The idea: run the same loop the engine runs — Ask, Read, Record
An answer engine recommends a business in three moves: it retrieves candidate sources, reads them, and selects who to name. Testing your visibility is the mirror image of that, done by hand. You Askthe engine the customer’s question, Read what it returns and where it got it, and Recordthe result so a single fuzzy answer becomes a measurement you can repeat. The whole point is to turn “I’m not sure if AI knows about us” into a dated, checkable score.
A screenshot of one good answer is a nice feeling. A dated series of the same prompts is evidence — and evidence is the only thing that tells you whether a change you made actually worked.
The test, step by step
Work through these in order. The first time takes about twenty minutes across the four engines; re-tests are faster once your prompts and scoring sheet are set up.
Write down the question your customer would actually ask
Phrase it the way a real person would, not the way you'd search. Aim for a recommendation request tied to your trade and city — e.g. “best emergency plumber in Tulsa” or “who should I hire to re-tile a bathroom near Tulsa?” Keep your business name OUT of it: you're testing whether the engine reaches for you unprompted, not whether it can repeat a name you fed it.
Run the same prompt in each engine, in a clean session
Open ChatGPT, Perplexity, Google (watch for the AI Overview at the top) and Gemini. Start a fresh chat or use a logged-out / incognito window so your own history and location don't quietly bias the result. Paste the identical prompt into each so the answers are comparable.
Read the answer for one thing: are you named, mentioned, or absent?
Score each engine on a simple three-way scale. Named: the engine recommends your business by name in the main answer. Mentioned: you appear in a list, a citation, or a follow-up but aren't a lead recommendation. Absent: you're nowhere. Note which competitors it does name — those are the businesses the engine currently trusts more than yours.
Check the sources the engine is actually reading
Perplexity and Google AI Overviews show citations; click them. ChatGPT and Gemini will usually tell you where local results come from if you ask “which sources are you using for this?” You're looking for whether your own site, your Google Business Profile, or directories that list you show up — that's the retrievable layer the AI is naming businesses from.
Confirm what the AI can even read about you (View Source test)
Open your main pages, right-click → View Source, and use find (Ctrl/Cmd-F) for your business name, address and phone number. If they aren’t in that raw HTML — because they’re baked into an image or only appear after JavaScript runs — the AI layer can’t read them either. This is the single most common reason an otherwise good business is absent (and there’s a companion guide on exactly why that happens).
Record the result so you can re-test and see movement
Log the date, the exact prompt, and the Named/Mentioned/Absent score per engine in a simple sheet. AI answers drift week to week, so one snapshot proves little — a dated series shows whether a change you made actually moved your visibility. Re-run the same prompts monthly, or after any meaningful site or profile change.
Copy-paste prompts
Swap the [bracketed]tokens for your trade, city and business name, then run each prompt in every engine. Keep your business name out of every prompt except the last one — the goal is to see whether the engine reaches for you on its own.
Best [trade] in [city]?
The bluntest test of who an engine reaches for first. If you're not in this answer, you're invisible for the highest-intent query there is.
Who should I hire to [specific job] near [city/neighborhood]?
Closer to how people actually phrase it. Engines often name different businesses here than for the blunt “best” query — run both.
Give me 3–5 reputable [trade] businesses in [city], with a sentence on each.
Surfaces the “consideration set” — the businesses the engine trusts enough to list even when it won't pick one. Being in the list is a real, recordable signal.
I need a [licensed/emergency/24-hour] [trade] in [city] this week — who do you recommend and why?
Tests whether your specifics (license, hours, service area) are readable. The “why” often reveals exactly which facts the engine has about each business.
What can you tell me about [your exact business name] in [city]?
The one place you DO use your name. It checks the entity the engine holds for you — right details, or wrong/none? Wrong facts here explain a lot of absences above.
How to read and record the answer
Resist the urge to judge the answer as “good” or “bad.” Score it on one axis only — where your business landed — so the result is comparable from week to week and across engines:
- Named. The engine recommends your business by name in the main answer. This is the win condition.
- Mentioned.You show up in a list, a citation, or only after a follow-up question — present, but not the lead recommendation. Partial credit, and a real foothold.
- Absent.You’re nowhere. Note which competitors werenamed — they’re the businesses the engine currently trusts more than yours.
Capture each run in a simple sheet: date, engine, the exact prompt, the score, and a note on who else was named and which sourcesthe engine cited. That last column is the most useful one over time — it shows you the pages and profiles the AI is actually reading you (or your competitors) from.
AI answers are probabilistic and change with model updates, your location and your account history, so treat any single result as a snapshot rather than a verdict. This is exactly the method we use in our own live, public experiment to measure whether content changes move AI visibility — and why we lean on dated, repeated runs instead of one-off screenshots.
What the engines are reading for local answers
When you check the sources behind a local recommendation, you’ll keep seeing the same kinds of places. Knowing what they are tells you where to look if your score is low:
Answer engines lean heavily on structured local data — business listings with a clear name, location, category, hours and reviews. For ChatGPT specifically, that local data reportedly comes from a combination of Foursquare and Google Business Profile data (on the order of 70% between them), with Foursquare reportedly the larger of the two. You can’t claim a Foursquare listing as directly, so an accurate, complete Google Business Profileis the practical lever most owners can actually pull — keep it correct and consistent with your website and you’ve strengthened a signal these engines read. It’s a signal to get right, not a service you have to buy.
Frequently asked questions
How do I check if AI search recommends my business?
How do I test my business visibility in ChatGPT specifically?
Why does the same prompt give me a different answer each time?
Do I need a paid account or special tool to test this?
What should I do if no AI engine names my business?
Keep going
Find out why AI can’t find your business yet
A plain-English walkthrough of why ChatGPT and other answer engines leave a business out — and the fixes that move you from absent to named.
Why ChatGPT isn’t recommending you