AI rank tracker: tracking your brand in ChatGPT, Gemini, Perplexity

An AI rank tracker puts a fixed list of questions to ChatGPT, Gemini and Perplexity, along with Google's AI Overviews and AI Mode, on a schedule, then records whether your brand is named, in what position, and next to which competitors. Because those answers shift from one session to the next, a rank in AI search is an average over many answers, reported as citation rate, average position and share of voice. Vendors sell the same job under other names too: AI search tracker, AI visibility tracker, LLM tracker.

You can see the gap with classic SEO on Google's own results page. When we checked the US results for “ai visibility tools” on October 22, 2026, the page opened with an AI Overview citing six sources, and none of them was among the first nine organic results. A rank tracker built for blue links reports none of that. Hypervisible, the AI rank tracker we publish, was made for that gap: it reads ChatGPT, Gemini and Perplexity along with both of Google's AI engines on the same questions every Monday. The other tools that track it are compared in our guide to the best AI visibility tools.

Why classic rank tracking misses AI answers

Classic rank tracking reports where a URL sits in a list of links. AI answers are paragraphs that name brands, sometimes with a link and sometimes without, and a ranking report has no column for that.

Google's own reporting shows the problem. Search Console's help page says that “an AI Overview occupies a single position in search results,” and all the links inside share it. The last link in the overview is credited with the same position as the first. Nothing tells you whether the text recommended your company or mentioned it in passing.

People's habits have moved as well. Pew Research Center logged what 900 US adults did on Google in March 2025, and clicks on traditional results were almost half as frequent on pages that carried an AI summary: 8% of visits, against 15% without one. When the summary answers the question, a share of searchers stops there, and what it says about you counts as much as your rank below it.

Then there are the assistants with no results page at all. ChatGPT, Gemini and Perplexity give one answer per question. There is no position 11 to climb from, only an answer that names you or leaves you out. A tracker that reads Google's blue links never puts a question to them, which is why several SEO suites now sell AI modules on top.

What a rank means inside an AI answer

Inside an AI answer, your rank is your place among the brands the answer names, in order of first appearance. If ChatGPT lists five CRM tools for a ten-person real estate team in Austin and yours comes third, your rank in that answer is 3. If you're not named, you have no rank at all.

Numbered lists make this easy to read. In a paragraph, count the order in which brands first appear. Keep brand mentions and linked sources apart, because AI engines mix them freely: an answer can recommend you on the strength of someone else's review page, or link to your site without recommending you.

The bigger difference with Google is that the rank moves. Part of that comes from how the models generate text. OpenAI's own cookbook says its Chat Completions API is “non-deterministic by default,” so outputs may differ from one request to the next. In September 2025, Thinking Machines Lab had an open model, Qwen3, answer the same prompt 1,000 times with temperature set to 0, the most predictable setting, and got 80 different completions. The lab traced most of the gap to server load at the moment of the request.

Your own history plays a part too. Google's AI Mode can draw on your previous searches if you're 18 or older and your search history is saved. With Memory turned on, ChatGPT uses what it has kept from your chats to personalize its responses.

And the models themselves differ and change. Google says AI Mode and AI Overviews “may use different models and techniques, so the set of responses and links they show will vary.” On May 19, 2026, at Google I/O, Google made Gemini 3.5 Flash the default model in AI Mode for all users. Same questions, new model behind the answers.

So one reading is an anecdote. A rank only means something as an average over many answers, followed for weeks. For the ChatGPT side of the story, see how ChatGPT chooses its sources.

Citation rate, average position, share of voice

Most AI rank tracking comes down to three numbers. Citation rate tells you how often you're named, and average position how early. Share of voice measures how much of the answer space you hold against competitors.

Read the first two together. An average position of 1.5 with a 5% citation rate means the AI puts you first on the rare occasions it names you. A position of 3 with a 60% rate describes a brand that shows up in most answers without leading them. Start with citation rate, because an absence doesn't count in an average of ranks.

Share of voice brings in the competition. Take two tracked brands: if your citation rate rises from 30% to 35% while your main competitor's climbs from 30% to 50%, your share of voice drops from 50% to 41%. You improved and still lost ground. AI share of voice has the full formula, with the variants some tools use.

Two more columns earn their place in any report: the sites the AI cites as sources, which show where your brand should be present, and the tone of what it says about you. For the definitions behind each number, and what counts as a decent score, see our explainer on AI visibility.

How many prompts to track, and how often

In our view, 20 to 50 fixed prompts per brand, checked weekly, is a sensible start; go daily only when you need to react within a day. The right number depends mostly on how many products and markets you sell in.

Volume matters because answers vary, and what you should count is answers: 25 prompts run on 4 engines give 100 per reading. On 20 answers, one mention more or less moves your citation rate by 5 points; on 100 answers, by 1 point. With too few answers, you'll read a different draw as a trend.

Build the list from the questions buyers ask before they buy, in their own words. Mix two kinds. Prompts that don't name you, like “best accounting firm in Denver for a small construction company,” show whether the AI recommends you when someone is shopping. Prompts that do, your brand plus “reviews” or “pricing,” show what people hear when they look you up. Track both. New customers come mostly through the first kind.

Once the list is written, stop editing it. A reworded prompt produces a different answer, and week-over-week comparisons lose their meaning. Add new prompts as a separate group instead of editing the old ones. Run them in the market you sell in, too: a brand named often in US answers may barely appear when the same question is asked from the UK.

For cadence, start from what you'll do with the numbers. Daily readings catch a change fast, which helps during a launch or a product recall. A weekly reading on unchanged prompts shows the trend clearly enough for a monthly report. Monthly readings leave long gaps, and a model change can move answers in between.

If you check by hand, cut personalization out. In ChatGPT, start a temporary chat and pick Unpersonalized, so it ignores your memories and custom instructions. In Google, sign out or pause your search history. For Google's AI summaries specifically, our AI Overviews tracker page covers what to record and how often.

Google rankings and AI answers in one report

Keep one row per topic, with the Google rank of the keyword next to your presence in AI answers to the matching question. No free tool fills every column.

Search Console covers the Google side. Its Performance report gives your pages' average position, and Google counts AI Overviews and AI Mode traffic in the “Web” search type, mixed in with the rest. The newer generative AI performance report adds impressions in AI responses, broken down by page and by country. It tracks URLs; your brand name and your competitors stay outside its view.

Several SEO suites have bolted AI tracking onto their rank trackers. SE Ranking, for one, sells a separate product called SE Visible: its Basic plan checks 75 prompts daily on ChatGPT, Gemini, AI Mode and AI Overviews for €80 a month, without Perplexity (read October 29, 2026). If your team already works in a suite every day, its AI module is the shortest route. Our roundup of AI SEO tools looks at them from the SEO side, and the AI toolkit of a large SEO suite gets a review of its own.

Then add a column for the gap between the two. A keyword where you rank first on Google but never appear in the AI Overview usually calls for a clearer page. A question where the AI names you while you don't rank often points to a third-party site doing the work for you.

Choosing an AI rank tracker

Pick the AI rank tracker that covers the engines your customers use, at the cadence you need, for a price you'll still be paying in a year. Six checks sort a shortlist:

• Engines in the price. A logo on a homepage can turn out to be an add-on, or missing from the plan you'd buy. Read that plan's engine list: SE Visible's, for example, leaves out Perplexity.
• Cadence. Most tools check daily, a few let you choose, and Hypervisible reads weekly.
• Real volume. Vendors count prompts, credits or checks. Ask each one how many answers you'll get per month.
• Country and language. If you sell in the US and the UK, get both markets in writing.
• History and export. A tracker is worth what its history is worth. Check how many months it keeps and whether the data can go into your own report.
• A trial or a free check. Judge on your own prompts before you pay.

Agencies also have to settle what the client sees each month, which is the subject of AI visibility reporting for agencies.

Hypervisible on the checklist

Hypervisible publishes this guide, so here is how our own tracker fares on the checks above.

Engines in the price: all five, ChatGPT, Perplexity and Gemini plus AI Overviews and AI Mode on Google, from the entry plan up. Nothing is sold as an extra, and that includes the two Google engines that sit above the blue links your old tracker reads.

Country and language: four markets are measured, the US, the UK, France and Germany, each with questions in its own language. A brand selling in Boston and in Leeds gets two readings, kept apart.

What you read: on a paid plan, the same questions run again each Monday, and every number shows how many points it moved since the last reading. You get citation rate and average position as defined above, where you place among your competitors when the same answers are compared, the sites the engines cite, and how they describe you, tone included. Plans are on the pricing page.

Who makes it: NEXUS GEO, a French agency with GEO as its sole business. The tool runs on data from Google and the AI engines.

Cadence is where you should weigh the trade-off. The reading is weekly, so if you need to see a change the day after you publish, a daily tracker will serve you better. Claude and Copilot aren't part of the reading either.

The sixth check costs nothing. The free AI visibility checker sends ten questions to those five engines, five on your industry without your name and five with it. You'll have the 50 answers read within 5 to 10 minutes and saved in your account, with no card number asked.

Frequently asked questions

Sources

• What are impressions, position, and clicks?, Search Console Help, read October 22, 2026
• AI features and your website, Google Search Central, updated December 10, 2025, read October 22, 2026
• New opportunities, control and insights for website owners, Google blog, June 3, 2026, updated August 31, 2026, read October 22, 2026
• Get AI-powered responses with AI Mode in Google Search, Google Search Help, read October 22, 2026
• Google Search's I/O 2026 updates, Google blog, May 19, 2026, read October 22, 2026
• Memory in ChatGPT, OpenAI Help Center, read October 22, 2026
• How to make your completions outputs reproducible with the new seed parameter, OpenAI Cookbook, November 2023, read October 22, 2026
• Defeating Nondeterminism in LLM Inference, Thinking Machines Lab, September 10, 2025, read October 22, 2026
• Google users are less likely to click on links when an AI summary appears in the results, Pew Research Center, July 22, 2025, read October 22, 2026
• SE Visible plans, SE Ranking, read October 22 and October 29, 2026

Read next