Study Finds AI Brand Recommendations Rarely Repeat Day to Day
Glen Allsopp of Detailed tracked more than 70,000 ChatGPT and Google AI Overviews responses to the same commercial prompts over 28 days. If you pick two random days, ChatGPT returned an identical set of brands for a prompt only 0.3% of the time. A small group of core brands stayed stable while most other names changed daily. Allsopp advises brands not to judge their AI visibility from a single day of results.
Allsopp checked more than 1,300 prompts once a day from 26 August to 22 September 2026, using a US account that was not logged in. The prompts were mostly commercial questions in categories such as software, agencies, finance, fashion and product recommendations. Detailed is owned by Ahrefs, where Allsopp works on marketing strategy and research.
The brand each platform named most often for a prompt appeared on 82% of days in ChatGPT and 89% in AI Overviews. In ChatGPT, brands named on at least 80% of days changed by 13% from one day to the next, while all other brands changed by 78%. The usual leader was named first on 50% of days in ChatGPT and 60% in AI Overviews. In ChatGPT, on days it was not named first, it was usually missing from the answer entirely.
The two platforms agreed on the most-present brand for the same prompt only 27% of the time. About a quarter of the pages ChatGPT cited for a prompt were cited again the next day, compared with about half in AI Overviews. Parse found a similar pattern in a separate analysis of 693,509 answers published in July 2026. In that analysis, repeat ChatGPT answers to the same prompt shared an average of 21.2% of their cited domains.
Some of the measured volatility comes from how brands are named. For 55 brands that Allsopp checked by hand, 36% of apparent drop-offs in ChatGPT were the same company appearing under a different brand or product name. He recommends that brands tracking their AI visibility count every variant of their name. He also advises them to focus on overall presence rather than being named first.
The number marketing teams should worry about is 0.3%. That is how often ChatGPT gave the same list of brands to the same question on two different days. Plenty of companies still judge their AI visibility by asking ChatGPT once and putting the screenshot in a board deck. Allsopp's data shows that screenshot captures one day's answer and little else. The stable part is small, usually one or two brands per prompt. A brand is either in that core or it rotates through the tail.
The 27% figure deserves as much attention. ChatGPT and Google agreed on the leading brand for only about one prompt in four, and people ask more assistants than those two. Allsopp himself names Claude and Google's AI Mode as the platforms he wants to add next. If two platforms already disagree this often, a report that covers only one or two of them leaves most of the picture blank.
The practical lesson is to measure rates across many answers. Ask many questions, phrased the way different customers phrase them, across the assistants people actually use, and count every variant of the brand's name. That is the idea behind the AI Readiness Check: thousands of data points across seven AI assistants, so that no single answer decides the result.
