How to Track Brand Mentions in AI Answers (ChatGPT, Claude, Perplexity) in 2026
The manual method first: a prompt-set spreadsheet you can start today. Then the tools, with verified pricing, for the day the spreadsheet stops scaling.
Contents (8)
Pick 20 to 30 questions your buyers actually ask, run each one in ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews on the same day every week, and log four fields per answer: mentioned, cited, position, competitors named. A spreadsheet handles one brand at small scale; dedicated tools automate it from about $17 to $29 per month.
Buyers now ask ChatGPT, Claude and Perplexity the questions they used to type into Google. Gartner predicted in February 2024 that traditional search engine volume would drop 25% by 2026 as AI assistants absorb those queries (Gartner, Press Release 2024). For what has actually happened since, see our 2026 AI search statistics. The only way to know whether those assistants name you is to measure it, repeatedly.
Why a single check misleads you
Ask an AI engine the same question twice and you can get two different answers, with different brands and different sources. That is not a bug in your test; it is how these systems work. One check tells you almost nothing. A trend across repeated, scheduled checks tells you everything.
The cause is documented: LLM inference is nondeterministic in production. Research from Thinking Machines Lab shows that even at temperature zero, server load and batch-size variation alone produce different outputs for identical prompts (Thinking Machines Lab, Defeating Nondeterminism in LLM Inference 2025).
The engines themselves also move constantly. Semrush tracked over 10 million keywords through 2025 and watched AI Overviews' presence climb from 6.49% of tracked queries in January to roughly 25% in July, then fall back to 15.69% by November (Semrush, AI Overviews Study 2025). A brand that "lost" its AI visibility in August may simply be caught in one of these swings.
How to track brand mentions in AI answers by hand (the DIY method)
The manual method needs a spreadsheet, one hour per week, and discipline. Run it for four weeks and you will know your mention rate, your citation rate, and which competitors AI engines prefer, before spending anything on tooling.
Step 1: Build a fixed prompt set
Write 20 to 30 questions your ideal buyer would ask an AI assistant. Not keywords, questions: "best invoicing software for freelancers in Germany", "[competitor] alternatives". Freeze the list. If the questions change every week, the trend line means nothing.
Step 2: Pick your engines
Cover the assistants your buyers use: ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews (search your question on Google and record what the AI box says). Google's box has its own sourcing rules, covered in our AI Overviews guide. Use fresh chats and a clean browser profile to reduce personalization.
Step 3: Run on a schedule, and sample
Same day, every week. Because answers vary run to run, ask each question 2 to 3 times per engine and record each run as its own row. Sampling is what turns noise into a measurement.
Step 4: Record the same fields every time
This is the sheet. Nine columns, one row per answer:
| Column | What to record | Why it matters |
|---|---|---|
| Date | Run date | Enables the trend line |
| Engine | ChatGPT / Claude / Perplexity / Gemini / AI Overviews | Visibility differs sharply per engine |
| Question | The exact prompt, verbatim | Keeps runs comparable |
| Mentioned? | Y/N: your brand named in the answer text | Your headline metric: mention rate |
| Cited? | Y/N: your domain linked as a source | Citations drive clicks and trust |
| Position | 1st, 2nd, 3rd... brand named, or absent | First-named brands win the recommendation |
| Competitors named | Every rival brand in the answer | Feeds share of voice |
| Sources cited | Domains the engine linked | Shows which pages to beat or get on |
| Notes | Sentiment, wrong facts, quoted phrases | Catches misinformation early |
Step 5: Score it monthly
At month end, compute four numbers per engine: mention rate (answers naming you ÷ total answers), citation rate (answers linking you ÷ total), share of voice (your mentions ÷ all brand mentions), and the delta versus last month. Those four numbers are the report. If you then want the pages that close the gaps, our playbook on how to get cited by ChatGPT covers the content side.
Engine differences that change what you measure
The five engines are not five copies of the same thing. Each retrieves and cites differently, which changes both how you check and how you improve. Treat each engine as its own channel with its own scoreboard.
| Engine | Retrieval behavior | What to watch when tracking |
|---|---|---|
| ChatGPT | Web answers lean heavily on Bing's index | Your Bing rankings; citations often mirror Bing top 10 |
| Perplexity | Search-first, numbered citations on nearly every answer | Citation rate matters as much as mention rate; freshness weighs heavily |
| Google AI Overviews | AI summary above classic results, presence fluctuates by query type | Whether the box appears at all for your questions, then who it links |
| Claude | Optional web search; often answers from model knowledge | Mentions without citations are common; record both separately |
| Gemini | Google-grounded answers | Overlap with your Google rankings, which does not guarantee an AI mention |
The ChatGPT row is the best-studied: Seer Interactive matched 500+ SearchGPT citations against search rankings and found 87% matched Bing's top organic results, against 56% for Google (Seer Interactive, SearchGPT Citations Study 2025). If ChatGPT ignores you, check Bing first; we broke down that failure mode in You rank #1 on Google but ChatGPT ignores you.
Perplexity attaches numbered, clickable citations to answers by design (Perplexity, Help Center 2026), so "cited" is unambiguous there in a way it is not in a casual ChatGPT reply. Its sources also behave differently; see how to show up as a Perplexity source.
The honest measurement caveats
Every method on this page, manual or tooled, has three limits. Vendors rarely say this part out loud, so we will.
1. API answers are not app answers. Many tracking tools query provider APIs, but the consumer apps wrap the same models in system prompts, memory and interface logic. Surfer ran 1,000 prompts through both the consumer apps and the APIs (13,779 answers in all) and found the brands mentioned overlapped only 15.5% to 23.8% of the time, with about 5% of cited pages matching for ChatGPT; Perplexity overlapped more, at 26.7% of cited domains (Surfer, Scraped vs API Study 2026). API tracking is still a consistent, comparable signal, but it is a proxy for the consumer experience, not a replica. This applies to GEOHATS too: our scans query provider APIs.
2. One sample is an anecdote. As covered above, identical prompts produce different answers. Any tracker (or spreadsheet) that reports a score from a single run per question is reporting noise. We put numbers on that in how accurate AI visibility trackers are. Look for sampling, or do it yourself.
3. Personalization skews spot checks. Logged-in accounts carry history and location context, and your own ChatGPT telling you about your own brand proves nothing. Use clean sessions, or accept a tool's neutral API vantage point as the more comparable baseline.
Where DIY breaks down
The spreadsheet method fails on arithmetic, not on principle. Thirty questions across five engines is 150 answers per week. Add the 2 to 3 samples per question that nondeterminism demands and you are reading, classifying and logging 300 to 450 AI answers every week, roughly 1,500 or more rows a month, by hand.
At one to two minutes per answer that is a 5 to 15 hour weekly job, forever. Most teams quietly stop by week six. That is what the paid tools actually sell: the discipline, automated. The cheapest ones are compared in affordable AI visibility tools.
Tools that track brand mentions in AI, with verified pricing
Five options worth shortlisting in 2026 (GEOHATS is ours, so it is listed last). Prices checked on each vendor's pricing page on November 16, 2026; confirm before buying, this market reprices often. A longer teardown lives in our best GEO tools for SaaS guide.
| Tool | Entry price | Engines at entry | Prompts at entry | Pricing model |
|---|---|---|---|---|
| Otterly.AI | $29/mo (Lite) | 3 of 7 engines you choose | 20 (2,000 credits) | Subscription (credits) |
| Peec AI | $95/mo (Starter) | 3 models you choose | 50 | Subscription |
| Profound | Free 7-day trial, then custom | Trial: ChatGPT, Gemini, AI Overviews | 50 (trial) | Enterprise contract |
| Trakkr | $100/mo (Growth, 14-day trial) | 8 engines incl. Claude, Grok, DeepSeek | 50 | Subscription |
| GEOHATS | $47 one-time (lifetime) | ChatGPT, Claude, Perplexity, Gemini, Grok (+AIO add-on) | Per-site question sets | Lifetime license + your own AI keys |
Otterly.AI $29/mo · verified Nov 2026
The cheapest monthly entry on this list. The Lite plan gives 2,000 credits a month, about 20 prompts tracked daily on 3 of Otterly's 7 engines, which you pick and can switch (Otterly.AI, Pricing 2026). Claude answers cost 3 credits instead of 1, Lite has no extra credit packs, API or MCP, and 20 prompts is tight. Best first paid step up from the spreadsheet. If you outgrow it, see Otterly.AI alternatives.
Peec AI from $95/mo · verified Nov 2026
A polished mid-market tracker with daily runs, position and sentiment metrics, and unlimited seats. The vendor publishes plans at peec.ai/pricing; Starter is $95/month for 50 prompts on 3 models, Pro $245 for 150. Strong dashboards with sentiment and source analysis, an MCP server on every plan and an API from Pro up, but extra models cost more and it doesn't write the content that fixes the gaps it finds. See Peec AI alternatives for what people switch to.
Profound Free trial, then custom · verified Nov 2026
The enterprise reference brand in this category, with deep answer-volume analytics. Brands now get a free 7-day trial (50 prompts on ChatGPT, Gemini and Google AI Overviews) and then a custom Enterprise contract; the old $99 and $399 plans are gone (Profound pricing). Its AI Marketer also drafts content. Excellent if you have the budget; oversized for a single founder. The trial runs a recommended industry prompt set rather than your own questions, so it shows the category picture, not your exact baseline. Details in Profound pricing.
Trakkr $100/mo · verified Nov 2026
The broadest engine coverage at an entry price in this group: all eight models it supports on every plan, including Claude and Grok, with no per-model fees. There is no free tier; you get a 14-day trial of Growth and are charged after it unless you cancel. Growth is $100/month ($83 billed yearly) for one brand, 50 prompts and 25 articles a month, which it can publish to WordPress, Webflow, Shopify or GitHub; extra brands are $50/month each (trakkr.ai/pricing).
GEOHATS $47 lifetime · our product
GEOHATS tracks your buyer questions across ChatGPT, Claude, Perplexity, Gemini and Grok on a schedule, records mentions and citations per engine, computes share of voice against competitors, and then does the part pure trackers skip: it writes the pages targeting the questions you are losing and publishes them to your CMS (WordPress, Webflow, Shopify and six more). Pricing is a one-time $47 lifetime license for 1 website (see GEOHATS pricing) (bring your own AI API keys; you pay providers at cost, typically cents per scan). How the tracking works is on the AI visibility tracking page.
One-time price instead of a subscription; tracking and content generation in one loop; MCP server so Cursor or Claude Code can drive it.
Newer product with a shorter track record than Profound or Peec; BYOK setup takes a few more minutes than a pure SaaS; publishing integrations beyond WordPress are new; API-based scans carry the same app-vs-API caveat as every tool above.
Disclosure: GEOHATS is our product. We have kept the comparison factual and linked every competitor's pricing; verify current features on each vendor's site.
What a good tracking setup measures
Whether you stay manual or buy a tool, judge the setup by whether it produces these four numbers per engine, per month. If it cannot, it is a demo, not a measurement system.
- Mention rate. Share of sampled answers that name your brand. The headline number, and the one nondeterminism distorts most without sampling.
- Citation rate. Share of answers that link your domain as a source. A mention without a citation is awareness; a citation is traffic and trust, and on Perplexity it is the whole game. To see that traffic in Google Analytics, add an AI channel with our free GA4 AI traffic regex.
- Share of voice. Your mentions divided by all brand mentions in your prompt set. The only metric that catches a competitor rising while your own numbers stay flat.
- Trend. All three, plotted week over week on a frozen prompt set. Given the volatility documented above, the direction of the line is the finding; any single week is noise.
Set expectations on timing, too: content changes typically take 2 to 8 weeks to show up in AI answers, because engines re-crawl and re-ground on their own schedules. Track weekly, judge monthly.
FAQ
What is the best way to track brand mentions in AI for free?
Use the spreadsheet method above: freeze 20 to 30 buyer questions, run them weekly in clean sessions across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews, and log mentioned, cited, position and competitors named for every answer. It costs an hour or two a week and produces a real mention-rate and share-of-voice trend within a month.
How often should I check whether ChatGPT or Perplexity mentions my brand?
Weekly, on the same day, with the same frozen question set. Daily checking adds noise faster than insight; monthly checking cannot separate a real trend from the platform swings Semrush documented across 2025. Weekly sampling with monthly scoring is the practical middle.
Why does ChatGPT give a different answer every time I ask the same question?
LLM inference is nondeterministic: sampling temperature plus server-side batch variation means identical prompts legitimately produce different outputs, as Thinking Machines Lab's research shows even at temperature zero. For tracking, that means one answer is an anecdote. Ask each question multiple times, record every run, and report the rate at which you appear, not whether you appeared once.
Do API-based tracking tools show what real users actually see?
Not exactly, and any honest vendor should say so. Surfer's 1,000-prompt comparison found brand overlap of only 15.5% to 23.8% between the consumer apps and their APIs, and about 5% of cited pages matching for ChatGPT. API tracking is still valuable because it is consistent over time, which is what a trend needs. Treat absolute numbers as a proxy and the trend direction as the finding. GEOHATS scans work this way too.
My brand ranks #1 on Google. Doesn't that mean AI engines mention me?
No. ChatGPT's web citations correlate with Bing, not Google: Seer Interactive found 87% of SearchGPT citations matched Bing's top organic results, versus 56% for Google. Claude often answers from model knowledge without searching at all. Track each engine separately and fix each one on its own terms.
Which metrics should a brand-mention tracker report?
Four per engine: mention rate, citation rate, share of voice against named competitors, and the week-over-week trend on a frozen prompt set. Position (are you the first brand named?) is a strong fifth. If a tool reports a single "AI visibility score" without exposing these components and its sampling, you cannot audit what moved or why.
How this page was made
Researched with AI assistance, then reviewed, edited and fact-checked by Daniel Borodin. Prices and product details were checked against each vendor's own pages on ; vendors change plans often, so confirm on their site before you buy. Spotted an error? Tell us.

Daniel builds GEOHATS and runs AI-search visibility for his own products and client sites. Posts here are researched with AI assistance; he reviews, edits and signs every one.
Keep reading
More in Playbooks →Can ChatGPT read your site?
See if GPTBot, ClaudeBot and PerplexityBot are blocked, and whether your pages load without JavaScript. Under a minute, no signup.
Run the free checkAlso available as plain Markdown for AI assistants and readers.