How to track brand mentions in AI
A repeatable prompt-set method for tracking brand mentions across AI engines, what to log, how often to check, and how to turn results into fixes.
By ShowUp AI · Published
Tracking brand mentions in AI means running the same set of realistic buyer questions through ChatGPT, Perplexity, Gemini, Grok and Google AI Overviews on a fixed schedule, logging whether you're mentioned, in what position, and what's said about you — then turning the gaps into a prioritised fix list. It's a repeatable process, not a one-off search.
Most brands have no idea what AI assistants say about them because nobody owns this. Here's how to build the habit without buying anything, and where dedicated tools start to earn their cost.
Why can't you just Google your brand name to check this?
Answer first: a normal search shows whether you rank, but AI assistants don't rank pages — they generate an answer from several sources and decide whether to name you at all. Your brand can rank #1 in Google and still be missing entirely from an AI Overview or a Perplexity answer, because those systems synthesise a response rather than list links.
The other reason manual spot-checks fail is that AI answers are not stable. The same prompt can return a different set of brands depending on the day, the account, and small wording changes. A single check tells you almost nothing; a logged series over weeks tells you a trend.
What actually differs from classic rank tracking:
- No fixed ranking position — you're checking presence and framing, not a numbered slot.
- Answers vary by engine. ChatGPT, Perplexity, Gemini and Grok pull from different sources and write differently.
- Sentiment matters as much as presence. Being mentioned negatively or inaccurately is its own problem worth logging separately.
- Citations are visible in some engines. Perplexity and Google AI Overviews often show sources — that's free diagnostic information about who's winning the retrieval step.
What's the manual prompt-set method for tracking mentions?
Answer first: build a fixed list of 15-30 real buyer questions covering category searches, comparisons, "best X for Y" phrasing and direct brand questions, run them through each major AI engine on a set schedule, and record the results in a simple spreadsheet. Anyone can do this with no software beyond a browser and a sheet.
Step by step:
- 1Write the prompt set. Pull from what your sales team actually hears: "best [category] for [use case]", "[you] vs [competitor]", "is [you] good for [audience]", "top [category] tools in [year]", and a few pure category questions with no brand name at all.
- 2Pick your engines. At minimum ChatGPT (with browsing/search on), Perplexity, Google AI Overviews and Gemini. Add Grok if your audience skews toward X.
- 3Run every prompt on every engine. Use a signed-out or fresh session where possible to reduce personalisation bias, though you can't fully eliminate it.
- 4Log the result for each prompt-engine pair using a consistent set of fields (below).
- 5Repeat on a fixed cadence — see the cadence section — and diff each run against the last.
What to log for every prompt-engine pair
| Field | What to record |
|---|---|
| Prompt | Exact wording used |
| Engine | ChatGPT, Perplexity, Gemini, Grok, AI Overviews |
| Mentioned? | Yes / no |
| Position | First named, buried in a list, footnote mention |
| Framing | Accurate, outdated, negative, generic |
| Competitors named | Which ones, and in what order |
| Sources cited | Any URLs shown (Perplexity, AI Overviews) |
| Date | Run date, so you can see drift over time |
Keep it in one spreadsheet, one row per prompt-engine-date combination. It gets long fast, but that's the point — the pattern only shows up in the aggregate, not in any single row.
How often should you actually re-run the checks?
Answer first: a weekly light check plus a full monthly run is enough for most brands; check daily only around a launch, a rebrand, or right after you've shipped fixes and want to see if they moved anything.
A workable cadence:
- Weekly (5-10 minutes): re-run your 5 highest-priority prompts across your top two engines. This catches sudden drops or new competitor mentions early.
- Monthly (30-60 minutes): run the full prompt set across all engines. This is your trend data.
- Ad hoc: any time you publish new content, fix your llms.txt or schema, or notice a competitor doing something differently, re-run the relevant prompts within a few days.
- Quarterly: revisit the prompt set itself. Buyer language shifts, new competitors appear, and old prompts stop being representative.
Don't over-index on daily monitoring for its own sake — AI answers fluctuate enough run to run that daily noise can look like a trend when it isn't. Cadence should match how often you can realistically act on what you find.
What does the tools landscape look like if you don't want to do this by hand?
Answer first: a growing set of dedicated AI-visibility platforms will run prompt sets on a schedule and give you dashboards, but most are new, priced for mid-market budgets, and still require you to interpret and act on the results yourself — the tool tells you the problem, not the fix.
Rough categories of what's out there:
- Dedicated AI-visibility trackers. Automate the prompt-running and log mentions, share of voice and sentiment over time. Useful once you've outgrown a spreadsheet, but check exactly which engines they actually query live versus which they estimate.
- SEO suites adding AI-mention modules. Several established SEO tools have bolted on AI-overview or LLM-mention tracking as a feature. Convenient if you already use the suite, often shallower than a dedicated tool.
- Manual/spreadsheet approach. Free, fully transparent, and forces you to actually read the answers instead of skimming a dashboard number — the trade-off is time.
- Done-for-you services. Someone else builds the prompt set, runs it, and hands you a report with a prioritised fix list rather than raw data.
None of this is settled yet — the category is young, engines change their retrieval behaviour without notice, and "share of voice" numbers from different tools can disagree because they're not sampling the same way. Treat any single tool's score as directional, not gospel.
If you want a fast baseline without building your own prompt set from scratch, the free AI visibility check runs a version of this for you and shows where you currently stand.
How do you turn a log of mentions into an actual fix list?
Answer first: group your logged gaps into a small number of causes — missing structured data, thin or outdated content, no presence on the sources the engine is citing, weak or absent reviews — then fix the highest-frequency cause first, because one root cause is usually behind several missed prompts at once.
A practical way to triage:
- 1Sort by frequency, not severity. A gap that shows up across ten prompts and three engines beats a dramatic miss on one obscure prompt.
- 2Check the cited sources. If Perplexity or AI Overviews show sources and you're not one of them, that's a content or listings gap you can act on directly — see which pages they did cite and ask why yours wasn't retrievable.
- 3Separate "not mentioned" from "mentioned wrong." Wrong or outdated framing (old pricing, a discontinued feature) is often a faster fix than earning a new mention — update the page or listing that's the likely source and re-check.
- 4Map each gap to an owner. "No FAQPage schema on pricing page" goes to dev/marketing ops. "Not listed on the review site competitors are cited from" goes to whoever owns partnerships or CS.
- 5Re-run the specific prompts after each fix, not the whole set — this tells you quickly whether the fix worked before you invest further.
A simple fix-list table works well for keeping this visible:
| Gap observed | Likely cause | Fix | Re-check date |
|---|---|---|---|
| Not mentioned for "best X for Y" | No content matching that exact use case | Publish a use-case page | +2 weeks |
| Mentioned with old pricing | Outdated info on a cited third-party page | Update listing, request refresh | +2 weeks |
| Competitor cited from a directory you're not on | Missing listing | Claim and complete profile | +1 week |
| No FAQPage/Organization schema | Never implemented | Add schema markup | +1 week |
Tracking without this last step is just data collection. The value is entirely in the loop — log, diagnose, fix, re-check — repeated often enough that your visibility trend actually moves instead of sitting flat for a quarter while you admire a dashboard.
Get named in AI answers — starting today
The $299 ShowUp AI Setup checks your live site and builds your report, technical kit (llms.txt, schema, meta, FAQs) and ready-to-publish content — with a 30-day re-check included. Monthly Watch plans are optional afterwards.
Related articles
How to Show Up in ChatGPT, Perplexity, Gemini & Google AI Overviews in 2026
Crawler access, answer-first content and consistent entity data are what decide whether an AI assistant names your brand. Here is the whole system, plus a 30-day plan.
Read itAEO vs SEO: What's the Difference and Why You Need Both in 2026
SEO gets you into the set of sources an AI engine considers. AEO decides whether it quotes you. A side-by-side comparison and a priority order.
Read itWhy Your Brand Is Invisible to AI Assistants (and How to Fix It)
AI assistants do not dislike your brand — they cannot find a sentence about you worth repeating. The five causes, how to confirm each, and the fixes by impact.
Read it