The 20-prompt test

Measurement starts with fixed questions. If you improvise new questions every month, you can never tell whether the answers changed or the questions did. So build a prompt set: about twenty questions, written in buyers' words, saved verbatim, that you will ask the tools every month in fresh sessions. Twenty is enough to see patterns and few enough to finish in one sitting.

Write them in four families, five each. Category prompts: “Best [category] in [town]” and its natural variants. Specialty prompts: the niche you chose in your master paragraph — “[category] that does [specialty], near [place].” Problem prompts: the situation before the search — “My [thing] broke / I need [outcome], who should I call?” Comparison prompts: “[You or a competitor] vs alternatives” and “Is [your business] good for [use case]?” Include your own name in a few — you need to know what the answers say when asked about you directly, not just whether you surface unprompted.

Run the test by hand: ChatGPT, Gemini, Google for AI Overviews, Perplexity if available — fresh chat each prompt, search or grounding on, no leading follow-ups. By hand is a feature, not a limitation: you read every answer whole, notice the wrong facts and the reasons given for competitors, and learn your category the way the tools see it. Automated dashboards can scale this later; the first months of reading raw answers are where the insight lives.

Record, for each prompt and tool: named or not, cited or not, who was named, and any factual claim about you. That is one row per answer in the spreadsheet the next lesson builds. Your Module 1 baseline was the first run of this test — from here on, it is a series, and a series is what turns anecdotes into evidence.

Score it in a spreadsheet

The spreadsheet needs three numbers per month, computed from your twenty prompts across the tools you test. Mention rate: the share of answers that name your business. Citation rate: the share that cite your site as a source. And share of answer: of all the businesses named across all answers, what share of the mentions are yours versus each competitor's. With twenty prompts and two tools, that is forty answers — an hour of testing and recording, tops.

Work an example. Forty answers: you are named in 14 (mention rate 35%), cited in 6 (citation rate 15%). Across all answers, 92 business mentions in total; yours are 14 of them, the top competitor has 31 — your share of answer is about 15% against their 34%. Those three numbers, month over month, tell the whole story of this course: the entity and content work should push citation rate, the authority work should push mention rate and share, and you can see which lever is lagging.

Use the calculator below to get a feel for the math with your own baseline. One warning about interpretation: small samples swing. A move from 30% to 35% in one month is weather; the same move held over three months is climate. Judge quarters, not months, and judge against the competitor set, not against perfection — in most local categories, the leader's mention rate is well under half. You do not need to be named in every answer. You need to be named in more of them than the business down the street.

Keep the raw answers, not just the scores. A tab per month, the full text saved, takes two extra minutes and becomes priceless later — when a fact about you changes in the answers, or a competitor suddenly surges, the archive tells you exactly when and, usually, why.

Read the answers like an auditor

The scores tell you where you stand; the answer text tells you what to do Monday morning. Read each month's answers with three auditor's questions, and keep notes in the spreadsheet next to the scores.

What is wrong? Every wrong fact about you — old address, dead service, invented price — is a work order with an address attached. Trace it to the published source (your stale page, the directory nobody updated, the article from three years ago) and fix it there; Module 2's discipline, applied forever. If a wrong claim has no findable source, out-publish it: make the correct fact the loudest, most repeated version, on your site and profiles, and watch it fade over the following months.

Why them? When a competitor is named or cited and you are not, open the cited page and read it as an extraction contestant. What passage did the tool lift? What format is the page — comparison, list, guide? What proof does it show that you do not? You are not copying their text; you are learning what the answer needed and where your record is thinner. Half of next month's content plan is sitting in this month's competitor citations.

What is missing? Questions where nobody in your category gives a good answer — thin citations, hedged language, sources from another state — are unclaimed ground. A clear page built for that exact question faces almost no competition at extraction. Keep a running list of these gaps; they are the highest-return pages you will ever write, because the contest for them has not started.

The monthly routine: 90 minutes that compound

Everything in this course converges on a routine small enough to survive your calendar. Ninety minutes, once a month, same week each month, four blocks in order — this is the entire maintenance contract for AI visibility, and it is deliberately boring. Boring is what compounds.

Test (30 minutes). Run the 20-prompt set across your tools, record named/cited/competitors/facts in the spreadsheet, save the raw answers. Audit (20 minutes). Read for wrong facts and trace each to its source; fix what is fixable on the spot — a profile edit, a stale page — and list the rest as tasks. Check your profiles for anything that drifted: hours, services, photos, new reviews unanswered.

Publish (25 minutes). Ship one improvement from the auditor's notes: refresh the page whose facts rotted, write the gap page nobody owns, add this season's numbers to the pricing page, re-sequence one more service page answer-first. One page a month is twelve a year — more citable surface than most competitors will build in total. Earn (15 minutes). One authority action: pitch the list editor, ask this month's happiest customers for reviews with the link ready, answer two community questions properly, send the podcast pitch. Small, monthly, forever.

Then stop. The discipline of this system is that it is finished — you are not behind, there is no daily grind, and the record you are building does its work between sessions while the tools re-read the web. Review the scores quarterly against the competitor set, adjust the prompt list only by adding, and let consistency do what consistency does. The businesses that will own AI answers in your category are not the ones that sprinted once. They are the ones still doing the boring ninety minutes next year.

• TEST (30 min): ran the fixed 20 prompts, scored and saved the answers
• AUDIT (20 min): wrong facts traced and fixed, profiles checked, reviews answered
• PUBLISH (25 min): one page created or refreshed from the auditor's notes
• EARN (15 min): one authority action — a pitch, a review ask, a community answer
• Scores compared against last quarter, not last month