How to Read Your AI Visibility Test Results: What the Numbers Actually Mean

You followed the playbook in How to Test Whether AI Recommends Your Business. You asked the engines the questions your buyers ask. You have results — maybe a spreadsheet, maybe screenshots, maybe a lump in your stomach.

This post is about the next step, where most owners go wrong: reading the numbers. Not collecting them — interpreting them. Because a mention rate is like a blood pressure reading: useless until you know what band it falls in, what it's trending toward, and which number to treat first.

Every example below is real, from 700 tracked checks of one Dallas–Fort Worth plumbing market between July 7 and August 3, 2026.

The nine rules for reading AI visibility results

1. One check is weather. A month is climate. AI answers are non-deterministic — the same prompt, same engine, same city can name you Tuesday and skip you Wednesday. A single check proves almost nothing in either direction. Rates over weeks are the signal; anything else is lying to yourself.
2. Read per engine. Never average. Our tracked company scored 92% on Gemini and 0% on Google AI Overviews in the same window. Averaging those to "46% visible" describes nothing that exists. Each engine researches differently and each is won separately.
3. A zero is a structure problem, not a quality problem. The same company that Gemini named nine times out of ten was cited by Google AI Overviews zero times in 84 checks. The work was identical; the surface was different. A hard zero on one engine while others name you means that engine's pipeline can't see or can't verify you — which is fixable, and is very different from being unknown everywhere.
4. Mention rate is not market share. 92% on Gemini means the engine named the business in 92% of our tracked prompts during the window. Real buyers phrase things their own way, from their own accounts, in their own suburbs. Treat the rate as signal strength, not as a count of customers seeing your name.
5. Position inside the answer matters. "Call X first" and "X also exists" are both mentions. When you log results, note whether you're the lead recommendation or the afterthought — moving from third-named to first-named is invisible in a mention rate and enormous on the phone.
6. The competitor ledger is half the value. Every check that skips you names someone else. Write those names down. After a few weeks you know exactly who owns each engine in your market — which is your target list, not a reason to feel bad.
7. The citation trail tells you why. When an engine shows sources, read them. In our window, the three research engines' most-consulted source about the tracked company was its own website, followed by Angi, BBB, Yelp, HomeAdvisor, and BuildZoom. The sources an engine consults are the surfaces you can fix — the full breakdown is in The Supplier Signal, our white paper on who AI actually trusts.
8. Compare like windows, not adjacent runs. Rates move for reasons that have nothing to do with you — engine updates, index refreshes, seasonality. Compare this month to last month, not this run to yesterday's run, and judge direction over four weeks minimum.
9. Volatility is normal. Trends deserve reactions; single dips don't. A 75% engine will hand you a 50% week eventually. React when three consecutive weeks move the same direction, or when a formerly warm engine goes cold and stays cold.

The reading bands

Once you have a real per-engine rate, place it on this scale:

One company, one window, five engines — and every band on the chart occupied at once. This is why averages lie.

What each band tells you to do

85%+ — defend. You are the default answer on that engine. Your job is boring: keep the sources that built the position consistent, keep reviews flowing, and keep your site parseable. Change nothing dramatic.

60–85% — push. The engine trusts you but hedges. Study the checks where you were skipped: what phrasing lost you? Who was named instead? Usually the gap is prompt coverage — you win "water heater replacement" and lose "tankless water heater installer" — which is a content and question-cluster problem.

25–60% — contest. The engine knows you exist and doesn't yet have a reason to prefer you. This is where the citation trail earns its keep: find what the engine consults in your market — directories, registries, the chamber — and make every one of those surfaces say the same, complete, current thing about you. The Local Entity Gap is the manual for this band.

1–25% — rebuild foundations. Don't chase advanced tactics from this band. Name-address-phone consistency, a Google Business Profile that's actually complete, directory claims, and a website that states your services and cities in plain, extractable language. The order of operations is the same one our audit prioritizes: local entity first, index layer second, answers after.

0% — diagnose the structure. A hard zero with healthy rates elsewhere means the surface itself can't process you: AI Overviews not citing you is a different failure than Siri not finding you, and each has its own fix. Zeros are also where the biggest gains hide — every point of visibility on an empty surface is a point your competitors don't have either.

A note on reading DFW results specifically

Dallas–Fort Worth is not one market; it's a mesh of suburb-level markets. An Arlington plumber can be the default answer in Arlington and an afterthought in Plano, twenty-five miles away. When you test, phrase prompts the way buyers in each city you serve would — "slab leak repair Mansfield," not just "plumber DFW" — and read each city as its own scoreboard. Exclusivity in our practice is per market per trade for exactly this reason.

The shortcut

Everything above assumes you're collecting and reading the data yourself, and the full DIY protocol is in the AI Search Measurement Playbook. If you'd rather have the reading done for you: the $27 AI Visibility Audit scores all six signals for your business and hands you the fix order, and the free check takes ninety seconds and shows you what AI likely says about you today.

---

This is the interpretation companion to How to Test Whether AI Recommends Your Business. The measured baseline behind every number in this post is published in We Ran 700 AI Visibility Checks in One Texas Market.