How to track your brand's visibility in ChatGPT, Perplexity, Gemini and Claude
Tracking AI visibility means running a fixed set of buyer prompts against each engine on a fixed schedule and recording four things every time: whether you were mentioned, where in the answer, which competitors appeared, and which sources the engine cited. Ad-hoc spot checks do not work, because engine answers vary between runs — only a repeated prompt set produces a trend you can act on.
Key takeaways
- Track prompts, not keywords — 20 to 60 prompts covers most B2B categories.
- Record mention, position, competitors and cited sources on every run.
- Re-run weekly; single runs are noise, trends are signal.
- Compare engines separately — they disagree constantly.
Step 1 — build the prompt set
Start from how a buyer actually asks, not from your product names. A procurement engineer types a requirement, not a brand. Three families of prompt cover most of the demand:
- Requirement prompts — "halogen-free flame retardant for PA66 at 25% glass fill".
- Shortlist prompts — "suppliers of ATEX-certified instrumentation for LNG terminals in Europe".
- Comparison and reputation prompts — "X versus Y for high-H2S corrosion inhibitors", "is X a reliable supplier".
Step 2 — decide what you record
For each prompt, on each engine, on each run, capture: mentioned yes/no; position within the answer; every other company named; every source URL cited; and, where the engine exposes it, the author of that source. The last two turn a score into a to-do list, because they name the pages you need to be on.
Step 3 — run every engine, separately
Engines disagree far more than search engines do. Perplexity leans on live retrieval and cites densely; ChatGPT blends memory with browsing; Gemini and Copilot pull differently again; DeepSeek behaves differently outside Western sources. Averaging them hides exactly the gaps you need. EGNITE runs ChatGPT, Perplexity, Gemini, Claude, Microsoft Copilot and DeepSeek independently and keeps the results separate.
Step 4 — set a cadence and hold it
Weekly is the right rhythm for most B2B categories: frequent enough to catch a competitor overtaking you, slow enough that you are not reacting to model noise. Keep the prompt set stable — every prompt you add or edit resets the trend for that prompt.
Step 5 — turn the data into work
- Prompts where you are absent and a competitor is present → content or structured-data gap on that topic.
- Prompts where you are named last → weak corroboration; you need third-party sources agreeing with you.
- Sources cited repeatedly across prompts → the publications and pages worth earning a presence on.
- Authors appearing repeatedly → the people shaping how your category is described.
Doing it manually versus doing it with a platform
You can do this by hand: a spreadsheet, six browser tabs and a disciplined Monday morning. It works for ten prompts and collapses at fifty, mostly because you cannot diff last week against this week reliably. A platform earns its place at the point where you need deltas, competitor segmentation and export — EGNITE runs the prompt set, stores every run, and shows the week-over-week movement per engine and per prompt.
Frequently asked questions
How many prompts should I track?
Twenty is enough to see a pattern; forty to sixty covers most industrial categories including sector, region and comparison variants. Beyond that you mostly add cost, not insight.
Why do answers change between runs?
Generative models sample rather than look up. Retrieval, session context and model updates all shift the output. That variance is exactly why a fixed prompt set run repeatedly beats occasional manual checks.
Can I track competitors too?
Yes, and you should. Every run records every company the engine named, so competitor presence comes for free — segmented by company type so you compare producers with producers, not with distributors.
See where you stand in AI answers
Run a free GEO scan: a 0–100 score across five dimensions, page-level evidence and a prioritized fix queue. No card required.
Keep reading
Last updated 8 September 2026 · Written by the EGNITE GEO team.