Buying intent appears in 6.9 percent of ChatGPT conversations in a US panel that SparkToro analyzed with data from Verso, and 87.9 percent of ChatGPT’s responses named no brand, product, or business. Rand Fishkin, SparkToro’s cofounder and CEO, published the findings on November 17. For marketers sizing demand inside AI assistants, it is one of the few measured counts available.

The dataset covers 118,163 prompts across 41,961 conversations from 12,321 opt-in panel users, collected between January 1 and October 29, 2026. Verso, which connects to users’ ChatGPT accounts with their consent, supplied the data after cofounder Aaron Aknin approached Fishkin. Fishkin removed the 131 heaviest users, about 1 percent of the panel, because they sent 21 percent of all prompts. Search Engine Journal, which also covered the report, noted that Verso’s terms let it license de-identified conversations to research and business customers.

Most of the figures come from a subset of 9,675 panelists, weighted to 2024 US Census targets. After weighting, Fishkin says, the largest shift in any usage category was 1.4 percentage points.

Of the weighted conversations, 4.9 percent involved a person weighing options before a purchase, per Search Engine Journal’s reading of the chart. Another 2.1 percent asked about one specific brand, and 93.1 percent showed no purchase intent. Brand mentions were also sparse: 86.7 percent of prompts sought no brand information, and 12.1 percent of responses mentioned at least one brand.

Fishkin reads this as weak news for OpenAI’s advertising plans. “I suspect this is bad news for ChatGPT’s advertising opportunities,” he wrote. He compared the figure with SparkToro’s 2024 study of roughly 332 million Google searches, where he says 15 percent or more matched his definition of purchase research, and he guessed that Google users run buying-journey queries at 16 times ChatGPT’s rate or higher.

That comparison needs care. Search Engine Journal pointed out that the 2024 study labeled 14.5 percent of searches commercial and 0.69 percent transactional, categories that do not map onto the ChatGPT purchase-intent measure. Treat the two percentages as separate yardsticks. The 16x figure is Fishkin’s own estimate, not a measured result.

The post also contains a visible inconsistency. Its takeaways put buyer-journey activity at 12 percent of ChatGPT use, while the purchase-intent chart says 6.9 percent. The 12.1 percent figure appears to describe responses that named a brand, which is a different measure. Teams quoting the report should use the chart number.

On what people actually do in ChatGPT, images and visual design led at 16.1 percent in the post’s text, followed by information and entertainment at 14.0 percent and writing and communication at 11.9 percent. Shopping, brands, and local services came to 5.7 percent. Chart and text figures differ by under two points per category.

The limits are substantial. Panelists lean young, female, and lower income. Phones dominated the recordings, so laptop and employer-account activity is probably undercounted, and API requests are absent. Fishkin says the data describes the content of questions much better than the volume of use. Anthropic’s Claude labeled each conversation from a written codebook, and a second labeler agreed on 90 percent of category labels. No agreement rate was given for the purchase-intent label specifically. Readers should also remember that SparkToro, the publisher of the analysis, sells audience research tools.

Fishkin’s conclusion is opinion: he argues AI visibility matters mainly because of Google’s generative features (AI Overviews, plus AI Mode) instead of ChatGPT itself, and that tracking ChatGPT makes sense only alongside Bing rankings and AI visibility. The panel data does not test that claim.

For budget planning, treat 6.9 percent as a snapshot of a light-use consumer panel, and compare it with the ChatGPT referrals and ad results in your own analytics before shifting spend.

Source: SparkToro, “We Analyzed 118,000 ChatGPT Prompts from 12,000+ Users” by Rand Fishkin, published November 17, 2026, with additional reporting from Search Engine Journal.