A mention rate scores fifth place the same as first
Almost every AI visibility tool leads with one number: the share of answers that name your brand. The IAB's August 2026 measurement framework defines it the same way, responses with a mention divided by all responses, and calls it the most basic metric there is. It is. If you aren't named, nothing else matters.
But AI answers to buying questions usually name several companies. From October 4 to 9, 2026 we sampled 7,362 answers from ChatGPT, Perplexity, Claude and Google AI Overviews, to 1,131 non-branded questions across 40 accounts. Of the 6,546 answers that named any company, the median named five. A buyer reads that list from the top, and a mention rate gives the first name and the fifth exactly the same credit.
Same 70 of 160 answers, first place 38 times against 15
In one business software category we asked 20 questions, twice each, on four engines: 160 answers. Two tracked companies were each named in 70 of them, a 44% mention rate apiece. Any tool that stops at mention rate shows them tied.
They weren't. One was named first in 38 of its 70 answers, 54% (95% interval 43% to 65%). The other was named first in 15 of 70, 21% (13% to 32%). The intervals don't overlap. Their average places in the list were 2.3 and 2.9.
A second category is starker. Across 120 answers, two companies were named 36 and 34 times, 30% and 28%. The first led the list in 31 of its 36. The second led in 2 of its 34.
In 6 of 18 categories, the most named company wasn't the one named first most
Those two pairs are the extremes, so here is the wider picture. We took every company named at least 30 times in a category and found 17 pairs whose mention rates sat within two points of each other. In 5 of the 17, their first place shares differed by 20 points or more.
Then we looked at the 18 categories with at least three such companies. In 12, the company named most often was also the one named first most often. In the other 6, it wasn't. So a leaderboard sorted by mention rate would have put the wrong company on top in a third of these categories, if what you care about is who the buyer sees first.
Five questions to put to any tool before you pay
The IAB framework lists position next to mention rate as a core brand metric, and says providers must disclose how they measure it in prose answers, because there's no standard method. Use that as your opening question, then go through the rest.
- Where am I in the answer when I'm named? Ask for an average place or a first place share per engine, and ask how it's computed when the answer is a paragraph and not a numbered list.
- How many answers is each number built on? The IAB treats fewer than 50 queries in a program as exploratory. A rate with no sample size and no range can't tell a real change from noise.
- Can I read the answers? Every number should open onto the answer text and the sources the engine cited, so you can check the tool's reading of it.
- Is it split by engine? ChatGPT, Perplexity, Claude and Google can rank the same field differently, and a blended number hides which one you need to work on.
- Who decides the competitor list? Share of voice and position both depend on which companies are counted, so you should be able to see and change the set.
What we show, and what we don't
Our Visibility page shows each tracked company's average position when named next to its mention rate, and the daily changes call out when an engine starts or stops naming you first. Mention rates carry Wilson intervals and you can open the answers behind them. We don't measure where a name falls on screen, scroll depth, or whether anyone clicked.
Where this could be wrong
Position here is a company's place in the list of companies an answer names, as our extractor reads it. The IAB describes position as what a user sees in the rendered answer, which can differ, especially in long prose.
First named isn't always recommended. An answer can open with the incumbent in order to argue for someone else. We didn't score that.
Each category in this sample covers two days and four engines in early October, and each company count is small, 30 to 90 namings. The categories are the ones our 40 accounts track, and another set could look different.
Placement may not change what buyers do. We haven't measured clicks or sales from first place.
