How to Track Your Brand Mentions in ChatGPT, Perplexity & AI Search (2026)

Somewhere right now, an AI engine is answering a question about your industry. Maybe it's recommending you. Maybe it's recommending your competitor. Maybe it's confidently repeating something wrong about your business. Most owners have never checked — which is wild, because checking is nearly free.

I check monthly, with a citation monitor I built myself, because my consulting practice depends partly on being the name Perplexity gives when people ask about frameworks I coined. This is the how-to I give clients, from free to fully automated.

Why track AI brand mentions at all?

Because the channel got big while nobody was watching the scoreboard. As of 2026, roughly 65% of consumers research purchases with AI tools, and B2B buyer studies put usage above 70%. AI answers name a handful of brands and stop — no page two. If you don't know what the engines say about you, you're flying blind in the fastest-growing research channel there is. The full argument is in my AI visibility guide; this post is purely the measurement half.

What exactly should you track? (Mention vs citation vs recommendation)

Don't record a fuzzy yes/no. Score three distinct levels — they move independently and mean different things:

• Mention: the engine names your brand anywhere in the answer. Signal: the model knows you exist.
• Citation: the engine links your site as a source. Signal: your content is trusted — this is what the AEO playbook optimizes for.
• Recommendation: the engine suggests you when asked what to buy or who to hire. Signal: revenue. This is the level buyers act on.

Also log who else appears. Competitor share-of-voice is half the value of tracking — when a rival owns a prompt you should own, that's your content roadmap writing itself.

Step 1: Build your prompt list

Ten to twenty prompts, written the way buyers actually talk. Mix four types:

• Category prompts: "best [your category] for [your customer type]" — the money queries.
• Problem prompts: "how do I fix [problem you solve]" — where citations are earned.
• Comparison prompts: "[you] vs [competitor]" and "[competitor] alternatives."
• Brand prompts: "what is [your business]" / "is [your business] legit" — checks what the engine believes about you, including errors.

Keep the list stable month to month. You're building a time series; changing the questions ruins the trend.

Step 2: Run manual spot-checks (free, ~1 hour/month)

The method that gets you 80% of the signal:

1. Use clean sessions. Fresh chat, no history — memory features skew answers toward what the engine knows you care about. Logged-out or a separate account is even better.
2. Run each prompt in ChatGPT, Perplexity, and Gemini. Add Google's AI Overviews for queries where they trigger. Perplexity is the most citation-transparent, which makes it the best diagnostic engine.
3. Score each answer in a spreadsheet: engine, prompt, mentioned? cited? recommended? competitors named. One row per run.
4. Run important prompts twice. AI answers are non-deterministic — the same question can produce different brand lists. Two runs tell you what's stable versus coin-flip.

An hour a month, total cost zero, and after three months you have a trend line most businesses would pay real money for.

One warning from experience: resist the urge to only run your brand-name prompts. "What is [my business]" is comforting and nearly useless — you'll usually show up, because you asked. The category and problem prompts, where you're competing against every other option the engine knows, are where the truth lives. Weight your list toward those.

Which engines should you prioritize?

You don't need to track everything with equal weight. My priority order for a small business, as of July 2026:

1. ChatGPT. The biggest audience by a wide margin — this is where the most buyer conversations happen, so a recommendation here is worth the most. Track it manually even though it's the hardest to automate cheaply, and run your money prompts in both regular and search-enabled modes, because the answers differ.

2. Perplexity. The best diagnostic engine, full stop. Every answer ships with visible citations, so you can see not just whether you appear but which page earned it and who beat you. It also has the friendliest API for automation. When I want to understand why a competitor owns a prompt, Perplexity shows me their receipts.

3. Google's AI surfaces (AI Overviews and AI Mode). Massive reach because they sit on top of default search behavior, and notably, Ahrefs' correlation data shows AI Mode tracks branded web mentions at 0.709 — even more mention-driven than ChatGPT. If your traditional SEO is decent, check whether the AI layer is inheriting it.

4. Gemini and Claude. Worth a quarterly pass rather than monthly. Smaller buyer volume for most niches, but Claude in particular skews toward technical and professional users — if that's your customer, promote it up this list.

The engines converge more than they diverge — fix your extractability and entity signals once and every engine benefits. So don't agonize over coverage; two engines tracked consistently beat five tracked sporadically.

Step 3: Decide if you need a paid tool

Paid tools automate exactly this loop at scale: more prompts, more engines, daily-or-weekly cadence, dashboards. As of July 2026 the credible range runs from Otterly.ai at $29/month (15 prompts) through LLMrefs at $79 flat (500 prompts), Peec AI from $95, Scrunch at $250, up to Profound at $399-$499+ with enterprise deployments in the thousands. I compared them all honestly — strengths included — in the best AI visibility tools guide.

My rule: buy a tool when the manual habit is established, the fixes are underway, and your monitoring time costs more than the subscription. A dashboard reporting "still invisible" every week is an expensive way to feel bad.

Step 4: Or build your own checker (what I did)

If you or anyone on your team can write a script, this is embarrassingly cheap. The Perplexity Sonar API is the workhorse because it does live web-grounded search and returns citations with every answer — exactly the data you're after.

The costs, verified July 2026: base Sonar runs $1 per million tokens in and out, plus a per-request search fee on the order of $5-$12 per thousand requests. A typical brand-check query — a few hundred tokens each way plus one search — costs about a penny. Sonar Pro ($3/$15 per million tokens, $6-$14 per 1K requests) runs a few cents per query. Even Perplexity's heavyweight deep-research queries top out around a dollar-plus each — and you don't need those for this job. My entire monthly audit costs less than a coffee. Not a nice coffee.

The build, in five steps:

1. Store your prompt list (the same 10-20 from Step 1) in a config file or database table.
2. Call the Sonar API for each prompt — one POST per prompt, ask for the answer plus citations.
3. Parse the response: does the answer text contain your brand name (mention)? Does the citation list contain your domain (citation)? Is your brand in a recommended/list context (recommendation)? Log competitors the same way.
4. Write results to a table with a timestamp — engine, prompt, the three booleans, competitors, raw answer.
5. Schedule it monthly (a cron job or background job) and diff against last month's run.

That's the whole architecture. Mine is a rake task plus an admin dashboard in the same Rails app that runs my site — it checks the prompts I care about, flags when a citation appears or disappears, and it's how I know (rather than hope) that Perplexity names me on the frameworks I've published. A weekend build, sub-$1 monthly run cost, and you own it. For ChatGPT-style answers you can run the same loop against the OpenAI API, though without live citations it's mention-tracking more than citation-tracking. If building this kind of small internal tool sounds useful beyond monitoring, that's the exact territory of my AI automation for small business guide.

Step 5: Set your cadence and act on the data

Monthly is the right default. Engines shift with model updates and index refreshes, but week-to-week noise mostly isn't actionable for a small business. Add ad-hoc checks after you publish a major page, after a competitor launches, and after big model releases.

Then — the part that matters — act on what you find:

• Invisible on a money prompt? That prompt needs a page: direct answer block, question headings, dated stats. The AEO playbook is the fix list, and the Princeton GEO research puts 30-40% visibility lifts on exactly those tactics.
• Mentioned but never cited? Your content isn't extractable enough, or your entity data is inconsistent. Run the AI discoverability checklist against your key pages.
• Competitor owns a prompt? Read what the engine cites for them. That's the bar. Beat it with something more current, more specific, and better structured.
• Engine says something wrong about you? Publish the correct fact prominently on your site with schema, and get it echoed off-site — engines fix beliefs when the web stops corroborating the error.

What does a real tracking setup look like?

Mine, concretely: a stable prompt list covering my consulting categories and coined frameworks; a monthly automated Perplexity run at about a penny per query; manual quarterly checks in ChatGPT and Gemini; results in an admin dashboard with month-over-month diffs. Total cost: roughly nothing. Total value: I watched myself go from unmentioned to named-by-Perplexity on three of my six frameworks, and I could tie each win to a specific page and its off-site echoes. That feedback loop — measure, fix, re-measure — is the entire game.

The zero-effort starting point

If you want the baseline without building anything: I run this exact diagnostic for businesses as a free AI visibility report — I check what ChatGPT, Perplexity, and Google's AI actually say about your business across real buyer prompts, score the mention/citation/recommendation levels, and show you which competitors are winning prompts you should own. It's the first thing I do for every client anyway, because you can't improve a number you've never measured.

Start there, or start with the spreadsheet. Just start — the engines are already talking about your market, with or without you in the answer.