Two different events you can observe
An AI answer can name you (your brand appears in the text) or cite you (one of your URLs is attached as a source). You can have either without the other. These four states are what you can record without guessing at anything inside the model.
| State | What the reader sees |
|---|---|
| Named and cited | Your brand in the sentence, your URL in the sources |
| Cited, not named | Your URL in the sources, brand absent from the text |
| Named, not cited | Your brand in the text, no link to you |
| Neither | No brand in the text, no URL in the sources |
Possible explanations, which you cannot observe
Engines do not publish which pages were fetched for an answer, what the model already held about your brand, or how the two were weighed. These explanations are worth testing, not facts:
- Cited, not named: your page was used as evidence for a claim while the brand was not the subject of the sentence, or the model named a competitor from other sources and attached your URL for a detail.
- Named, not cited: the name came from training data or from a third-party page retrieved instead of yours. The answer alone cannot tell you which.
- Neither: your page was not fetched for that prompt, or was fetched and not used. The two look identical from outside.
- Search on or off changes all of this. With search off, every name is memory; with search on, the source list is the only evidence of what was fetched.
The second row breaks most reporting. Semrush's June 2026 study of 3,981 domain appearances across four engines found that 61.7% of citations linked the URL without saying the brand. If your report counts citations as wins, most of those wins did not name the brand.
Why one score hides the thing you need to know
Tools collapse named and cited into one visibility number. That number rises when your blog post is cited as background on a category question, and when your brand is recommended on a buying question. Leadership makes different decisions depending on which one moved. Keep two columns. If you must show one headline figure, show named on buying prompts, with cited beside it.
How to count it, by hand, this week
- Fix a prompt set of eight to sixteen questions your buyers type, in their words. Freeze the wording.
- Run each prompt in ChatGPT with search on, and in Perplexity. Paste the answer and source list into a sheet with the run metadata (model name as shown, date, locale, search on or off, logged in or not). Without those five fields the row cannot be compared with next week's.
- For each prompt and engine, mark three cells. Named (yes/no), cited (yes/no, which URL), competitor named (which).
- A week later, run the identical prompts again and mark the same cells.
- Report named on N of M prompts and cited on N of M prompts, per engine, with both dates and the metadata. Expect some rows to change with no action on your side.
| Prompt | Engine, model, search, locale | Dates | Named (wk 1 to wk 2) | Cited URL (wk 1 to wk 2) | Competitor named |
|---|---|---|---|---|---|
| tool to see why ChatGPT recommends a competitor | ChatGPT, model as shown, search on, en-IN | wk 1 date, wk 2 date | no, then yes | none, then /guides/why-chatgpt-names-a-competitor | two others, then one |
| how to measure brand mentions in ChatGPT | Perplexity, default model, search on, en-IN | wk 1 date, wk 2 date | no, then no | none, then none | same two |
Reading the result
- Cited went up, named did not. The page is probably used as evidence, not as the answer. Check whether the page's first paragraph answers this prompt or a neighbouring one.
- Named went up, cited did not. A third-party page or training data probably carries the name. Check which URLs are cited on that prompt, and whether you could be on them.
- Both went up on the same prompt. Record the date, exact prompt and metadata, and keep running it. One week of both is an observation, not a trend.
- Neither moved. Common in the first week after a change, and common when nothing you did mattered. Give it a second and third run before concluding anything.