AI visibility tracking: how to check your brand in ChatGPT, Gemini and Perplexity

November 5, 2026

Type "best project management tool for a 20-person agency" into ChatGPT and you get a short list of five or six names with a line of reasoning behind each. That list is the new page one. If your brand isn't on it, the buyer never sees your website, your ads or your reviews.

AI visibility tracking is the work of measuring that list over time: how often an assistant mentions you, where in the answer you appear, how it describes you and which websites it relied on. This guide covers how to do it properly, by hand first and then with a tool.

What AI visibility actually means

Search rankings are a position on a page that looks roughly the same for everyone. AI answers are different in three ways that shape how you track them:

• The answer is generated each time. Ask the same question twice and the list of brands can change. A single check tells you very little.
• There is no fixed keyword. People ask long, specific questions ("CRM for a dental clinic that integrates with Google Calendar"). You track prompts, and you choose them.
• The answer has a source layer. Assistants with web search (ChatGPT search, Perplexity, Gemini, Google AI Overviews) pull from pages and cite them. Those citations explain a lot of what gets said about you.

So "AI visibility" is not one number. It's a set of measurements over a sample of prompts, assistants and runs.

How to check your brand visibility in ChatGPT manually

You can get a useful first picture in an afternoon with nothing but a spreadsheet.

1. Log out, or open a temporary chat. Memory and past conversations bias the answer toward brands you have talked about. A clean session is closer to what a new buyer sees.
2. Write 15–20 prompts a real buyer would type. Mix three kinds: category questions ("best email marketing tools for ecommerce"), problem questions ("how do I reduce cart abandonment emails going to spam") and comparison questions ("Klaviyo vs Mailchimp for a small Shopify store"). Avoid prompts that contain your brand name, except a few you keep separately to check how you are described.
3. Ask each prompt in each assistant you care about. At minimum ChatGPT, Gemini and Perplexity. If you sell to Google searchers, check AI Overviews for the same queries.
4. Record five things per answer: was your brand mentioned, its position among the brands listed, the tone (positive, neutral, negative), which competitors appeared and which URLs were cited.
5. Repeat the whole set at least twice more on different days. Then look at the rates, not the individual answers.

This takes about two to three hours per round for 20 prompts across three assistants. It's the right way to start because it forces you to read the answers. It stops being practical once you want weekly data, more than one country or more than a handful of competitors.

The four metrics worth reporting

Most AI visibility tools, ours included, report some version of these. They are simple to calculate from the spreadsheet above.

Two notes on reading them.

Visibility and share of voice move differently. If assistants start listing more brands per answer, your visibility can rise while your share of voice falls. Report both.

Position only counts where you appear. An average position of 1.8 with 12% visibility means "rarely mentioned, but first when it is". That's a different problem from 60% visibility at position 5.

Why one check is not a measurement

AI answers are sampled, not looked up. The same prompt, asked three times on the same day, can name you in one answer and leave you out of the next. Treat any single answer as one draw from a distribution.

Practical rules that hold up:

• Use at least 20 prompts per topic before you quote a visibility rate. With five prompts, one answer swings the number by 20 points.
• Compare runs, not answers. A change from 30% to 34% between two weekly runs is noise. A trend over four runs, or a jump of 15+ points, is worth investigating.
• Keep the prompt set fixed. If you add prompts, track the new ones as a separate group for a few runs. Otherwise your chart measures your editing, not the market.
• Fix the country. Answers with web search depend on location. A prompt asked from Germany can return a different shortlist than the same prompt asked from the US.

API answers vs what users see in the chat

There are two ways to collect answers at scale, and they don't return identical results.

API collection sends the prompt to the model's developer API, usually from a server in a data center, with the country written into the prompt ("answer as if I'm in Germany"). It's fast, cheap and repeatable, and it's what most trackers run on.

Web app collection asks the question on the assistant's own site (chatgpt.com, gemini.google.com, perplexity.ai, a Google results page), the way a buyer would. It's slower and costs more, but it captures what the consumer product actually shows.

The difference isn't cosmetic. The consumer apps have their own system prompts, their own search behaviour and their own source selection, and they localize by where the visitor connects from, not by what the prompt says. An API answer can name a different set of brands than the app does for the same question.

Location matters just as much. If the request comes from a US data center, "best bank for freelancers" gets US results no matter what the prompt says about Germany. To see what a German buyer sees, the question has to come from a German connection, ideally a residential one, because assistants and search engines treat data center traffic differently.

The most accurate setup therefore combines three things: the real web app, a residential connection in the target country, and a fresh logged-out session. Use the API with web search only for assistants that don't have a public web app.

Tracking competitors and AI share of voice

The prompt set you built for yourself is also your competitor tracker. For every answer, log all brands mentioned, not just yours. Within a few runs you will see three groups:

• Known competitors you already compete with in search.
• Discovered competitors, brands the assistants recommend that you haven't been tracking. These are often the most useful finding of a first report.
• Non-brands, generic advice or open-source options that take a slot in the list.

AI share of voice is your mentions divided by the mentions of every brand in the set. Watching it per assistant is useful: it's normal to lead in Perplexity and trail in ChatGPT, because they lean on different sources.

Sources: the part you can act on

When an assistant searches the web, its answer leans on the pages it found. Every citation is a clue about what to fix.

Group the cited URLs by domain and sort by how often each one appears in answers where competitors are mentioned and you are not. Typical patterns:

• Listicles and reviews ("best X tools in 2026") on third-party sites. If you're missing from the article, you're often missing from the answer.
• Reddit and community threads. Assistants quote them heavily for opinion-style questions.
• Your own pages, cited for factual questions about you (pricing, features, integrations). If outdated pages are cited, the answer will be outdated too.
• Competitor pages, sometimes cited even in comparison answers, which means they wrote the comparison you didn't.

This is where tracking turns into a to-do list. Our guide on how to rank in ChatGPT covers what to do with each type of source.

When to move from a spreadsheet to an AI visibility tracker

The manual method breaks down at roughly the same point for most teams:

• more than about 50 prompts, or more than three assistants
• more than one country or language
• a weekly or daily cadence
• the need to show the exact answer behind every number in a report

An AI visibility tracker automates the asking, the labelling and the history. When you evaluate one, check the following:

1. Which assistants are included in the base price and which are add-ons. Some tools charge extra for Gemini or Claude.
2. How answers are collected: the assistant's real web app or its API, and from which location. Ask whether the country is set by the connection or only written into the prompt.
3. How mentions and sentiment are labelled, and whether you can correct a wrong label.
4. Country and language control per report.
5. Pricing model. Per-prompt plans suit a stable set; per-check pricing suits teams that test many prompts and keep only the useful ones.
6. Export. You should be able to get every answer, label and source out as CSV.

We compare the main options, with current prices, in Peec AI alternatives: 8 AI visibility tools compared.

How AI Monitor does it

AI Monitor follows the method above. You add your brand, its spellings and your competitors, write or generate the prompts buyers ask, pick the assistants (Google AI Overviews, ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek, Mistral) and the country to ask from.

Wherever an assistant has a public web app, AI Monitor asks there, in a fresh logged-out session, through a residential connection in the report's country. It checks where the connection actually was and flags any answer that came from the wrong country, so you can re-ask it. Assistants without a public web app, such as Claude, are asked through their API with web search always on. Each run stores the full answers and every cited source. A separate model labels every brand mention with its position, tone and the sentence it was based on, and you can correct any label. Every number in the report links to the answers it was counted from.

There's no subscription needed to try it: new accounts get free credit and you pay per check, or move to a monthly plan once you know your prompt set. See pricing.

FAQ

What is an AI visibility tracker?

An AI visibility tracker is a tool that regularly asks AI assistants such as ChatGPT, Gemini and Perplexity a fixed set of prompts and records whether your brand is mentioned, in what position, with what sentiment and which sources were cited. It turns individual answers into trends you can report on.

How do I check if ChatGPT mentions my brand?

Open a temporary chat or log out, ask 15–20 questions a buyer in your category would ask (without your brand name in them), and note whether you appear, where, and how you're described. Repeat on two more days, because answers vary between runs.

How often should I track AI visibility?

Weekly is enough for most brands. Daily tracking helps during a launch, a PR crisis or right after you publish content aimed at AI answers, but it mostly adds noise for stable categories.

Is AI visibility the same as SEO rankings?

No. Rankings are a position on a page for a keyword. AI visibility is the rate at which you appear across generated answers to many prompts. They're connected, because assistants with web search cite pages that rank, but a brand can rank well and still be left out of AI answers.

How many prompts do I need?

Start with 20 per product line or topic. That's enough to see a stable rate after two or three runs. Large brands with several categories usually track 100–300 prompts.