By Alex Jordan, Founder
The step
We query the engines about a brand from training data alone — no web access, no help. Then we compare each answer against what the brand's own homepage actually says.
We run it across four engines: ChatGPT, Claude, Gemini and Perplexity. Every answer is scored against what the brand's own site says, and the result comes back as the three components below: Recognition, Accuracy and Positioning.
The gap between those two things is the entire finding. It is where a model has invented a feature, aged out a price, attached the wrong founder story, or quietly recommended a competitor instead.
What the score is made of
One Brand AI Visibility Score out of 100, blended from three components, each scored 0–100:
Recognition — 30%. Do the engines know the brand exists from training data alone?
Accuracy — 30%. When they describe it, how right are they? Every hallucination costs here.
Positioning — 40%. Does the brand surface on category and industry queries, or does the model reach for a competitor?
Positioning carries the most weight because it is the one that costs money. A model can know exactly who you are and still hand the buyer to someone else.
The finding that lands hardest
A heritage leather goods brand — decades of trading, a real catalogue, a name their customers know. Asked cold, Claude offered that it might be a whiskey company. Possibly a restaurant.
Not a wrong detail. No recognition at all.
That is what a score in the low 40s looks like from the inside, and it is why we start here rather than with content.
What a bad result looks like, and a good one
At the low end, the model hedges on who you are. It opens with "if you mean", mixes you up with a similarly named company, and answers for the wrong audience. The leather brand that got called a possible whiskey company is the low end.
At the high end, it names you straight away, describes what you sell the way you would describe it, and aims the answer at the buyer you actually sell to.
Why it decides everything downstream
There is no point optimising content for engines that do not know who you are. Identity comes first, and this is how we find out whether it exists.
Once we know whether that identity exists, the next question is structural: can the engines read the site at all? That is what the AI Readiness Audit checks, and it runs next.
What we found across 900 brands
We ran this at scale. The average score was 46 out of 100. 22% scored under 30. 53% scored under 50.
The number that surprised us: brands doing $20M+ averaged 46.1 against 46.0 for smaller brands. Size buys you nothing here. A twenty-year-old catalogue and a large marketing budget do not make a model understand what you sell.
The proprietary part
The batch version runs against hundreds of brands at once, which is what made that study possible — and what makes it useful in a sales conversation rather than as a novelty.
What it does not tell you
It reads what the models learned in training, not the live web, so anything you changed on your site recently will not show up yet.
It tells you how you are described, not whether you get cited when someone is actually choosing what to buy. That is what the Competitor Scorecard measures.
There is no backlink or index data in it. And it is a snapshot: the models update, and the score moves with them.
LLM Perception is the first of the nine tools in our method. See all nine, in the order we use them →
Powers
Identity Architecture — making sure the models know who you are before anything else is attempted.
We will run this on your brand and send you the results as a free AI Perception Report.
Book the call →
Frequently asked questions
Why does ChatGPT describe my brand well when I ask, but not when my customers do?
When you ask ChatGPT about your brand while logged into your own account, it has your history, your phrasing and your context, so you get the most flattering version of your brand you will ever see. Your buyers do not get that version. Firon's LLM Perception step removes that advantage. It queries four engines, ChatGPT, Claude, Gemini and Perplexity, about a brand from training data alone, with no web access and no help, then compares each answer against what the brand's own homepage says. The gap between the two is where a model has invented a feature, aged out a price, attached the wrong founder story or recommended a competitor instead.
How is the Brand AI Visibility Score calculated?
The Brand AI Visibility Score is a single score out of 100 produced by Firon's LLM Perception step. It blends three components, each scored from 0 to 100. Recognition, weighted at 30%, asks whether the engines know the brand exists from training data alone. Accuracy, also 30%, measures how right they are when they describe it, and every hallucination costs points. Positioning, weighted at 40%, asks whether the brand surfaces on category and industry queries or whether the model reaches for a competitor. Positioning carries the most weight because it is the one that costs money: a model can know exactly who you are and still hand the buyer to someone else.
Does a bigger brand get better recognition from AI models?
Not in Firon's data. When Firon ran LLM Perception across 900 brands, the average Brand AI Visibility Score was 46 out of 100. 22% of brands scored under 30 and 53% scored under 50. Brands doing $20M or more averaged 46.1, against 46.0 for smaller brands, so size bought nothing. A twenty-year-old catalogue and a large marketing budget do not make a model understand what you sell.
Should we fix our content first or check whether AI recognises our brand?
Check recognition first. There is no point optimising content for engines that do not know who you are, which is why Firon starts with LLM Perception rather than with content. Identity comes first, and LLM Perception is how Firon finds out whether it exists. Once that is measured, the next step is structural: the AI Readiness Audit checks whether the engines can read the site at all, across crawl permissions, JavaScript rendering, schema completeness, entity consistency and content structure. Grading existing content, planning new content and producing it all come later in the sequence.
Alex Jordan
Founder, Firon Marketing
15+ years scaling brands. A strategic partner for high-growth founders, focused on sustainable revenue models and long-term equity value.
