signups — a 23x conversion premium; AI visitors browsed ~50% plus pages per session. Industry-wide premium is ~4–4,4x. Source
• Seulement 16% of brands systematically track AI search performances (McKinsey, September 2025) — the opportunity gap.
• AI-visibility correlations (Ahrefs’ 75 000-brand study by Louise Linehan and Xibeijia Guan): branded web mentions 0,664, branded anchor text 0,527, branded search volume 0,392, Domain Rating 0,326. Source
• Seulement 38% of pages cited in Google AI Overviews ranked in the traditional top 10 (bas from 76%) — LLM visibility ≠ blue-link rankings.
• 3,6% of Ahrefs’ AI assistant trafic went to non-existent (hallucinated) pages — a measurement wrinkle with aucun traditional-search equivalent. Source
• The Great Decoupling: Ahrefs blog clicks/impressions correlation flipped from +0,425 to −0,352 après AI Overviews expanded; AIOs lié to a 34,5% CTR reduction. Source
• Google AI Overviews grew 58% from February 2025 to February 2026 — BrightEdge enterprise benchmark données, illustrating the scale of AI Overview expansion over the même period AI trafic was accelerating.

Journal des modifications