SparkToro and Verso: 6.9% of 118,000 ChatGPT prompts show purchase intent

SparkToro and Verso: Only 6.9% of 118,000 ChatGPT prompts show purchase intent

AINovember 18, 2026
By Antonio Fernandez

TL;DR

  • SparkToro published on 17 November 2026 an analysis of 118,163 prompts from 12,321 US panel users (Verso data, January to October): 6.9% showed anything related to researching a brand to buy from.
  • 87.9% of ChatGPT responses named no brand, product or company; 12.1% named at least one. Fishkin cites 15%+ purchase-journey matches in his earlier 332 million Google query study.
  • Limits: opt-in US panel skewing young, female and lower income, mostly phones, no API traffic, no Thai data, and no conversion data. SEJ notes the report's 12% and 6.9% figures differ.

SparkToro published an analysis on 17 November 2026 of 118,163 ChatGPT prompts from a US panel supplied by Verso, and found that only 6.9% of prompts showed anything related to researching a brand to buy from. In the same data, 87.9% of ChatGPT responses named no brand, product or company, and 12.1% named at least one. Rand Fishkin, SparkToro's co-founder, calls the ChatGPT purchase intent figure bad news for ChatGPT's advertising opportunities, while arguing that AI visibility still matters a great deal. This is one opt-in US panel, not all ChatGPT usage, and the sections below separate what was measured from what was concluded.

What SparkToro and Verso analysed

According to the SparkToro report, Verso makes opt-in, anonymized connections to OpenAI, and Verso co-founder Aaron Aknin supplied the data behind it. The dataset covers 118,163 prompts across 41,961 complete ChatGPT conversations from 12,321 panel users, between 1 January and 29 October 2026.

SparkToro removed the 131 heaviest users, defined as anyone averaging more than 33 prompts per month. They were about 1% of panelists but sent 21% of all prompts, so leaving them in would have made the results describe a small group of power users. Most of the charts in the report are per user rather than per prompt, so each person counts equally. SparkToro also reports a census-weighted subset: 9,675 panelists with demographic records were weighted to 2024 US Census targets for age, gender, region, education and household income. Search Engine Journal, which covered the study the same day, adds that each member of the weighted group sent an average of 9.9 prompts across the period.

How the conversations were labelled

SparkToro says it built a codebook of 34 fine-grained intents grouped into 12 broader categories, and that each conversation received exactly one label based on its main goal. When a person asked ChatGPT to produce something, the output type won over the topic: a request to write a refund email to PayPal counted as writing and communication, not shopping.

Every conversation was read and labelled by an AI model, Anthropic's Claude, working from the written codebook rather than keyword rules. For each conversation the labeller saw the auto-generated title, up to six of the user's prompts, the opening of ChatGPT's first few replies, and any product or place cards and bolded names in the replies. A second, independent labeller re-coded a random sample without seeing the first labels. The two agreed on 90% of category labels (kappa of about 0.88) and on 97% to 99% of the brand tags. SparkToro adds that Fishkin hand-checked the numbers and saw better than 98 out of 100 accuracy. Search Engine Journal notes that the article gives no agreement figure for purchase intent specifically.

What people asked ChatGPT, by category

SparkToro's per-user breakdown of the 12 top-level categories was as follows:

  • Images and visual design: 16.1%
  • Information and entertainment: 14.0%
  • Everyday life and admin: 12.2%
  • Writing and communication: 11.9%
  • Health and wellness: 10.3%
  • Money, work and business: 8.5%
  • Schoolwork and learning: 7.6%
  • Shopping, brands and local services: 5.7%
  • Technology support and coding: 4.4%
  • Relationships and emotional support: 3.4%
  • Chit-chat and other: 3.0%
  • Creative writing and roleplay: 2.9%

The shopping category is defined by SparkToro as deciding what to buy or where to go, comparing products, or finding a nearby business and its hours, phone number or address. It is a separate measure from the purchase-intent question, which looked at whether a conversation suggested someone was researching, comparing or deciding on a purchase. SparkToro also split out a group it labelled tech workers, who ask far more coding and technical questions and fewer homework, image and relationship questions than the rest of the panel.

The purchase-intent and brand figures

For brands, SparkToro made three classifications of prompts: no request for a brand, product or company; an active request for them; and a question about one specific brand. For purchase intent it used three classes of conversation: no purchase intent, researching options to buy, and analysing one specific brand or product with purchase-relevant questions. The figures below combine SparkToro's text with the chart breakdown reported by Search Engine Journal (writer Matt G. Southern).

The purchase-intent and brand figures
MeasureFigureWhere it comes from
Prompts showing anything related to researching a brand to buy from6.9%SparkToro text and SEJ chart headline
Conversations planning a purchase and comparing options4.9%SEJ chart
Conversations asking about a specific brand, product or service2.1%SEJ chart
Conversations with no purchase intent93.1%SEJ chart
Prompts that did not ask for brand, product or company information86.7%SEJ report of the weighted group (6.6% asked about several brands, 6.7% about one)
Responses that named no brand, product or business87.9%SEJ report of the weighted group (7.0% named several, 5.1% named one)
Responses that named at least one brand, product or company12.1%SparkToro text

Two details matter when quoting these numbers. SparkToro's text says "prompts", while Search Engine Journal describes the same 6.9% as conversations. Analysis: the 4.9% and 2.1% sum to 7.0% rather than 6.9%, which looks like rounding, though neither source addresses it.

The comparison with Google, in Fishkin's words and numbers

SparkToro compares the 6.9% with its earlier study of 332 million Google queries, conducted two years ago, in which more than 15% matched the same definition. Fishkin writes that Google is about 5.5 times bigger than ChatGPT in searches versus prompts and roughly 2.8 times bigger in the share with purchase intent. He then offers a conservative guess, labelled as a guess, that people ask Google purchase-journey queries at 16 times the rate or more of ChatGPT, and notes that this is not unlike the Google to Bing ratio.

His takeaways, as stated in the report, are that ChatGPT is a heavily consumer app; that the competition with Google has been overblown by marketing press and influencers because people treat ChatGPT as a confidant, creator tool, entertainment product or assistant rather than a search engine; and that 12% of ChatGPT usage is part of the buyer journey, which he describes as closer to the size of Bing than Google. He also writes that AI visibility matters a huge amount, but not because of ChatGPT: he attributes its importance to Google AI Overviews and AI Mode, and says tracking ChatGPT visibility is worthwhile only alongside keyword rankings and Bing visibility.

Search Engine Journal points out that the 12% in Fishkin's takeaways and the 6.9% in the purchase-intent chart differ. The report does not explain the gap in the passages available to us, so quoting either figure should name which one it is.

Limits and what the source did not say

SparkToro lists several limits itself. Like any opt-in panel, this one does not perfectly match the US population: it skews younger, female and lower income. It tested weighting to Census targets and reports that no category moved by more than 1.4 percentage points. Most panelists were recorded on phones, so desktop and work-account use is likely under-represented, and API requests are missing. SparkToro states the consequence plainly: the panel tells us what people ask ChatGPT far better than how much they use it. Fishkin guesses that API use may be 50% to 75% of all usage, and says it probably carries very different prompts. That is his estimate, not a measurement.

It is also worth listing what the report does not contain, based on the text we read:

  • No data from Thailand or any market outside the United States.
  • No data on conversions, sales or revenue that followed a ChatGPT conversation. Purchase intent here is a label on what was asked, not a result.
  • No measurement of Google AI Overviews or AI Mode. Fishkin's claim that those surfaces drive the importance of AI visibility is his conclusion, not a finding of this dataset.
  • No explanation of the 12% versus 6.9% difference.
  • Nothing on whether or when ChatGPT advertising is available in any particular country.

Disclosure for context: SparkToro sells audience-research software, and Verso is a tool that analyses users' linked ChatGPT history. The labelling model was Anthropic's Claude. None of that invalidates the numbers, but it is the full set of parties involved in producing them.

What to check in your own data

Analysis, not from the source: a panel average says little about your category, so test the claim against your own evidence before changing budgets.

  • In Google Analytics 4, look at acquisition reports for visits that arrive from ChatGPT, and compare their landing pages and conversion rates with organic search. Referral data from AI tools can be incomplete, so treat the count as a floor.
  • Sample the questions your buyers actually ask. Run a fixed list of 20 to 30 category and brand questions in ChatGPT on a schedule and record whether your brand, a competitor or no brand is named. That reproduces, at small scale, the brand-in-response measure SparkToro used.
  • Check Search Console for the queries that lead to your pages, and compare them with the prompts you collected. If the purchase-journey questions in your market still arrive mainly through Google, that is consistent with SparkToro's reading.
  • Separate reporting for ChatGPT, Google AI Overviews and AI Mode wherever your tools allow it, so a single AI visibility number does not hide three different surfaces.

What this means for Thai marketers

Analysis, with the study's scope in mind: the panel is American, so it cannot tell us how Thai users prompt ChatGPT or what share of their conversations involve buying. What it offers is a sizing reference. If a US panel with heavy consumer use shows 6.9% purchase-intent conversations, then a business that expects ChatGPT to replace Google for product research has a number to test that expectation against. The report does not establish that the Thai figure is similar, higher or lower, and nobody should present it as such.

A practical reading is to treat ChatGPT work as one part of a wider AI visibility plan rather than a separate budget line sized by hype. Fishkin's own conclusion points the same direction: the surfaces he considers most influential for AI visibility are Google's. For brands building that plan, our pages on ChatGPT SEO and generative engine optimization cover the visibility side, and the ChatGPT Ads page covers the paid side. The study says nothing about ChatGPT ads in Thailand, so any paid plan there should rest on current availability checked directly, not on this report.

FAQ

Does only 6.9% of ChatGPT use involve buying?

In this US panel, 6.9% of prompts showed anything related to researching a brand to buy from, according to SparkToro. The dataset covered 118,163 prompts from 12,321 users between January and October 2026, and it excludes API traffic, so it does not describe all ChatGPT usage.

Why does Fishkin also say 12% of ChatGPT usage is part of the buyer journey?

The report does not explain the difference in the passages we read. Search Engine Journal flagged that the 12% in the takeaways and the 6.9% in the chart do not match, and the 12.1% figure elsewhere in the report refers to responses that mention at least one brand.

Does this mean AI visibility does not matter?

No, and Fishkin says the opposite: he writes that AI visibility matters a huge amount. His argument is that the importance comes from Google AI Overviews and AI Mode rather than from ChatGPT, and that ChatGPT tracking makes sense mainly alongside keyword rankings and Bing visibility.

Is the panel representative of US ChatGPT users?

Partly. SparkToro says the panel skews younger, female and lower income, and that desktop and work use is likely under-reported. After weighting 9,675 panelists to 2024 Census targets, it reports that no category moved by more than 1.4 percentage points.

Does this apply to Thailand?

The source does not say. The data is a US panel, so Thai ChatGPT usage and purchase-intent share would need to be measured separately, for example through your own analytics and prompt sampling.

If you want help separating ChatGPT, Google AI Overviews and AI Mode in your reporting, Relevant Audience can review your current setup. Start with the GEO service page or send a note through the contact form.

Antonio Fernandez

Antonio Fernandez

Founder and CEO of Relevant Audience. With over 15 years of experience in digital marketing strategy, he leads teams across southeast Asia in delivering exceptional results for clients through performance-focused digital solutions.

Share to:
Copy link:

Read us often? Add Relevant Audience as a preferred source so our articles surface more in your Google results.