What Counts as AI Brand Mentions? Named, Linked and Recommended

An AI brand mention is your brand's name appearing in the text of an answer from ChatGPT, Gemini or Google's AI Overviews. Some reports blur AI brand mentions, links, source lists and recommendations into one number, and tools such as Firemark and Advanced Web Ranking draw those lines in different places. The same ChatGPT answer can count once in one report and two or three times in another.

Does a name in the source list count in Firemark or Advanced Web Ranking? Is a recommendation from Gemini a mention? What do you call a ChatGPT answer that praises your rival and lists you last?

Settle the words first, and the numbers in Firemark or any other report get easier to trust.

Key takeaways

• A mention is the brand's name in an answer from ChatGPT, Gemini or Google's AI Overviews. A link, a source-list entry and a recommendation are separate facts, as five answers about Ledgerly show.
• Firemark's two verdicts, named and linked, give four outcomes. Recommended is a label you apply by reading ChatGPT or Gemini answers against a written test.
• Firemark records position and sentiment only once a brand is named, so they sit a layer below Named.
• Advanced Web Ranking and Firemark define a mention differently, so check the definition first.

What counts as AI brand mentions?

An AI brand mention counts when the brand's name appears in the text of the answer, and everything else is a separate fact about that answer. Firemark's method page spells out four of those facts in its own words.

Firemark's named and linked verdicts give four outcomes: named and linked, named, linked but not named, and absent. Firemark's plain definition of AI visibility lays out all four.

Recommended isn't on Firemark's list. Firemark's method page has no recommended verdict, so a recommendation is a judgment you make by reading the ChatGPT or Gemini answer, not something Firemark scores.

What does each outcome look like?

Each outcome reads differently on the page, so the clearest definition is five answers to one question about Ledgerly. The question, the answers and the brands Ledgerly, Brightbooks and Billhive are invented.

Say the tracked brand is Ledgerly and the question is "Which invoicing software suits a five-person agency?" Here are five answers, trimmed to the lines that matter.

Answer 1 has everything: Ledgerly is named, its site is in the sources and the text picks Ledgerly. Answer 3 names Ledgerly too, and it still hurts, because the text steers the reader to Brightbooks.

Answer 4 is the odd one. The sources hold a ledgerly.example page, so the answer is linked, but the text never says the name Ledgerly. Under Firemark's wording that is linked but not named, and a report that counts source lists as mentions would score answer 4 as one.

What is a source-list-only appearance?

A source-list-only appearance in Firemark is when a link to the brand's site, or to a profile or listing it has added, appears in the answer's sources and the answer text never says the name. Under Firemark's rules that answer is linked but not named. A source title that only contains the name, on a page that isn't yours, is neither named nor linked in Firemark.

ChatGPT and Google's AI features both show sources apart from the prose. OpenAI's help page says responses that use web search may include citations, and that its Sources view, when available, lists the cited sources and other relevant links. A page in ChatGPT's Sources list may not be one the answer cited.

Google surfaces links too. Its AI Overviews and AI Mode link to supporting pages, and Google Search Central says they can display a wider set of links than classic search. The Search Console Generative AI performance report reports impressions, the number of times links to your site were shown in one of those features, and Google says it had reached all websites worldwide as of August 31, 2026.

Search Console's report counts links shown. It can't tell you whether Google's answer text said your name. To log the sources column next to the name column by hand, see Firemark's guide to tracking brand mentions in ChatGPT.

What does recommended mean?

Recommended means the ChatGPT or Gemini answer text points the reader to the brand as the choice for their stated need, instead of only listing it among other names. That's a working definition for this article, not a Firemark verdict, so write the test down before you label anything.

A three-step test keeps the labels consistent across ChatGPT, Gemini and Google:

1. Check for the name: If the ChatGPT or Gemini answer text doesn't name Ledgerly, stop. Ledgerly can't be recommended by an answer that never says Ledgerly.
2. Look for a pick: Phrases like "I'd pick", "start with" or "best for" aimed at the buyer's stated need count as a pick. A bare list of Ledgerly, Brightbooks and Billhive doesn't.
3. Look for a steer away: A warning, or a stronger pick for Brightbooks, means Ledgerly is named but not recommended.

When a call is close, label it listed. Then hand the same ChatGPT answers to a second person and compare labels, because the disagreements show where the test is vague.

What do position and sentiment add?

Firemark's position and sentiment describe a mention that already exists. Position says which third of the answer holds the first mention, and sentiment says whether the wording reads positive, neutral or negative.

Both need a named brand: Firemark records sentiment only when you're named, and position points at the first mention. In the five answers above, Ledgerly comes first in answer 1 and only after Brightbooks in answer 3, and answer 3 is the one most likely to read as negative or mixed. Firemark records which third of the full answer holds the first mention, which trimmed excerpts can't show.

Firemark's changelog says sentiment shares appear once 20 or more answers are rated, and that below that you see counts. Each negative answer is one click from its text in Firemark, which is the right way to read them: the label is a reading of wording, so check the negative ones yourself. Read both across many ChatGPT and Gemini answers, because one answer's position is an anecdote.

Why do tools count mentions differently?

Tools draw the line between a mention, a link and a citation in different places, so a mention total from Firemark can't be compared with another tool's. The definition decides the number, whether it's Firemark's or Advanced Web Ranking's.

Advanced Web Ranking's help page, for example, counts plain text mentions, inline links and cited source URLs together as AI brand mentions in Google's AI Overviews and AI Mode. In a tool that counts all three, answers 1 to 4 above would each add to the total, provided the brand name is set up as a reference. Firemark counts named and linked apart.

The gap between being named and being cited in Google AI Mode or Gemini can be wide. In BuzzStream's study of 12,000 responses, collected through Xofu from Google AI Mode, Google AI Overview, Gemini and GPT-5.4-mini, 23.1% of brand mentions came with a citation of the brand's domain in the same response, a figure that was 39.0% on single-brand prompts and 7.2% on list and category prompts. That's one vendor's sample with its own definitions, so read the direction, not the decimals.

Counting rules move totals the way denominators do, as Firemark's guide to AI share of voice shows.

How do you count them?

Count named first, then linked, then label recommended, position and sentiment only on the named answers, and give every count its own base, in Firemark or a Google Sheet. The base for recommended is the 18 named answers, not all 40, as the sample below shows.

Here is an invented sample of 40 answers on one engine, say ChatGPT, one answer per prompt. The numbers are arithmetic, not data from a real brand or from Firemark.

Five of 18 named answers is 28%, and five of 40 answers is about 13%. Same five ChatGPT answers, two headlines, so say which base you used.

Eighteen is a small base to split three ways, which is why the table shows counts, and Firemark's five-step method walks through a full run and the rules for calling a change.

What does a mention not tell you?

A mention records what ChatGPT or Gemini said, not whether the answer was right or what the reader did next.

OpenAI's help page warns that search results and citations can be incomplete, outdated or incorrect, and says to open a cited source to check that it supports the answer. The method page says Firemark doesn't track accuracy or the clicks, visits or sales that follow an answer. Firemark asks each engine logged out, from a US location, in English. Google's AI Overviews help page gives the same warning, that AI Overviews can and will make mistakes.

Read your name questions in ChatGPT by hand. Firemark tells you where you appear, and only you know whether the sentence is true.

See how AI answers treat your brand

Named, linked and recommended are three readings of one answer, and a report from Firemark or Advanced Web Ranking is only as clear as the line it draws between them.

Firemark counts named and linked separately, records position and sentiment on named answers, and shows counts with their base. It measures what ChatGPT, Gemini and Google answers say and doesn't change what any engine shows. Open Firemark's demo to see a full report, or start with a small set of questions.