Your brand now lives in the model's memory. The tools cannot agree on how to count it
72% of ChatGPT memories mention a brand. OpenAI's Business Hub counts only linked mentions. The gap between the two is the new SEO.
- 72% of ChatGPT memories mention at least one brand, and 47% of memories contain at least one buying signal, in Evertune's analysis of 1,657 ChatGPT memory records. Self-reported m:…6826
- 24% of ChatGPT responses currently draw on stored memory, and prompts asking for recommendations trigger memory 73% more often than average. Self-reported m:…6826
- For online businesses, a mention in Business Hub is defined as a reference to the business's website, and brand mentions without a link to the website occur nearly 50% of the time. Self-reported m:…2205 m:…2160
- OpenAI's Business Hub launched on November 10, 2026, and Business Hub data windows are limited to seven-day and 28-day views. Self-reported m:…2205
The memory is the market
Evertune analyzed 1,657 ChatGPT memory records from Gener8 panelists, and the headline is that the model's memory is already a brand channel. 72% of ChatGPT memories mention at least one brand, and 47% of memories contain at least one buying signal. Brand and product preferences account for 25% of all memories, and 90% of memories mention 10 or fewer brands. Apple is the most mentioned brand in ChatGPT memories, appearing in 258 records. m:…6826
Memory is still early in how much it shapes answers. 24% of ChatGPT responses currently draw on stored memory, and Evertune expects the 24% memory usage rate to increase as AI companies improve personalization. Prompts asking for recommendations trigger memory 73% more often than average, and prompts referring to the user trigger memory 41% more often than those that do not. Over 80% of AI prompts are unique, and the median AI prompt length is 19 words. m:…6826
The implication is straightforward. If the model remembers your brand, and the user asks for a recommendation, the memory is in the room. If the model does not remember you, you are not in the conversation at all. And because over 80% of prompts are unique, the old keyword game does not work: you cannot optimize for a query nobody repeats. You have to be in the memory. m:…6826
What the Hub counts
OpenAI launched Businesses in ChatGPT in beta this week, and OpenAI's Business Hub launched on November 10, 2026. Eligible businesses can see how often ChatGPT references their website, how many recorded clicks those links receive, and which topics those conversations cover. m:…2160 m:…2205
But the Hub counts something narrower than a brand mention. For online businesses, a mention in Business Hub is defined as a reference to the business's website, and for local businesses, a mention is defined as showing a result for that location. Business Hub mentions behave like citations because they follow the domain rather than the brand name, and brands sold through retailers and marketplaces are disadvantaged by the link-based mention definition. Brand mentions without a link to the website occur nearly 50% of the time, which means the Hub misses about half of what the model actually says about a brand. m:…2205 m:…2160
OpenAI chose link-based counting because domain references map cleanly to accounts and avoid brand name ambiguity, and website clicks in the Hub are defined as clicks on links from supported ChatGPT surfaces. Business Hub insights are only displayed once a sufficient volume of conversations or interactions is reached, and data windows are limited to seven-day and 28-day views. Since October 9, 2026, many ChatGPT visits arrive without a referrer or UTM tag, which makes the Hub's first-party data more valuable and harder to cross-check. m:…2205
The agency layer
The measurement gap is already a business. TRYSEO is a six-person agency based in Magdeburg, Germany, that began offering GEO services in 2023, and treats GEO as an extension of a strong SEO foundation. TRYSEO's GEO work includes prompt tracking, technical changes, Digital PR, media placements, and ongoing measurement, and uses OtterlyAI for prompt tracking, Brand Reports, and citation tracking. TRYSEO's positioning interview process draws knowledge from people close to the product, particularly engineers and sales specialists. m:…8867
The results TRYSEO reports are large. Spreadshop saw a 355% increase in impressions following TRYSEO's engagement, and a 133% increase in non-branded clicks following the engagement. TRYSEO charges a one-time setup fee between €4,000 and €6,000, and a monthly retainer between €2,000 and €4,000, and does not offer success-based fees. Profound, meanwhile, runs tracked prompt sets across ChatGPT, Google AI Mode, Google AI Overviews, Gemini, Perplexity, Claude, and DeepSeek, and reports share of voice and rank against competitors broken down by prompt, topic, model, region, and persona. m:…8867 m:…2160
The lesson is that AI search has scoreboards that disagree. The memory scoreboard says your brand is in 72% of memories and half the mentions have no link. The Hub scoreboard says only the linked mentions count. The agencies selling GEO are arbitraging the gap, and the brands that win will be the ones that show up in both: remembered by the model, and cited with a link when the model answers. m:…6826 m:…2160 m:…2205
What is still unresolved
- How fast the 24% memory usage rate rises as personalization improves.
- Whether the Hub expands beyond seven-day and 28-day views, or adds brand-name tracking alongside domain citations.
- How much of TRYSEO's reported impression growth is attributable to GEO versus the underlying SEO work.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.