How to Track Your AI Visibility With a Spreadsheet and 12 Prompts

Only 22% of marketers track how their brand shows up in AI answers, while 72% of marketing leaders expect AI search to overtake SEO within three years. That gap is from Yext's 2026 research, and it means most companies have no idea what ChatGPT, Perplexity, and Google's AI Overviews tell buyers about them, in the channel where 51% of software research now starts.

There is a growing market of AI visibility platforms that will monitor this for you. Some are good. But you should not buy one before you have run the manual version, for the same reason you should not buy a CRM before you have a sales process: the tool automates a practice, and you do not have the practice yet. The manual version costs an hour a month and a spreadsheet.

The method in one paragraph

Write a fixed set of buyer-shaped prompts. Run the same set against the same engines every month. Log whether you were mentioned, where, described how, and citing what. Fix what the log tells you. The discipline is in the word fixed: if you change the prompts each month, you have opinions, not a trendline.

The 12 prompts

Adapt the bracketed parts to your category. Keep the phrasing buyer-shaped: how a real person types, not how a marketer would.

1. best [category] tools for [ICP, e.g. mid-market B2B SaaS]
2. [your product] vs [main competitor], which should I choose
3. [your product] alternatives
4. is [your product] worth it in 2026
5. [main competitor] alternatives
6. how do I solve [the core problem your product solves]
7. [category] tools that integrate with [the platform your buyers live in]
8. [category] for [second segment you care about, e.g. regulated industries]
9. what is [your product] and who is it for
10. [category] pricing, what should I expect to pay
11. cheapest way to [job to be done] without buying [category incumbent]
12. [your category] recommendations from people who actually use them

Prompts 1, 3, and 5 are where shortlists form. Prompt 9 tells you what the models believe you are, which is your positioning reflected back at you. Prompt 11 catches the build-versus-buy answer that quietly removes you from consideration.

The log

One spreadsheet, one row per prompt per engine per month. Six columns:

Column	What to record
Mentioned	yes or no
Position	1st, 2nd, 3rd named vendor, or absent
Description	the exact words the engine used about you
Sources cited	which pages the answer linked or named
Competitors named	who else appeared, in what order
Delta	what changed since last month

Run each prompt in a fresh session with no account context where possible. Log the answer as given. Do not re-roll until you get a flattering one; you are measuring the median buyer's experience, not your best case.

Reading the results

Three patterns show up in almost every first run.

You are described in old language. The engine's description of your product lags your current messaging, sometimes by years, because it is synthesized from everything ever published about you. The fix is consistency: your site, your review profiles, your partner pages, and your PR all using the same category language and claims. Models reward the version of you that appears most often, not most recently.

A competitor owns a comparison you never wrote. If prompt 2 or 5 cites a competitor's comparison page, that page is doing your positioning for you. Write the comparison yourself, honestly, with a real feature table. Sourced, structured comparison content is among the most-cited formats in AI answers.

Your mentions cite sources you do not control. Review sites, forum threads, an analyst blurb. Per Ahrefs' 2025 research, brand mentions across the web predict AI citations about three times more strongly than backlinks do. Earned mentions are now visibility infrastructure, which moves PR from a brand function to a demand function.

When to graduate to a tool

Automate when two things are true: the prompt set has been stable for three months, and someone acts on the log every month. The platforms add daily sampling, more engines, and share-of-voice math, which is real value once the practice exists. Until then a tool produces dashboards nobody reads. If you would rather wire the manual version into n8n and run it on a schedule, the pattern is the same as the other systems in our workflows post.

The strategic context for all of this, why answer engines reward authority and statistics, how citation behavior actually works, and what it changes about the PMM role, is Chapter 2 of Product Marketing in the Age of AI. The executive summary and Part I are free on the page.

Common questions

How often should I check my brand's AI visibility?

Monthly is enough to see movement without chasing noise. AI answers vary run to run, so what you are tracking is the trend across a fixed prompt set, not any single answer. Weekly checks make sense only during a launch, a rebrand, or active PR.

Do I need to track every AI engine?

No. Start with the two or three your buyers actually use: for most B2B categories that is ChatGPT, Perplexity, and Google's AI Overviews. Each engine has different citation behavior, so log them separately rather than averaging.

Can I just ask the AI why it did not mention my brand?

You can, and the answer will sound convincing, but models cannot reliably introspect their own retrieval. Trust the pattern in your log over the model's explanation of itself. The reliable levers are consistent entity language, sourced statistics, comparison content, and earned mentions.