How do I track which AI engines mention my brand?

Q: How do I track which AI engines mention my brand?

Three options: run prompts manually across ChatGPT, Perplexity, Gemini, Claude, and DeepSeek; pay a GEO tool to do it on a schedule; or script it yourself against the engines' APIs. Manual is free, tools are fastest, DIY is cheapest at scale.

Q: Why are AI mentions harder to track than search rankings?

AI engines are non-deterministic. The same prompt can produce different brand orderings on each run. Honest tracking means sampling repeatedly so noise averages out, not single-shot ranking.

Q: What does a DIY tracking script cost?

At 30 prompts, five engines, four runs per week, the API bill lands around 10 to 20 dollars per month using cheap models like Haiku or Gemini Flash. Engineering time to build and maintain it is the bigger cost.

Q: Which GEO tools track all five major engines?

avisibli covers ChatGPT, Perplexity, Gemini, Claude, and DeepSeek. Otterly and Peec cover four engines by default with add-ons. Profound's entry tier is ChatGPT-only, expanding at higher tiers.

Question

How do I track which AI engines mention my brand?

Maciej Grabek · Accepted Answer

There are three honest ways to track AI-engine mentions: run prompts yourself across the five engines on a schedule, pay a GEO tool to do it for you, or write your own API scripts. Manual is free but does not scale. Tools are fastest. DIY APIs are cheapest at scale if you have engineering time. None of them is perfect because the engines themselves are non-deterministic. Why this is harder than tracking SEO Google rankings are deterministic from a given location at a given time. AI-engine answers are not. Ask ChatGPT "best CRMs for small teams" five times in a row and you can get five different brand orderings. Temperature, model version, conversation history, and account-level personalisation all shift the output. So tracking is really sampling: run the same prompts often enough that the noise averages out. Add to that: each engine has different access. Perplexity has a public API. ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), and DeepSeek all have APIs but with caveats - the API model is not always the same as the consumer-app model, and the consumer app sometimes adds browsing or system prompts that the API does not. So an honest tracker either accepts that the API is a proxy, or scrapes the consumer apps (which most ToS forbid). Method 1: manual prompting Pick 10-30 prompts a real customer would ask. Run each through ChatGPT, Perplexity, Gemini, Claude, and DeepSeek once a week. Note whether your brand appears, where it ranks, and what surrounding brands get mentioned. Pros: free, gives you visceral feel for how each engine talks about your category. Cons: time-consuming (an hour per round, easily), single-sample (you see one of many possible outputs), no historical chart. Good for: founders or marketers in the first month who want intuition before paying for a tool. Stop doing this manually past month one. Method 2: GEO tracking tools Several tools exist now. They all run roughly the same loop: take your prompts, run them across engines on a schedule, parse responses for brand and citation mentions, dashboard the results. avisibli - the platform that publishes this answer library. Tracks all five engines, citation share-of-voice, sentiment, competitor comparison. Otterly.AI - one of the earliest entrants, four engines by default, cheapest entry tier. Peec.ai - similar coverage, EU-based, modular pricing with paid engine add-ons. Profound - enterprise-tilted, ChatGPT-only at the entry tier, broader at higher tiers. Gauge , Writesonic GEO , Ahrefs Brand Radar - other entrants with varying coverage and price points. Pros: hands-off, historical charts, multi-prompt-per-run sampling, alerts on rank drops. Cons: cost ($30 to $500+ per month depending on tool and tier), and you still need to pick the right prompts. Tools that hand you a giant prompt library without trimming it to your category will produce noisy data. Good for: anyone past the "do I have a problem" stage. If you are publishing content and want to know whether it moves the needle, a tool pays for itself in saved time inside a month. Method 3: DIY API scripts If you have engineering time, you can wire this up yourself. Roughly: Pick 20-50 prompts. Schedule a daily or weekly job (cron, GitHub Actions, a worker process). For each prompt: call the OpenAI, Anthropic, Perplexity, Google Gemini, and DeepSeek APIs. Parse the response for your brand and competitors. Regex on brand names plus citation URLs. Store results in a database. Chart over time. Concrete cost example. 30 prompts x 5 engines x 4 runs per week x 4 weeks = 2,400 API calls per month. At an average of $0.005 per call across cheap models like Haiku and Gemini Flash, that is $12 per month in API fees. Plus your engineering time. Pros: cheapest at scale, fully customisable, you own the data. Cons: API model is not the consumer app, no UI unless you build one, you maintain it forever. Worth doing if you are an engineer who enjoys this kind of plumbing or if your prompt set is so specific no tool covers it. Which method fits where Pick by stage: Pre-launch or first month of GEO: manual , just to learn the engines. Active GEO program, one to twenty competitors to watch: tool . Massive prompt library, custom internal needs, an engineer on staff: DIY API . The honest answer is that most teams running GEO seriously land on a tool plus occasional manual spot-checks. The DIY route makes sense if your prompt list crosses a few hundred or you want to track variants no commercial tool will sample. avisibli is the GEO platform that publishes this answer library. Self-references are limited to topics where a tool-based answer is genuinely useful to readers.