Most teams treat AI search visibility as a single yes-or-no question: did the chatbot say our name? That answer is too coarse to act on. A brand can appear in an answer and still lose the deal, because appearing is only the first of three separate measurements.
The three measurements
Mentioned. Did the answer name your brand at all? This is the baseline. It is binary per answer, and it is the weakest form of visibility because a passing reference carries little weight.
Position. Where in the list did you appear? Answers usually present options in an order, and readers anchor on what comes first. A brand in the first slot is read differently from a brand in the sixth, even when both are technically mentioned.
Cited source. Did the answer point to your content as the source it drew from? This is separate from being named. An answer can name your brand because it read a review site, a comparison post, or a competitor's page. Being the cited source means the model treated your page as evidence.
These three move independently. You can be mentioned without being cited. You can be cited without being named in the prose. You can be named in the last position after four competitors. Each combination calls for a different fix, which is why collapsing them into one score hides the useful part.
Why the distinction matters
A mention tells you the model knows you exist. Position tells you how the model ranks you against alternatives it also knows about. Citation tells you whether your own pages are doing the persuading.
If you are consistently mentioned but never cited, your problem is content, not awareness. The model has learned your name from somewhere else — directories, press, third-party listicles — but does not treat your site as a source worth quoting. That points at page structure, specificity, and whether your pages actually answer the question being asked.
If you are cited but ranked last, your problem is positioning. The model is reading you and still putting others first. That often comes down to how clearly your page states what you are best for, and whether third-party sources agree.
If you are neither mentioned nor cited, the question is whether the model can reach and parse your pages at all. Crawler access comes before content quality. A blocked crawler makes everything downstream irrelevant.
A worked example
Suppose you sell project management software and you ask an assistant a buyer-style question: "What are the best project management tools for a 20-person agency?"
The answer comes back as a short list with a sentence or two on each, followed by links.
Read it once and write down three things.
Step 1: Mark every brand named. Say the answer names six tools. Your brand is one of them. Mentioned: yes, 1 of 6.
Step 2: Record the order. Your brand appears fourth. Position: 4 of 6. Note whether the first three get longer descriptions — often they do, which widens the gap beyond raw ordering.
Step 3: Check the citations. The answer links to five sources. Maybe two are third-party review sites, one is a competitor's pricing page, one is a Reddit thread, and one is a help article from a vendor. Your domain is not among them. Cited source: no.
Now you have a profile instead of a verdict. You are mentioned, mid-list, uncited. The action is not "get mentioned more." It is "become the source." That might mean publishing a page that directly answers the agency-size question with the specifics a model can lift — team size thresholds, pricing at that scale, what breaks first. It might mean earning inclusion in the third-party pages the model already trusts.
Run the same three steps on a second question, such as "project management software for agencies with client billing," and compare. If you are cited on one and not the other, you have found a content gap rather than a visibility problem. If you are absent from both, start with crawler access.
Doing this by hand is fine at first
You do not need a platform to get value from this. Pick five buyer questions, ask each one in the assistants your buyers actually use, and score the three measurements in a spreadsheet. Five questions across three assistants is fifteen rows. That is an afternoon, and it will tell you more than a single aggregate score.
Two cautions. First, answers vary between runs and between users, so treat any single answer as a sample, not a fixed rank. Second, the same question asked in different assistants can produce very different lists, because each one retrieves and weights sources differently.
If you want a faster first read before building the spreadsheet, our free AI visibility check generates three buyer questions for your brand and category, asks Perplexity with web search, and shows whether your brand is named, which competitors are named, and which sources are cited. It gives you three checks per day and does not track results over time, so it is a starting snapshot rather than a monitoring tool.
What you can and cannot control
You cannot make an assistant mention you. No vendor documents a way to buy or guarantee inclusion in AI answers, and anyone promising a specific placement is guessing.
What you can control is narrower and more concrete:
- Crawler access. Each assistant uses different bots for different jobs. OpenAI's OAI-SearchBot surfaces sites in ChatGPT search, and sites opted out of it are not shown in ChatGPT search answers. Anthropic's Claude-SearchBot indexes content for search, while Claude-User fetches pages when a user asks a question. PerplexityBot surfaces and links sites in Perplexity results. If you block the wrong bot, you remove yourself from one surface while believing you have only opted out of training.
- Whether your pages are quotable. Clear claims, specific numbers, defined audiences, and direct answers to real questions give a model something to cite.
- Whether third parties describe you consistently. Models draw on sources beyond your site. If review sites and communities describe you the same way, that description is more likely to survive into an answer.
Note that robots.txt expresses intent rather than enforcing access, and changes can take about 24 hours to propagate with some vendors. Treat it as a signal, not a lock.
Start with one question
Pick the single question your best customer asks before they find you. Ask it in one assistant, then score the answer on the three measurements: mentioned, position, cited. Write down the competitors named and the sources linked. That one row tells you whether your next move is access, content, or positioning — and it takes about five minutes.