LLM visibility tracking means asking AI assistants such as ChatGPT, Claude, Gemini and Perplexity the questions your customers ask, recording whether your brand gets mentioned and which sources get cited, then repeating those checks over time. The part most people miss: an AI answer isn’t a ranking. The same prompt can name different brands on different runs, so you measure a mention rate across many prompts and runs, not a position. The SEO MCP’s AI Visibility tools return each assistant’s answer to a prompt from inside your own AI chat, at 12 to 18 credits per answer on Growth Engine. The checks run when you ask; you keep the record.
Last updated: November 2026
Why “rank” is the wrong word for AI answers
Everyone knows the idea by now: people ask assistants for recommendations, so you’d like to be in the answer. The trouble starts when people track it the way they track Google.
A Google position is fairly stable from one hour to the next. An AI answer isn’t. Ask the same assistant the same question twice and the wording can change, and sometimes so do the brands it names. One screenshot of ChatGPT recommending you proves that it happened once. That’s all.
Most people don’t realize how much that variation matters until they try to repeat a result. So treat every response as one sample.
Local SEO already works this way. A geo-grid scan checks your Google Maps rank from up to 49 points around a location, because the result from one spot isn’t the whole picture. AI visibility needs the same many-samples thinking, only the samples are prompts and repeat runs instead of map coordinates.
The four things worth recording
For each prompt and each assistant, write down:
- Mentioned or not. Across all your samples, this becomes your mention rate, the share of answers that name you.
- Where you appear. First recommendation, one of several, or a footnote near the end.
- Sources cited. Which URLs the assistant points to, and whether any are yours.
- Who shows up instead. The competitors named when you’re missing are your real rivals in AI answers, and they’re not always the ones you rank against in Google.
With those four numbers you could finally tell a real gain from a lucky run. Number three is the one I’d watch most closely. A cited page is something you can act on: you can earn a mention on it, or publish a better answer to the question it covers.
Build a prompt set that reflects real demand
Most tracking goes wrong before the first check runs, because the prompts are wrong.
Start from customer questions, not brand questions. Asking “is Acme Bookkeeping any good?” mostly tests whether the model has heard of you. Asking “what’s the best bookkeeping service for a small bakery?” tests whether you win the recommendation, which is the thing you actually care about.
Then write two or three phrasings of each question. People ask the same thing in different words, and assistants answer those differently.
To pick which topics deserve a slot, check demand first. The SEO MCP’s ai_keyword_data_search_volume tool estimates how often people ask AI tools (ChatGPT, Gemini and Perplexity) about a keyword, with a 12-month trend. It costs 29 credits per started block of 100 keywords at Growth Engine rates, so a list of up to 100 candidate keywords costs 29 credits.
Running the checks with The SEO MCP
The AI Visibility category on our tools page has 8 tools. The ones that do the day-to-day work, at Growth Engine rates:
| Tool | What it returns | Credits |
|---|---|---|
llm_responses_chatgpt |
ChatGPT’s answer to a prompt | 18 |
llm_responses_claude |
Claude’s answer to a prompt | 18 |
llm_responses_gemini |
Gemini’s answer to a prompt | 17 |
llm_responses_perplexity |
Perplexity’s answer, with cited sources | 12 |
llm_scraper_chatgpt |
ChatGPT search results for a keyword, with cited sources and brand mentions | 6 |
Two more, which count keyword or domain mentions across many LLM answers, are available on request.
You don’t call these by name. You’d type something like:
“Ask ChatGPT, Claude, Gemini and Perplexity: ‘what’s the best bookkeeping service for a small bakery?’ Run each one twice. Make a table of which businesses each answer mentions, in what order, and list the sources Perplexity cites.”
What a round costs
One prompt across all four assistants is 65 credits at Growth Engine rates (18 + 18 + 17 + 12). From there:
- 20 prompts, one run each: 1,300 credits
- The same 20 prompts, two runs each: 2,600 credits
- Two runs a week for a month: 10,400 credits
Growth Engine includes 200,000 credits a month. Local Dominator, the $99 plan, uses fewer credits per call than these figures, and Full Stack uses more but each credit costs less. Our pricing page lists all four plans.
The honest catch: you keep the record
Nothing in the catalog is a scheduler. Every check is a live call that runs when you ask and returns fresh results at that moment. To see change over time, save each round’s table and paste last week’s back in with “what changed?”
You can wire the server into n8n, Make or a custom agent, but plan around sign-in. Access tokens last 24 hours and there are no refresh tokens, so someone signs in again about once a day. It suits checks done alongside a person, not jobs left running unattended.
If you’d rather not keep the record yourself, look at dedicated trackers and check how each one runs its checks. Local Falcon, for one, lists AI visibility tracking alongside its local rank grids. The SEO MCP’s case is different: the checks sit on the same credit balance as keyword, SERP and page audit tools, so the next question (“why does Perplexity cite that page and not mine?”) is answerable in the same chat.
What to do with the numbers
My rule: ignore a single week’s swing. Act on a pattern that holds across two or three rounds.
When a pattern does hold, start with the cited sources. If assistants keep quoting a page you don’t own, look at what that page covers that yours doesn’t. The on_page_content_parsing tool (44 credits) breaks down a page’s structure, and comparing it with your own is a quick way to look for the gap. Our Perplexity SEO guide goes further on earning citations in that one assistant.
If you’re missing from ChatGPT in particular, our ChatGPT SEO MCP post shows how to run these same checks from inside ChatGPT itself, which is oddly satisfying.
One limit to accept up front. No tool can tell you exactly why a model picked one brand over another. You can see what it said and what it cited. The reasons stay inside the model.
Frequently asked questions
What is LLM visibility tracking?
LLM visibility tracking is checking, over time, whether AI assistants such as ChatGPT, Claude, Gemini and Perplexity mention your brand when someone asks a question you’d want to win. Because answers vary between runs, you measure how often you’re mentioned across a set of prompts and repeat runs, plus which sources get cited, rather than a single position.
Does The SEO MCP track my brand in ChatGPT automatically?
No. The SEO MCP’s AI Visibility tools run when you ask, and nothing in its catalog is a scheduler. Each call returns an assistant’s answer at that moment. To track change, save each round of results and ask your assistant to compare them with the last round. Two tools that count mentions across many LLM answers are available on request.
Which AI assistants can The SEO MCP check?
Its tools return answers to a prompt from ChatGPT, Claude, Gemini and Perplexity, with Perplexity’s cited sources included. A separate tool returns ChatGPT search results for a keyword, with cited sources and brand mentions. Another estimates how often people ask ChatGPT, Gemini and Perplexity about a keyword, with a 12-month trend. All are listed on the tools page.
How many prompts should I track?
Start small. Fifteen to twenty-five customer questions, each written two or three ways, is my rule of thumb, not an industry standard. That’s enough to spot a pattern without drowning in tables. Add prompts when a new product, service or competitor appears, and drop the ones that never change. Run each prompt more than once before trusting the result.