I ran a client's site through the ChatGPT API and asked it to describe each page. It told me the brand made appliances.
It doesn't. It's a premium direct-to-consumer retail brand, and it was doing well in Google at the time, with organic revenue up 149% year over year. Rankings just couldn't tell me how AI understood it.
So now I test for it. Here's how, and what to do with what you find.
Why it matters if AI misreads your brand
When someone asks an AI assistant a question, it often breaks that question into smaller searches, pulls sources and writes an answer. That's query fan-out. If the model can't connect your name to what you sell, you're a harder brand to recommend.
AI traffic is still small for most sites, but I wouldn't wait on this. On another account, leads from AI assistants converted about 1.4x better than organic search.
How I ran the test
- Crawl the site in Screaming Frog and include the pages that earn revenue.
- Connect the ChatGPT API in Screaming Frog.
- For each page, send only its name or main heading.
- Ask what the page is about, and what searches the model would run to answer questions about it.
- Export everything to a spreadsheet and read the answers yourself.
Here's a simple prompt to start with:
Here is a page title: [H1]
In 2 to 3 sentences, explain what this is.
Then list 5 searches you would run to learn more about it.I kept the context thin on purpose. A full page gives the model plenty of clues. I wanted to know whether the name could stand on its own, because a brand name has to do a lot of work on its own.
My run covered 173 pages and 860 fan-out queries. Only 28 of the 173 answers mentioned the brand by name.
What came back
| Page | How ChatGPT read it |
|---|---|
| Brand homepage | An appliance manufacturer |
| Brand story page | A farming initiative |
| Core product | A vehicle accessory |
| Signature product | A product from a competitor |
Each of those guesses makes sense once you see it. The brand and product names overlap with other things in the world, and nothing in the name pointed the model back to this company.
What to look for in your own results
- Wrong industry. The model thinks you're in a different business.
- Competitor credit. It connects your product name to another brand.
- Generic answers. It describes the category and never mentions you.
- Stale queries. Fan-outs include old years, a sign you're looking at older training data.
How to fix what it gets wrong
- Pair distinctive product names with plain descriptions in titles, H1s and the first lines of copy.
- Say exactly what you sell in the first sentence of your homepage and About page.
- Use the same brand and product names everywhere.
- Add Organization and Product schema so the relationships are explicit.
You probably can't fix every page at once. I rank pages by three things: ranking upside, commercial value and AI risk. You can see how that worked in my e-commerce case study.
What this test can't tell you
This shows how clearly your names communicate with very little context. It doesn't show what a shopper sees in ChatGPT with web search turned on, and answers can change from one run to the next.
Check the sources, too. My export listed example.com as a citation, which is a placeholder address. Open every URL before you put one in a report.
Where to start
Pick your ten most important pages and run them this week. If the model can't tell what you sell, that's a useful thing to find out.
If you try it, I'd love to hear what you find.