ChatGPT: SparkToro Measures 6.9% Purchase Intent

SparkToro identifies a purchase intent in 6.9% of ChatGPT conversations from a U.S. panel, observed between January 1 and October 29, 2026. This measure does not constitute a count of commercial exchanges across the entire service: the most active users were excluded, and some usages remain underrepresented.
SparkToro removes 131 profiles sending more than 33 monthly queries
Before calculations, the 131 most active participants were excluded. They submitted an average of over 33 queries per active month and accounted for about 1% of the panel, but 21% of the queries. After this exclusion, the average of the weighted group reaches 9.9 queries per member between January and October. SparkToro notes that this data provides more insight into the nature of requests than the frequency of use.
The data collection was primarily conducted via phone, with a possible underestimation of usage on computers or professional accounts, SparkToro acknowledges. API calls are absent. The panel also overrepresents young people, women, and individuals with low incomes. Demographic information from 9,675 members allowed for weighting according to the 2024 U.S. census; according to SparkToro, no usage category varies by more than 1.4 percentage points.
The initial data provided by Verso includes 41,961 conversations from 12,321 individuals, with most results based on a weighted subsample. Verso collects the history of individuals who connect their accounts and offers to analyze the content for free. Its terms stipulate that imported conversations will be retained and then integrated without identifiers into datasets licensed to research and business clients. According to SparkToro, Aaron Aknin, co-founder of Verso, reached out to the company to provide this data.
SparkToro distinguishes between planned purchases and brand presence
In the details of the purchase indicator, SparkToro classifies 4.9% of conversations as those where a purchase is planned, but options are still being explored. Questions regarding the purchase of a specific brand, product, or service represent 2.1%. Conversely, no purchase intent is identified in 93.1% of exchanges from the weighted group.
Commercial presence in the questions is counted separately. According to SparkToro, 86.7% of queries do not request any information about a brand, product, or company. Queries concerning multiple brands account for 6.6% of the total, compared to 6.7% for a single brand.
On the response side, SparkToro notes an absence of brand, product, or company in 87.9% of cases. Multiple brands are mentioned in 7.0% of responses, and a single brand in 5.1%, for a reported proportion of 12.1% containing at least one brand. SparkToro also qualifies 12% of ChatGPT usage as part of the purchasing journey in its conclusions, although this proportion does not correspond to its purchase intent indicator.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Comparisons with Google change categories
Rand Fishkin, co-founder and CEO of SparkToro, a company that sells audience research software, considers these results unfavorable for ChatGPT's advertising prospects. He primarily attributes the importance of visibility in AI to Google’s AI Overviews and AI Mode. He also questions the value of tracking presence on ChatGPT without simultaneously observing Bing rankings and visibility in AI.
For Google, Rand Fishkin mentions that 15% or more of searches correspond to his definition. SparkToro analyzed approximately 332 million searches conducted in browsers by a U.S. panel between January 2023 and September 2024. It classified 14.5% of searches as commercial, a category that includes finance, business services, and commerce, without assuming an immediate purchase; transactional searches accounted for 0.69%. These categories differ from those applied to ChatGPT: their proportions do not directly compare.
Another comparison made by SparkToro: OpenAI's Signals data for September places more emphasis on writing and technical assistance, while the panel prioritizes images and practical requests. Rand Fishkin defends the comparison by suggesting that OpenAI almost certainly includes API calls. However, OpenAI specifies that it only counts messages from ChatGPT Free, Go, Plus, and Pro accounts, excluding Codex, business uses, and APIs. The units and territories also differ: OpenAI counts messages globally, while Verso counts U.S. conversations; nonetheless, Rand Fishkin wishes to continue research with Verso.
Claude categorizes requests by their primary objective
According to SparkToro's graphical distribution, images and visual design top the usage with 17.6% of conversations. Information and entertainment follow at 14.4%, ahead of writing and communication at 12.7%. The category encompassing purchases, brands, and local services reaches 5.3%. These graphical values differ from those in the text accompanying the results by less than two percentage points per category.
Each conversation received only one category, determined by its primary objective. When a creation is requested, the expected format takes precedence over the theme: a message requesting a refund from PayPal falls under writing, not purchasing. The classification was performed using Claude, Anthropic's model, based on a written grid. It utilized the conversation title, up to six user requests, the first lines of the initial responses, as well as any product sheets or bolded names.
A second coder independently reviewed a random sample. SparkToro reports a 90% agreement on categories and 97% to 99% on brand labels. No specific agreement rate for detecting purchase intent is provided.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.