The AI Invisibility Report: 55% of brands are never recommended by ChatGPT
Over the last nine weeks, we ran AI [IN]visibility reports on 217 brands. Each report asks ChatGPT the questions a buyer would ask about the brand's offerings, then counts two things: how often the brand is mentioned, and how often it is recommended.
I expected the numbers to be bad. They were worse.
Each square is one brand. Hollow squares are brands ChatGPT never recommended in any answer.
The headline numbers
- 53% were never mentioned. Not once, across every question we asked.
- 55% were never recommended.
- The median brand scored 0% on both.
- Only 18 brands (8%) were recommended in half or more of the answers.
- 66% had every visibility gap in their report still open.
The average recommendation rate was 11.9%, but that average is carried by a small group at the top. Most brands sit at zero.
Found, understood, chosen
We think about AI visibility as three steps. The model has to find you, understand you correctly, and then choose you. The study shows where brands fall out.
Of 217 brands, 102 were mentioned at least once. Only 18 were recommended in half the answers or more.
The first wall is getting into the answer at all: 115 brands never cleared it. Once a brand is in the answer, it is usually recommended too. Only 5 brands were mentioned without ever being recommended. The hard part is doing it consistently: 39 brands reached one answer in four, and 18 reached one in two.
What the winners have in common
Three patterns came up again and again. Each one gets its own post in this series.
- Memory helps, but it doesn't decide. Brands ChatGPT knew moderately or well averaged 27% recommendation, against 7% for brands it didn't know. Yet 10 of those 49 well-known brands were never mentioned at all.
- Specific beats famous. Ten brands with no model memory were recommended in 40% or more of answers. They do one narrow thing: a music festival directory, a last-mile delivery service, an industrial brakes maker.
- Sector matters. 75% of marketing agencies were never mentioned, the worst of any sector with 10 or more brands. Healthcare did best: only 27% were never mentioned.
What I would do if I ran a brand
- Check whether you are in the answer at all. Ask ChatGPT the question your buyer asks, not your brand name.
- Fix your facts before you write more content. If the model has your offering wrong, more content repeats the error.
- Write for the question, not the keyword. The brands that won without fame answered one specific need clearly.
- Measure the same questions over time. One run is a snapshot, not a score.
How we measured
Each report asks ChatGPT a set of buyer-style questions about a brand's main offerings. Mention rate is the share of answers that name the brand; recommendation rate is the share that recommend it. Model memory is how much ChatGPT already knows about the brand. The 217 brands are ones whose owners ran a free report, plus brands our team audited, between Sep 1 and Nov 5, 2026. It is one ChatGPT run per brand, not a random sample of the market.
Want to see where your brand stands? Run your own AI [IN]visibility report at lumenario.com.
