For three years, marketers have been asked to bet on AI assistants without a single reliable number about demand. Google gives you keyword volumes. ChatGPT gives you nothing.
That changed a little this week. And the first real number is smaller than most vendors would like.
What 118,000 prompts actually show
Rand Fishkin published the analysis on the SparkToro blog, built on an opt-in panel from Verso. The dataset covers 118,163 prompts across 41,961 complete ChatGPT conversations from 12,321 US users, January through October 2026. He removed 131 heavy users who sent 21% of all prompts, so the results describe typical people rather than power users.
The headline finding: only 6.9% of prompts showed any sign of researching a brand or product to buy. When SparkToro looked at 332 million Google queries two years ago, more than 15% matched the same definition.
Brands do show up more often in answers than in questions. About 12.1% of ChatGPT responses named one or more brands, products or companies. That is a meaningful surface. It is not a replacement for search.
The ratio that should reset your plan
Fishkin's estimate is that Google is roughly 5.5 times bigger than ChatGPT in total searches versus prompts. Add the purchase-intent gap and he lands on Google seeing buying-journey queries at something like 16 times the rate of ChatGPT.
That ratio looks a lot like Google versus Bing. Nobody serious ignores Bing. Nobody serious moves half their search budget there either.
This is the operational reality I keep repeating to clients. AI visibility is real, growing and worth measuring. But a board deck that treats ChatGPT as the new front door for purchases is selling a story the cold data does not yet support.
Why the panel probably undercounts the influence
Before anyone swings too far the other way, the data has limits worth respecting. The panel is consumer-heavy and excludes API usage. Fishkin himself notes that marketers are outliers, and his own usage looks nothing like the average.
There is also a second-order effect the panel cannot see. A prompt that asks "how do I fix plantar fasciitis" carries no purchase intent by the codebook. Yet the answer may mention a brand of insoles that the user searches on Google two days later.
I wrote about exactly this pattern in AI sends you traffic under a different name. The assistant shapes the shortlist. The transaction gets credited somewhere else. A 6.9% purchase-intent share understates influence, but it does correct the fantasy that assistants are a high-volume shopping mall.
Category matters too. A software buyer comparing CRM vendors behaves very differently from someone asking for a lasagna recipe, and the panel averages both. In B2B, where buying committees research for weeks, the share of commercially relevant prompts is likely higher than 6.9%.
The direction of the finding holds. The exact number for your category is something you have to measure yourself.
What this means for how you spend
The practical move is proportion. If your category has real purchase research in AI assistants, such as software, travel, health products or professional services, then measure your presence there and fix the gaps. If your category is mostly impulse retail or local services, Google and Maps still carry the weight.
The measurement question becomes sharper too. Brands appear in 12.1% of answers, so the real question is not "are people shopping in ChatGPT?" but "when the topic touches my category, am I in the answer, and why?"
If you want to measure where your brand actually stands in AI-generated answers, GEOflux.ai is built specifically for this. It maps not just whether your brand is mentioned, but why, and what to do about it.
I would also treat this as a reminder about how we size any new channel. In AI visibility is a billboard, not a pipe, I argued that assistants build preference rather than send traffic. The SparkToro numbers fit that model well.
A billboard is valuable. You just do not judge it by how many people walk straight from the billboard into the shop.
The advertising angle nobody wants to price
There is one more implication, and it is commercial. OpenAI is building an advertising business. Advertising businesses are valued on commercial intent, and Google became one of the richest companies on Earth because a large share of its queries are people about to spend money.
If only about 7% of consumer prompts carry purchase intent, ChatGPT ads will look more like display and social than like search. That is not a death sentence. Meta built an empire on interruption rather than intent.
But it changes what advertisers should expect. Lower intent usually means lower conversion rates, more reliance on attribution models, and more pressure to prove incrementality. Buy it like a brand channel, measure it like a brand channel, and do not let anyone sell it to you at search prices.
The first real dataset rarely confirms the hype. It usually confirms the fundamentals. People still go to Google when they are ready to buy. They go to ChatGPT when they are trying to understand something.
Be in both places. Spend in proportion to the data, not the noise.