adtestbench Higgsfield
EN

AI news · ChatGPT · purchase intent

SparkToro finds shopping intent in 6.9% of ChatGPT chats

SparkToro’s co-founder reads a US panel of 12,321 ChatGPT users as bad news for ChatGPT’s ads; Search Engine Journal says his Google comparison uses different categories.

By Alexandre S. , 07:30 UTC

Only 6.9% of ChatGPT conversations in a US panel showed any sign that the person was weighing something to buy, according to an analysis by SparkToro, the audience research company, covered by Search Engine Journal on 17 November 2026. Verso, a service that logs the ChatGPT history of people who connect their accounts, supplied the data. The full set is 41,961 conversations from 12,321 users between 1 January and 29 October 2026, and most figures come from a group weighted to mirror the US population.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a shopping cart standing on the horizon.
Drawn by adtestbench from “We analyzed 11,800 ChatGPT prompts from 12,000 users”,

Most of the chats, 93.1%, showed no buying intent at all. In 2.1% the person asked purchase questions about one specific brand, product or service, and in 4.9% they meant to buy and were still weighing options. ChatGPT’s own replies rarely pointed anywhere: in the weighted group, 87.9% of them mentioned no company, brand or product.

Co-founder Rand Fishkin compares that with SparkToro’s earlier study of about 332 million Google searches, in which 15% or more matched the same buying definition. He suspects the result is “bad news for ChatGPT’s advertising opportunities”, and concludes that Google’s AI Overviews and AI Mode are what make it important for a brand to be named in AI answers. Search Engine Journal adds a caution: the Google study sorted searches into different categories, so the two percentages do not line up exactly.

For an advertiser weighing ChatGPT’s ads, the panel puts buying-related chats at under 7%, and 87.9% of the answers named no brand at all; Search Engine Journal notes the panel reflects light consumer use, recorded mostly on phones, with desktop and work use likely under-counted and no API traffic.