ChatGPT’s share of AI prompts fell. Visibility work still has to cover Gemini and Claude.
Digiday’s Sep 22 write-up of Comscore’s Q2 2026 AI Intelligence Report says ChatGPT’s share of prompt volume fell from 70% to 50% between January and June 2026 while Gemini and Claude gained, and desktop Google searches with an AI Overview rose from 25.8% (Jul 2025) to 39.4% (Jun 2026). The same report split “sourced” from “cited.” Measure assistant by assistant. First-party pages still have to be crawlable.
By James Brady
Chief AI Officer, Product & AI Operations
ChatGPT is still huge. It is no longer the whole map.
That is not a vibe take. On September 22, 2026, Digiday published Comscore’s Q2 2026 AI Intelligence Report numbers: ChatGPT’s share of prompt volume fell from 70% to 50% between January and June 2026, Gemini’s share nearly doubled from 17% to 30%, and Claude climbed from 2% to 11% over the same window. Digiday also reports Comscore’s panel finding that Google desktop searches with an AI Overview grew from 25.8% in July 2025 to 39.4% in June 2026. Separately, Digiday quotes Comscore on a travel example: Tripadvisor accounted for 61% of travel sourcing by AI platforms but only 21% of citations.
Sources: Comscore data shows how AI discovery is splintering beyond ChatGPT, Digiday, September 22, 2026 (opened this fire); AI features and your website, Google Search Central (opened this fire); Optimizing your website for generative AI features on Google Search, Google Search Central (opened this fire); List of Google's common crawlers — Google-Extended, Google Crawling Infrastructure (opened this fire). Adjacent clock only: Research Shows Google AI Mode Sends Less Clicks & Is A Poor User Experience, Search Engine Journal / Roger Montti, September 21, 2026 (opened this fire; UPenn/Northeastern draft paper summarized there — we did not open the PDF).
The short version
- Digiday (Sep 22) + Comscore Q2 2026 AI Intelligence Report (via Digiday only): ChatGPT prompt-share 70% → 50% (Jan–Jun 2026); Gemini 17% → 30%; Claude 2% → 11%.
- Same Digiday piece: Google desktop searches with an AI Overview 25.8% (Jul 2025) → 39.4% (Jun 2026).
- Same Digiday piece: Tripadvisor travel example — 61% sourcing vs 21% citations. Sourced ≠ cited ≠ recommended.
- Digiday also reports June 2026 assistant visitation: 35% of desktop users and 29% of mobile users visited an AI assistant (vs 25% / 12% in January 2025), plus conversation counts for ChatGPT / Gemini / Claude — treat as panel estimates, not Google or OpenAI product metrics.
- Google Search Central still says supporting links in AI Overviews / AI Mode need pages that are indexed and snippet-eligible, with no extra technical requirements, and the generative AI optimization guide says allow crawling in robots.txt and by any CDN or hosting infrastructure.
- Google’s Google-Extended product token manages Gemini Apps / Vertex grounding and training uses of crawled content. Google says it does not impact Search inclusion and is not a ranking signal. That is not a multi-assistant citation dashboard.
- Operator split: measure visibility per assistant (and AI Overviews separately). Do not treat a ChatGPT screenshot as a GEO program. First-party proof pages still have to fetch.
- Monitoring ≠ execution. A Digiday/Comscore chart will not rewrite your service pages.
What Digiday / Comscore actually reported (via Digiday)
Digiday attributes the figures to Comscore’s Q2 2026 AI Intelligence Report, published the same day as Digiday’s article, using Comscore’s opt-in panel (Digiday: respondent base between 500,000 and 1 million users). Digiday quotes Comscore’s Smriti Sharma that overall AI assistant usage is growing while prompt share spreads across platforms, and that publishers should stop asking only how they show up in ChatGPT and start asking where their audience actually asks questions. Digiday also quotes Michael King (iPullRank) on measuring visibility per assistant and weighting by where readers are.
We opened Digiday. We did not open the Comscore PDF. If CoS needs primary Comscore pages for a number fight, hold ship until that fetch lands.
What Google confirms on the Search side
Google Search Central’s AI features guide: to be eligible as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible for a Search snippet. No additional technical requirements. The generative AI optimization guide reframes AEO/GEO as still SEO from Google’s perspective, tells owners to keep crawling allowed at robots.txt and CDN/hosting layers, and points measurement at Search Console’s Generative AI performance report — not at third-party “internal ranking” claims. Google-Extended is a separate robots token for Gemini Apps / Vertex training and grounding; Google says it does not affect Search inclusion.
Google did not publish Digiday’s Comscore prompt-share or AI Overview prevalence percentages. Do not put those numbers in Google’s mouth.
What this means for a New Reward operator
- Multi-assistant is the default measurement frame. If your weekly GEO deck is still one ChatGPT thread, you are measuring a shrinking share of the prompt pie Digiday attributes to Comscore.
- AI Overviews are a Search surface, not a chatbot. Digiday’s 25.8% → 39.4% desktop AIO prevalence (Comscore via Digiday) is about Google Search results pages. Fix crawl, index, and snippet eligibility the way Google documents — then measure generative impressions in Search Console when the report is available for the property.
- Sourced ≠ cited. Digiday’s Tripadvisor 61% / 21% split is the operator language: retrieval can use you without showing you. Track citations and recommendations separately from “the model saw the page.”
- First-party pages still win the boring way. Clear service facts, proof, entity identity, crawlable HTML. Google’s guide still prioritizes unique, non-commodity content over GEO file hacks. New Reward’s scan and Field Guide posts are about that loop — not about chasing one assistant’s UI this quarter.
- Do not collapse CDN training policy into Search crawl. Adjacent note for CoS queue: Cloudflare’s Sep 15 Disallow AI Training path (press + SEJ opened this fire) is how Accountable mixed-use crawlers keep search access while training preferences go to robots tokens such as Google-Extended. A Training Block can still stop Googlebot. That is a different post; do not ship the outdated Sep 11 draft.
Honesty block
- OpenAI did not confirm Digiday/Comscore prompt-share or conversation counts in any page we opened this fire.
- Google did not confirm Digiday/Comscore AI Overview prevalence percentages in any page we opened this fire. Google’s official pages we opened describe eligibility and measurement mechanics, not those panel rates.
- We did not open the Comscore Q2 2026 AI Intelligence Report PDF. Figures above are Digiday’s reporting of Comscore.
- Digiday’s Tripadvisor 61% / 21% example is Comscore-via-Digiday for travel — not a universal citation rate for every category, and not a New Reward client result.
- SEJ’s Sep 21 Montti write-up of the UPenn/Northeastern AI Mode field experiment is an adjacent click-behavior clock. We opened SEJ; we did not open the academic PDF. Do not import Montti’s editorial “is Google lying” framing into the Field Guide body.
- Google Search Central does not promise that multi-assistant citation work changes rankings. Indexing and serving are never guaranteed.
- No invented caps on how often New Reward customers get cited after a scan.
Related New Reward pages (live, HEAD 200 this fire)
- SEO and GEO move together
- GEO has a measurement layer now
- AI visibility tools: ChatGPT, Perplexity, brand tracking
- Why good brands disappear from AI search
- How to appear in AI search results
- Why your business isn’t showing up in ChatGPT
- How to rank in Perplexity
- AI Visibility Score
- Run a free AI visibility scan
- SEO / AEO / GEO offering · Scan · FAQ
Perplexity
Grok