Track your AI visibility with a prompt set and a spreadsheet

Measurement & Maintenanceeffort Simpact 3/5

Assistant answers are generated, not looked up. Ask the same question twice in two fresh sessions and you can get two different source lists, which means one check tells you almost nothing: you cannot separate a real change from the variance.

So measuring here is a sampling problem, and it takes the ordinary fix. Freeze the wording of the question, repeat it on a schedule, score the result the same way every time. Five to ten prompts, one run a week per engine, three possible scores. The signal lives in the shape of the column, not in any single row.

It costs twenty minutes a week and no subscription, and it pays back fastest right after you ship something, because citations move in weeks where rankings move in months.

What you get

A zero-cost weekly measure of whether ChatGPT, Perplexity and AI Overviews mention you, with a trend line. No subscription tool.

Steps

Fix a prompt set: five to ten prompts your buyers actually ask (best-of, alternatives-to, how-to-choose). Freeze the wording; a moving question cannot show a trend.
Once a week, on the same day, run every prompt in a fresh session (no history, no memory) in ChatGPT, in Perplexity, and in Google where AI Overviews appear.
Score each run in a sheet, one row per prompt-engine-week: 0 absent, 1 cited (your URL among the sources), 2 named in the answer text.
Log which competitors and sources appear each run; that column becomes the target list for the other AI Search plays.
In Search Console, add a query filter with the regex .{75,} — unusually long queries are a usable proxy for assistant-originated searches; note the count monthly.
In your analytics, segment referral sessions from chatgpt.com and perplexity.ai. AI Overview clicks stay inside Google organic; Search Console reports them under the Web search type with no separate breakdown.
Chart mention rate per engine monthly, and annotate the chart with what you shipped, so movements have causes attached.
After shipping any change from the AI Search plays, re-run the set within days.

How to verify it worked

The sheet itself: a mention-rate trend per engine, with nonzero rows appearing within four to eight weeks of running the AI Search plays.
The referral segment shows sessions from chatgpt.com or perplexity.ai once citations exist.

Why it works

The competitor column is the part people skip and then miss. Every run tells you which sources the engines reach for on your questions, so the same sheet that measures you also produces the target list the other AI Search plays work from.

Evidence

The method assembles the practitioner audit ritual in our research: a fixed multi-prompt weekly audit, plus the .{75,} Search Console regex, which practitioners report catches roughly 80 percent of assistant-originated queries (claim as reported).

MagicSpace SEO's AI and LLM SEO course (magicspace.agency) uses this loop and sets the working targets: appearing in 20-30 percent of runs by month one, 50-70 percent by month three (course claims, not our measurements).

Google's AI-features documentation (reviewed 2026-09-01): AI Overview traffic is reported inside Search Console's Web search type — there is no separate report to wait for.

Last reviewed: 2026-09-01. A play whose evidence stops holding up is rewritten or retired.

Related plays

Browse all free plays