ChatGPT's share of AI prompts fell to 50% from 70% in six months, and most GEO dashboards still watch only one assistant

Software marketers who track AI visibility by watching ChatGPT alone are now missing half the picture, according to Comscore's Q2 2026 AI Intelligence Report. In the six months to June 2026, ChatGPT's slice of all AI prompts slid to 50% from 70%, while Google's Gemini nearly doubled its own slice and Anthropic's Claude more than quintupled it.

What changed

The report, released September 22 and built from a Comscore panel of 500,000 to 1 million opted-in respondents, still shows ChatGPT well ahead on raw usage. By June 2026 it was fielding 168 million desktop conversations a month, and its footprint across multiple platforms grew to 99 million visitors from 59 million a year prior. What's shrinking is its share of the total. Gemini climbed to 30% of prompt volume from 17% over the same span, a jump Comscore and independent analysts tie to its bundling inside Google Search, Android and Workspace. Claude's trajectory is the steepest: usage started near 1.1 million desktop conversations back in October 2025, climbed toward a high point of nearly 40 million by that April, then eased to 22.3 million by June — still roughly twenty times where it began less than a year earlier.

The shift isn't confined to standalone chatbots. Google's AI Overviews accounted for 39.4% of desktop searches by June 2026, up sharply from 25.8% the prior July, and Microsoft's Copilot Search rose to 17.3% of search volume from 11.8% across the same window. Assistant reach broadened too: by June 2026, more than a third of desktop visitors and roughly 29% on mobile had used an AI assistant, up from a quarter and 12% respectively in January 2025.

Comscore's data also separates two things marketers tend to conflate: being pulled into an AI answer, and being named in it. Tripadvisor supplies 61% of the travel results AI platforms draw on, per the report, yet shows up in only 21% of the citations those platforms actually display to users.

"Being used by AI is not equal to being visible in AI," says Comscore chief analytics and technology officer Smriti Sharma.

Who it affects

Any team reporting AI visibility as a single ChatGPT-mentions number is now working from a partial dataset — and so is anyone buying ChatGPT ad placements as though that's the entire paid AI channel available to them. Comscore also points to AI search activity building inside TikTok, Meta and YouTube, surfaces most marketing teams aren't tracking at all yet. Paid and demand-gen teams have a second reason to pay attention: sponsored placements inside AI answers are expanding fast. Among hotel-search prompts that carried source links, the sponsored share jumped to 24% in May 2026 from 6% two months earlier — evidence that assistants are starting to charge for the same answer real estate brands are chasing for free.

Content and PR teams have their own stake in this. Publishers including Forbes and Time have started pitching their citation rates across AI answer engines to advertisers as a standalone metric, even though there's still no agreed way to measure AI visibility consistently across platforms, and even they have found that a high citation rate doesn't scale into proportional referral traffic. Software vendors watching competitors court the same citation-as-currency pitch should expect it to show up in analyst briefings and partner decks before it shows up in a standard reporting template.

What a software marketer should do differently

The fix isn't a bigger ChatGPT dashboard — it's a different measurement model. Michael King, founder and CEO of SEO agency iPullRank, told Digiday that sourcing and citation are separate stages that deserve separate tracking: "Retrieval happens in two stages. First the assistant fans out into a bunch of sub-queries and pulls in a wide set of pages. That's sourcing." Citation, he said, only attaches to the passages a model leans on directly, which makes citation counts alone a lagging signal rather than a complete one. His recommendation is to track the whole chain instead of one link in it: whether a page gets pulled into retrieval at all, whether it earns a citation once retrieval happens, and whether either step eventually shows up as a referral or a signed deal. A team that only watches the citation step is optimizing for the metric that's easiest to screenshot, not the one that predicts revenue.

• Split AI-visibility reporting by assistant — ChatGPT, Gemini, Claude and Copilot each retrieve and cite differently, so one dashboard number hides the shift.
• Weight tracking toward where your buyers actually prompt, not toward whichever assistant currently has the most volume.
• Build pages around original data and self-contained answers — the trait Comscore and King both tie to earning a citation rather than just a retrieval.
• Watch your category's AI answers for sponsored placements now, before a competitor buys the slot you assumed was earned media.