AI NewsGo-to-marketReported

SparkToro finds 6.9 percent of ChatGPT chats show any purchase intent

SparkToro coded 41,961 ChatGPT conversations from 12,321 US panel users. Only 6.9 percent showed any purchase intent, and 87.9 percent of answers named no brand.

AI News

Editorial2 min read

LinkedInX
Pie chart of the 12 categories SparkToro coded for ChatGPT conversations by US panel users

Image: SparkToro

Why it mattersA GEO plan that treats ChatGPT like a buying channel is aiming at a small slice. Most answers do not name any brand at all.

The GEO work most brands are paying for right now is aimed at a small slice of ChatGPT chats. SparkToro coded ten months of real ChatGPT use and found that most answers name no brand at all. Rand Fishkin and his team published the study on 17 November 2026.

SparkToro worked with Verso, an opt-in panel service that collects chat history from users who link their ChatGPT accounts. The panel covered 12,321 US users. SparkToro coded 118,163 prompts across 41,961 conversations, from 1 January to 29 October 2026. Fishkin removed 131 heavy users whose volume would have skewed the shares.

How the chats were coded

Claude labelled each conversation against a 34-item codebook grouped into 12 categories. A second rater checked the labels on a subset. SparkToro reports agreement of about 90 percent on the categories and 97 to 99 percent on brand tags. SparkToro weighted the final shares to match the 2024 US Census.

Purchase intent is thin

Only 6.9 percent of conversations touched buying in any way. SparkToro breaks that number into two parts. The first is researching a purchase among options, which Fishkin groups with general product research. The second is asking about a specific brand or product, which is what shows real intent to spend money.

Fishkin states that Google receives purchase-journey queries at more than 16 times the rate of ChatGPT. He calls the result bad news for ChatGPT's advertising opportunities. SparkToro's own number for shopping, brands or local services across the whole panel is 5.3 percent of chats.

Most answers name no brand

Of all the answers in the panel, 87.9 percent contained no brand, product or company name. Only 12.1 percent of answers named at least one. Most GEO work is aimed at that smaller share of answers.

Image and design work was 16.1 percent of chats, which includes photo editing, poster design and other visual tasks. That is a larger share than every brand mention put together.

A reader buying GEO services should ask whose numbers back the pitch. The SparkToro study is one of the only recent panel reads of real ChatGPT chats at this scale. Its shares will not hold for every audience or every brand. A business whose category sits in that 12.1 percent should still work on being named. A business whose category does not should check that assumption before it spends.

Source

We analyzed 41,961 ChatGPT conversations to answer: how do Americans use AI tools in 2026? by Rand Fishkin at SparkToro. Covered by Search Engine Journal.

Reported bySparkToro

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project