How to Measure AI Visibility: A Monthly Checklist

AI answers do not stay still. Ask the same question this week and next month and you may get a different company, a different description, or no company at all. That movement is normal, which is why guessing from one search is useless. Measuring once a month, the same way every time, is not. Thirty minutes, on the same day each month, tells you whether the work you shipped is being seen, described, and trusted — or whether it is still invisible.

The first rule is sameness. AI answers vary from run to run, so a comparison only means something when the conditions match. Use the same prompt set every month, the same engines, a fresh or logged-out session, and the same day of the month. Daily checking is the opposite of measuring: it mostly measures noise.

Start with the prompts your customers actually ask. The five-prompt self-test at https://eric-yim.com/notebook/five-prompts-to-test-ai-visibility gives you the shape of it — who does this trade near me, what does it cost, who can come soon — without needing to repeat its exact questions here. Write five to eight prompts in the words homeowners use, covering your main service, your specialty, and your price question. Keep that list frozen for the month.

Run each prompt in ChatGPT, Perplexity, and Gemini, and score what comes back three ways: named, mentioned vaguely, or absent. Named means your company is named, with the service right. Mentioned vaguely means the answer talks about your trade without naming you, or names you for the wrong job. Absent means you do not appear at all. Record the result verbatim, with the date and the engine. One line per prompt per engine is enough.

The vague middle is where most companies live, and it is worth reading closely. An answer that says “many local plumbers offer emergency service” has not recommended you; it has described a category you happen to belong to. An answer that names you but attaches the wrong service, the wrong town, or a price you do not charge is not a win either. Do not count vague as named. The whole point of the scorecard is to stop flattering yourself.

Next, track citations separately from mentions. When an answer does name or describe you, note which page it cited, if any: your own page, or a third-party list or “best of” article. Note the exact sentence it lifted, too. That sentence is the description the system currently believes about you. If the citation is a third-party page, you are being recommended through someone else’s description — which works until it does not.

Search Console is the other half of the picture, and it must be read separately from clicks. AI-answer sessions rarely click the source they cite, so a month with no new clicks can still be a month your impressions grew. Look instead at non-branded impressions, the actual queries, average position, and click-through rate for the pages you shipped. Are new queries appearing — the homeowner phrases your problem pages were written for? Impressions tell you about being seen.

Then re-check the two places AI looks before it reads your site: your Google Business Profile and your reviews. Count the reviews and check their recency, and read whether the recent ones mention specific services in plain words, the way the reviews piece at https://eric-yim.com/notebook/reviews-fuel-ai-recommendations describes. Then check the profile itself with the CLEAR Profile Check at https://eric-yim.com/notebook/google-business-profile-setup-for-ai-search: services in plain words, hours current, name, address, and phone consistent everywhere.

Log what you shipped alongside the scores, in the same place, every month. One problem page, three captioned job photos, a price range added, a profile service corrected — write it down next to that month’s results. This log is not proof that one change caused one answer; AI answers have too many inputs for that claim. It is pattern-spotting over three or four months, and it is how you learn which kind of work shows up.

Now turn the scores into a decision, starting with the simplest case. Absent everywhere, on every engine, for your main service, is a findability problem. Before rewriting anything, check the basics: does the page for that service exist, is it in plain customer words, is the profile complete, and is the business described consistently? The pillar guide at https://eric-yim.com/notebook/ai-visibility-for-service-businesses lays out the playbook, and the Three Gates test at https://eric-yim.com/notebook/how-to-get-recommended-by-ai-search sorts findability from description and trust.

Named but vague, or named wrong, is a description problem. The system found you and could not say what you do, or said it wrong. That usually means a page is missing the plain facts: the problem in the customer’s words, what the job covers, the price range, and how to book. The vertical walkthroughs show the fix trade by trade — HVAC at https://eric-yim.com/notebook/ai-visibility-for-hvac-companies, plumbing at https://eric-yim.com/notebook/ai-visibility-for-plumbing-companies, cleaning at https://eric-yim.com/notebook/ai-visibility-for-cleaning-companies, and detailing at https://eric-yim.com/notebook/ai-visibility-for-detailing-companies. Blank on a specialty means that specialty needs its own page and proof.

Cited only through a third party is a different decision: build your own page for that exact question. If a “best of” list is the page AI quotes when someone asks what your service costs, write the page that answers it in your own words, with your own price range and your own captioned job. The GEO versus SEO guide at https://eric-yim.com/notebook/geo-vs-seo-for-service-businesses explains why a page built to be quoted — direct answer first, plain facts, one quotable sentence — is the kind that earns the citation itself.

Keep the whole check on one page. A simple log with five columns is enough: date, engine, prompt, result, and citation URL, plus a short line at the bottom for what shipped that month. Use the same sheet every month so the history stays in one place. The template matters less than the habit: same prompts, same engines, same day, verbatim results, no tidying up the answers that embarrass you. The embarrassing ones are the checklist.

Do not game the measurement. Searching your own prompts fifty times will not train an AI answer to pick you, and refreshing daily will not turn a vague mention into a recommendation. Buying reviews, buying citations, or stuffing pages with town names will do worse than nothing: the first real customer who checks finds the gap. The job-photos piece at https://eric-yim.com/notebook/job-photos-content-ai-can-cite makes the honest case — captioned finished jobs, with permission, are proof no competitor can fake quickly.

What does good movement look like? Over three months, expect vague to turn named on one or two prompts, not all of them at once. Expect a third-party citation to be joined, then replaced, by your own page for one question. Expect impressions to broaden first — new homeowner phrases appearing — before clicks follow. Some months nothing moves, and that is information too: it usually means nothing new was shipped.

Start this month, on a day you will keep. Run the prompts, fill the scorecard, check the citations, read Search Console for impressions rather than clicks, re-check the profile and the reviews, and write down what shipped. Then do the same thing next month, under the same conditions, and compare. Thirty minutes, same day, every month — movement, not miracles.