Ask AI About Your Brand And It Usually Gets It Right
A teammate said it in the weekly meeting: “I asked ChatGPT about us, and it was pretty accurate.” Case closed, as far as the room was concerned.
I was relieved too, in the moment. That report read as the AI-visibility check, done.
AI visibility means how a brand shows up inside AI answers — how often it’s named, and what gets said about it when it is.
Then a measurement across 150 brands landed, and the check that room ran turns out to test something narrower than it sounds like.
The number comes from Victorious, a US SEO agency, in its own Q2 2026 Quarterly Search Report.
AI Recognizes Most Brands When You Say The Name
Most brands get described correctly. Eight AIs described 150 brands, and 96% of those descriptions came back accurate.
The eight: ChatGPT, Claude, Gemini, Copilot, Perplexity, Google AI Overviews, Google AI Mode, and Meta AI.
By industry, five of the eight stayed at 83% or higher across the board — Google AI Mode, Gemini, ChatGPT, Google AI Overviews, and Copilot.
The other three dipped below that line somewhere: Claude ran 70-91%, Perplexity 47-100%, Meta AI 46-83%. Perplexity’s low point and Meta AI’s low point landed in the same industry — SaaS, at 47% and 46%.
Which of these eight is the one your actual buyers open?
This is the side where AI comes out ahead: it knows the brand. It’s also why a direct-name check usually comes back clean.
Ask The Category Question, And Recognition Stops Mattering
The same publisher ran the same 150 brands through a second test — a category question, no brand name in it. 89% of those brands never came up once.
The prompt shape: something like “who are the best law firms in the US” — the phrasing a buyer would actually type.
The dataset behind it: 5 industries (ecommerce/retail, SaaS, healthcare, legal, finance), 5,830 responses, 49,391 citations pulled out of them.
That 89% is 89% of the 150 measured brands.
Victorious’s own page also has a sentence that reads as “89% of the 96% that got described correctly” — but the summary at the top states it as a share of the full 150, and that’s the reading this article uses.
So the name check and the category check are two different measurements of the same AI, and they don’t move together.
Why ‘Knowing’ And ‘Mentioning’ Pull Apart
A name-in prompt pulls on what the model already knows. If a brand sits in the training data, it usually gets described right.
A category question is a different process — the model has to pick which brands to name as candidates. Knowing a fact and surfacing it in an answer are separate steps inside the model.
So a clean score on the name check says nothing about the category-mention rate.
The kind of source the model leans on shifts too, depending on which stage it’s answering.
Problem-awareness questions (“how do I know if I need a lawyer”) pulled mostly from video platforms and government sites — a 0.10% mention rate per response.
Category-comparison questions (“who’s the best law firm in the US”) pulled from directories, rankings, and comparison sites instead — 1.29% per response. Victorious calls that “more than 12x.”
Same brand, same set of AIs — the score depends on which stage you ask at. Track it quarter over quarter and you have to hold the stage fixed, or the number moving isn’t the model, it’s your own prompt drifting.
Reading It
Six conditions come with this 96% and 89%, from how the study was run and who ran it.
- Correlation, not causation. Mention probability correlates with third-party links (a 0.49 correlation with referring domains), but that isn’t an experiment showing more mentions cause more AI citations.
- 150 brands across 5 industries. How many brands sit in each industry isn’t published.
- The window is Q2 2026, start and end dates undisclosed. A Google core update (May 21-June 2, 2026) and a spam update (June 24-26) landed in that same window; Victorious names both as possible explanations for shifts in organic visibility.
- Claude and Meta AI had web search limited this round, so the citation figures run on the remaining six AIs only.
- Victorious sells SEO services. “Go get more third-party mentions” happens to be their business too.
- None of this measures what AI says about a brand — only whether the description is accurate, and whether the name shows up at all.
Discount for the correlation and for who published it, and one thing still holds: the name check and the category check measure two different things, and they don’t move together.
So which one is the prompt sitting in your own reporting deck?
Split Your Check Into Two, Starting Today
A name-in-the-prompt check measures accuracy. A category question measures whether a brand shows up at all — two different instruments, not one number wearing two names.
Today’s move: pull up whatever prompts are already running, and mark the ones that include the brand’s own name.
If every one of them does, add at least one category question — the kind a real buyer would type — and from next quarter on, track the two separately.
I ran that check on my own five prompts. Four had our name baked in. Zero were category questions.
What “knowing” without “surfacing” looks like from the model’s own side is in Say The Same Fact Lots Of Ways And ChatGPT Can Recall It — this piece adds the next step: once a model can surface a brand, does it actually put the name in the answer.
Swap one prompt, and a blind spot you couldn’t see before gets a name.
Sources
- Victorious, “Quarterly Search Report (Q2 2026),” https://victorious.com/quarterly-search-report/
- The URL above rolls forward each quarter. Archived snapshot of the Q2 2026 edition used here: Internet Archive (captured 2026-08-04)
