How to Track Brand Mentions in ChatGPT, Gemini and Perplexity (Free Method, Then Tools)

To track brand mentions in ChatGPT, Gemini and Perplexity, ask each one your buyers' questions 3 times a week and log who gets named and which sources it shows. The repeats matter: in Ahrefs' 20-day study (Oct 29, 2026), around 60% of the websites ChatGPT cited one day were gone the next⁸.

A spreadsheet handles about 20 questions for free. Past that, a tracker does the asking, and the fair way to compare trackers is by answers a month.

Tracking is half the job. Getting named in the first place is the other half, covered in how to get your brand recommended by ChatGPT.

Write the prompts your buyers would type

Start with 10 to 20 questions a buyer types before they've picked a brand, in their words. Leave your brand name out. A question that names you only tests what the assistant knows about you, and the number you want is how often it brings you up unasked.

Mix four kinds: the category (“best mattress for back pain”), a comparison of two rivals, alternatives to a rival, and the problem before the product. Borrow the wording from places buyers already type it: your Search Console queries, sales calls, support tickets and the threads in your niche.

Keep questions about your own brand (“is Snoozly any good”, pricing, the return policy) on a separate tab. Ahrefs recommends tracking them apart from discovery questions, so a strong score on your own pricing doesn't hide a weak one for “best X”⁸. Then freeze the wording. Change one word and you're measuring a different question.

Run each prompt 3 times, in a clean chat

Your own account is the worst place to check. With Memory on, ChatGPT knows what you've told it about your company, and a Temporary chat alone doesn't fix that: OpenAI's help page says you pick Personalized or Unpersonalized before the first message, and only Unpersonalized skips your memories and custom instructions. For Gemini, Perplexity and Google's AI Mode, use a private window, signed out where the product lets you.

Then ask each question 3 times. The Ahrefs study followed 37,500 citations across 5 AI tools, and Gemini and Copilot churned hardest, with about 80% of their citations gone by the next day. About 75% of the lost citations came back within the 20 days, so Ahrefs suggests counting how many of the last 7 days you were cited instead of trusting one check⁸.

Same question, different sources

Record mentions, sources and position

Log one row per answer. The columns: assistant, prompt, run, whether it searched the web (it shows sources when it does), your brand (recommended, named or missing), its spot, who else got named, and the sources. Your number is the mention rate: answers that name you divided by answers, worked out per assistant, because each one leans on different sources.

Mentions and citations move separately

A mention is the assistant naming your brand. A citation is a link to a page as a source, which may be yours or a review that names you. Ahrefs looked at brands that were both named and cited on Google's AI tools and ChatGPT: the next day, 12 to 16% had lost the citation but kept the name, and 9 to 12% the reverse. Names were steadier overall, with about 1 in 5 brand names showing up on just one day against 50 to 70% of cited websites⁸.

So log name variants as one brand (Ahrefs grouped “HubSpot” and “HubSpot CRM” before counting), and keep answers that searched apart from those that didn't. An answer that didn't search cites nothing, so the brands it names come from training. In Ahrefs' sample of 66.6 million free, logged-out ChatGPT responses from August and September 2026, only 14.4% contained fan-out queries, the web searches it runs before answering⁹.

Count answers, not prompts

Trackers put prompts on the price tag. What you get is answers: prompts × assistants × runs × checks a month. Peec AI's pricing page counts it that way, 50 prompts × 3 models × 30 days for 4,500 answers on Starter², and Ahrefs counts Brand Radar prompts in checks, where 5 prompts on 1 platform, daily, is 150 a month⁵.

Answers also decide how far you can trust the number. If your brand shows up in half the answers, 10 answers can only tell you it's somewhere between about 20% and 80%. With 60 answers that narrows to 13 points either way, and with 400 to 5 (our arithmetic, a standard 95% range).

The default setup is 720 answers a month, typed by hand. (A tracker starts to look reasonable around here.) Peec AI's and Ahrefs' counts assume one run per prompt per check, which makes a single day noisy and a month solid, so read a tracker's chart as a weekly or monthly trend.

Trackers that do it for you

Six trackers, each read on its own pricing page on Nov 3, 2026. The answers column is our arithmetic from their published limits, assuming every prompt runs on every included engine, which is the reading most generous to them.

• Otterly.AI fits if you want the cheapest daily tracker and your buyers use ChatGPT, Perplexity and Google. Gemini, AI Mode and Claude are add-ons (Gemini is $9 a month on Lite), and Lite drops to $25 a month billed annually.
• Peec AI fits if you want the answer count printed on the plan. Its 6 self-serve engines include ChatGPT, Gemini and Google's AI answers. Perplexity and Claude sit on Enterprise.
• Semrush or Ahrefs fit if you already pay for that SEO suite. Ahrefs Lite includes 5 Brand Radar prompts at no extra cost.
• LLMrefs fits if you'd rather track keywords than write prompts and want every engine at one price, updated weekly (SlopMog vs LLMrefs).
• Profound fits an enterprise team that needs up to 9 engines, SSO and SOC 2. Trial prompts can't be edited (SlopMog vs Profound).

Before you buy, ask how the tool asks. Some trackers query each assistant's app, others its API with web search on, and a logged-in buyer with memory sees a third version. No method matches every customer's screen, so pick one and keep it. For a one-off look, Ahrefs' free AI Visibility Checker gives a snapshot of mentions and cited pages, top 5 results per section, with no signup⁶.

When you're not in the answer

A low mention rate is a reading, and the Sources column says what to change. Group a month of rows by the kind of source and work down the list.

Check which assistant each source came from before you spend a week on Reddit. ChatGPT cited Reddit in about 1 in 43 answers by late September 2026, down from 1 in 5 in March⁹, while Google's AI answers, Gemini and Perplexity barely moved in August¹⁰ (the engine-by-engine numbers). After each change, rerun the same prompt list, and wait for the mention rate to hold for several weeks before you credit the change.

Getting into the cited threads is the job SlopMog does. Its High-Value discovery writes 40 buyer-style questions without your brand name, asks each one to 4 AI models with live web search (Gemini, Claude, GPT and Grok), and ranks the Reddit and YouTube threads they cite. Those are API models, so the list can differ from a buyer's screen (more on what the 4 models cite). SlopMog then drafts a comment for each thread, and you approve it or switch the site to hands-free.

Test the Waters is $99 a month for 1,000 credits, about 166 regular comments at $0.60 or 16 High-Value ones at $6, with a 7-day refund on the first payment and no free trial. Make Waves ($249 a month) and up also track when AI chatbots mention you (every plan, side by side).