Brand AI Mentions

Element Code: TE-013

What counts as a brand AI mention

A brand AI mention is any time a generative assistant names your brand inside an answer: a shortlist for "best invoicing software for freelancers", a comparison against a competitor, a how-to answer that recommends your product as the tool for the job, or a direct "is this company legit" reply. In practice you will see three flavors, and they are worth tracking separately:

• Unlinked mention: the model names you but cites nothing. This usually comes from training data, which means it is slow to change and can be years stale.
• Cited mention: the assistant names you and links one of your own URLs as a source. This is the strongest outcome, because it comes from live retrieval you can influence.
• Third party mention: the assistant names you because a listicle, review site, or Reddit thread it retrieved names you. You control this indirectly, by being present on the sources the model trusts.

The distinction between training memory and live retrieval matters more than most guides admit. When ChatGPT answers with search enabled, or Perplexity answers anything, the brand list is being assembled from retrieved documents at answer time. That is a surface you can actually work on this quarter. A mention baked into model weights is not.

Why this matters

The uncomfortable part of assistant answers is that there is no SERP to screenshot. When a buyer asks Google for "best CRM for a two person agency", you can at least see who ranks. When they ask an assistant, the shortlist is generated privately, per user, per session, and if your brand is not in it you were never in the running. No impression data, no ranking report, nothing in Search Console. The only way to know is to ask the assistants yourself, systematically.

Mentions also compound. Assistants lean heavily on a small set of citation sources per topic, those sources cross reference each other, and the brands already present tend to stay present. The earlier you find out you are missing from that loop, the cheaper it is to fix.

One more thing that trips people up: answers are non deterministic. Run the same prompt five times and the shortlist shuffles. A single spot check tells you almost nothing. You need sampling and a mention rate, not a screenshot from one lucky run.

How to measure it, step by step

1. Build a prompt set from real buyer language. Pull 30 to 100 questions from sales calls, support tickets, People Also Ask, and your paid search queries. Cover four patterns: category prompts ("best X for Y"), comparisons ("us vs competitor"), problem prompts ("how do I fix Z"), and brand prompts ("is BRAND worth it").
2. Run each prompt across the assistants your market uses. ChatGPT with search enabled, Perplexity, Gemini, Claude, and Copilot is a sane default set. Run each prompt several times, because single runs lie.
3. Log the answer, not just yes or no. Record: mentioned or not, position in the shortlist, which URL got cited, how the brand was described, and every competitor named. The competitor column is where the strategy comes from.
4. Automate once the manual baseline proves the point. Semrush's AI toolkit, Ahrefs Brand Radar, Profound, Peec AI, and Otterly.AI all track assistant mentions on a schedule. If you want raw data in your own dashboards, DataForSEO exposes LLM mention endpoints you can wire into Looker Studio or a spreadsheet.
5. Compute share of voice. Your mentions divided by all brand mentions across the prompt set, per model. That single number, tracked monthly, is the KPI.

How to move the number

You cannot edit a model. You edit what it reads. The playbook that actually works starts with the citation log from step three: for every category prompt where you were absent, look at which pages the assistant did cite. Nine times out of ten it is a handful of listicles, review platforms, and community threads. Getting included on those specific pages moves your mention rate faster than anything you publish on your own domain.

Then fix the owned side. Keep your entity clean and consistent: one canonical brand name everywhere, Organization schema with sameAs links, an About page that states plainly what you do, for whom, at what price. Publish comparison and alternatives pages that answer the exact prompt patterns buyers use, because when retrieval runs, a page titled the way the question was asked is an easy chunk to grab. And check your robots.txt: if you blanket blocked AI crawlers in 2023 and forgot, you are opting out of the cited mention category entirely. I have seen that exact self-inflicted wound on more than one audit, and it is a quietly expensive one.

What good looks like

You appear in the majority of category prompts where you genuinely belong, your description is accurate and current across models, at least some mentions cite your own domain rather than only third parties, and you have a monthly trend line you can put next to organic traffic in reporting. You also know exactly which prompts you lose, to whom, and which source pages caused it. That last part turns a vanity metric into a work queue.

FAQ