ChatGPT shows purchase intent in just 6.9% of conversations, new study finds
A SparkToro analysis of more than 40,000 real ChatGPT conversations finds buying research is a small slice of usage, and argues AI-visibility budgets are pointed at the wrong assistant.

A new analysis of 118,163 real ChatGPT prompts from 12,321 U.S. users found that only 6.9% of conversations involved someone actively researching a brand to buy from, a far smaller share than the purchase-related query volume SparkToro has measured on Google.
The study comes from SparkToro co-founder Rand Fishkin, built on conversation data from Verso, a tool that lets people import and analyze their own ChatGPT history. Verso co-founder Aaron Aknin supplied the dataset: 41,961 complete conversations collected between Jan. 1 and Oct. 29, 2026, after Fishkin excluded the heaviest 1% of users, who alone generated 21% of all prompts. A subset of 9,675 panelists with demographic data was weighted against U.S. Census figures for age, gender, region, education and income; no category moved by more than 1.4 percentage points once weighted.
What the data shows
Each conversation received one label from a 34-category codebook Fishkin built, grouped into 12 broader categories. Claude did the labeling against that codebook, and a second human rater re-coded a random sample, agreeing with the machine on 90% of category calls and 97-99% of brand tags. Images and visual design led overall usage at 16.1% of conversations, ahead of information and entertainment (14.0%) and everyday life and admin (12.2%). Shopping, brands and local services made up 5.7% of conversations - present, but far from dominant.
The rest of the usage mix, in descending order: money, work and business (8.5%), schoolwork and learning (7.6%), technology support and coding (4.4%), relationships and emotional support (3.4%), chit-chat (3.0%) and creative writing or roleplay (2.9%). Fishkin also separated out a smaller group of self-identified tech workers, who asked far more coding and technical questions than the panel overall, and noticeably fewer schoolwork, image or relationship questions - a reminder that this panel's averages flatten real differences between consumer and professional use.
Within that mix, 12.1% of ChatGPT's responses named at least one brand, product or company, and 6.9% of prompts showed someone actively researching a purchase. Search Engine Journal's independent write-up of the same report breaks that 6.9% down further: 4.9% of conversations involved a person still comparing options, 2.1% were asking about one specific brand or product, and 93.1% showed no buying intent at all. Fishkin says he hand-checked the labeling and found better than 98% accuracy against his own read of the transcripts.
Why the Google comparison doesn't hold
Fishkin set that 6.9% against a 2024 SparkToro analysis of 332 million Google queries, in which at least 15% matched his definition of purchase-journey intent - a rate he estimates runs more than 16 times higher than what this ChatGPT panel shows. That gap supports the argument running through his post: ChatGPT behaves less like a search engine and more like a general-purpose assistant that occasionally brushes against commerce. "It's not a search engine. People don't treat it that way," he wrote. Search Engine Journal's Matt G. Southern quoted Fishkin putting the implication more bluntly: "I suspect this is bad news for ChatGPT's advertising opportunities."
Fishkin's own conclusion is that AI visibility still matters a great deal to marketers - just mostly because of Google's AI Overviews and AI Mode, which inherit Google's existing purchase-intent query volume, rather than because of ChatGPT on its own. He put ChatGPT's 12% buyer-journey share closer in scale to Bing's footprint than to Google's.
What a software marketer should do with this
The panel has real limits for a B2B audience, and Fishkin says so himself. It skews younger, female and lower-income, leans heavily toward phone use over desktop, and excludes API traffic entirely - which Fishkin guesses could account for half to three-quarters of all ChatGPT usage. A product evaluation running through an API-connected agent, or research done from a work account on desktop, is exactly the kind of B2B behavior this consumer panel is least likely to catch.
That caveat cuts against treating the 6.9% figure as proof that AI visibility doesn't matter for software marketing. It argues instead for where to point the budget already committed to it. Most GEO and AI-visibility tools built in the past two years lead with ChatGPT citations as the headline metric, because ChatGPT was the first chat assistant with mass consumer reach and the easiest one to prompt and screenshot. This data doesn't say that tracking is worthless - it says the tracking is currently weighted toward the assistant that research like this suggests carries the smallest share of a software buyer's actual purchase research, and away from the two properties, Google's AI surfaces and Bing, where Fishkin's numbers put more of that research happening.
- Weight ChatGPT-specific visibility work against Google's AI Overviews and AI Mode, where purchase-intent query volume is measured far higher
- Add Bing to the visibility tracking list - this data puts ChatGPT's buyer-journey share closer to Bing's scale than Google's
- Read this panel as directional for B2B, not definitive - it under-samples desktop use, work accounts and API-driven agent traffic
Don't cut ChatGPT-specific GEO work on the strength of one consumer panel, but don't let it crowd out Google AI Overviews, AI Mode and Bing visibility either - on this data, those are where software buyers' purchase-research queries actually concentrate.
Tools & AnalyticsA bug in Optimizely's signup flow let an attacker inject malware into 178 companies' websites
Sienna Mcpherson · 4 min read
Tools & AnalyticsBloomreach acquires AI-visibility platform AthenaHQ, tying AI citations to revenue
Sienna Mcpherson · 5 min read
AI