How to run the test
Set up once, then ask 20 questions twice each.
- Open a temporary chat with personalization off. ChatGPT can use your saved memories when it searches (OpenAI). In a temporary chat choose Unpersonalized: "Does not use memory, custom instructions, or plugins" (OpenAI). Start a new one for each question.
- Name the city in every question. ChatGPT otherwise guesses your location from your IP address. OpenAI's own fix for wrong-area results: "Include your city, neighborhood, or postal code in your question" (OpenAI).
- Check that it searched. Answers that searched the web usually show citations and a Sources button (OpenAI). If there are none, ask again.
- Ask each question twice. Repeat ChatGPT result sets overlapped only 23% on average (BrightLocal, 2026). One run is luck.
- Log every answer in the sheet below.
The 20 questions
Write them the way customers talk, not the way you describe your business. Half generic, half specific.
| Type | How many | Plumber example |
|---|---|---|
| Best or near me | 5 | "Best plumber in Mesa, AZ" |
| A specific service | 5 | "Who installs tankless water heaters in Mesa?" |
| A specific situation | 5 | "Plumber in Mesa for a slab leak in an older house" |
| Price or comparison | 3 | "How much does a Mesa plumber charge to replace a water heater, and who's fair?" |
| Urgency, language or customer type | 2 | "Emergency plumber in Mesa open Sunday" |
Your specific questions should match the jobs you most want. Those are where a small business can win, because each one is a narrower search (see why).
The scoring sheet
One row per question. Two runs each makes 40 answers.
| Question | Run 1: named? | Run 2: named? | Who was named instead | Sources shown |
|---|---|---|---|---|
| Best plumber in Mesa, AZ | No | No | Rival A, Rival B, Rival C | Yelp, rival sites |
| Tankless water heater install, Mesa | Yes | No | Rival A | Your site, Yelp |
Add up three numbers: your mentions out of 40, your top competitor's out of 40, and how often each source appears.
Does ChatGPT recommend businesses?
Yes, a handful per answer. ChatGPT averages 4.1 businesses per response, and 10% of its local answers name none (BrightLocal, 2026). It is pickier than Maps:
"Google Maps mentions a tracked business in 66% of searches vs 32-38% for AI surfaces." (BrightLocal Local AI Visibility Study)
So don't expect to be in every answer. Expect to be in about as many as your strongest competitor.
How to read your score
| What you see | What it means | Next step |
|---|---|---|
| 0 of 40, rivals named, Sources show Yelp | ChatGPT is reading listings you haven't filled in | Fix listings first |
| Named only on specific questions | Normal for a small business; generic questions go to the biggest names | Add a page for each need you serve well |
| Named, but wrong hours, phone or services | Your listings disagree with each other | Correct Yelp, Bing Places and your ChatGPT listing |
| Named in some runs, not others | Expected; a business seen once reappears in about half of later tries (BrightLocal, 2026) | Compare totals, not single answers |
| About level with your top rival | You're in the pool | Retest monthly with the same 20 questions |
The Sources column is the most useful part. It shows which listings and pages ChatGPT used for the businesses it named. If a rival's Yelp page shows up again and again, that's where you're missing.
Is there a better way than asking by hand?
In the US, partly: claim your existing ChatGPT listing and pass the ID check. Its insights count your mentions over the past 28 days from real questions, though counts may be incomplete (OpenAI).
"For a local business, a mention counts when ChatGPT shows a result for that location, even if the customer doesn’t open the listing or visit the website." (OpenAI Help Center)
You won't see the questions themselves; OpenAI shows only "Discovery topics", grouped themes (OpenAI). Keep the 20-question test for knowing who beats you and why. Local businesses can claim only existing US listings for now (OpenAI), so in Canada the hand test is the main tool.
Example
A Mesa plumber
Score: 6 of 40. All 6 on tankless or slab-leak questions, none on "best plumber". Rival A: 22 of 40, with Yelp in almost every Sources list. Reading: the plumber's pages work for specific jobs; the gap is Yelp. Fix the Yelp page and retest in a month.
A Vancouver physiotherapist
Score: 14 of 40; the top rival 15. Two answers gave an old address. Reading: in the pool, but listings disagree. Correct Yelp and Bing Places, then retest.
A new landscaping company
Score: 0 of 40. No Sources list ever shows its site. Reading: the robots.txt file blocks all crawlers. OpenAI's crawler must be allowed before anything else can work (OpenAI).
Different for you?
- You scored low and want the fixes: how to get ChatGPT to recommend your business.
- You want to know which listing matters most: where ChatGPT finds local businesses.
- You want to test Google's AI answers too: AI Overviews and AI Mode for local services.
- Someone is selling you GEO with a score like this: is GEO worth paying for?
- You'd rather have it tracked every month: SEO & GEO Agent.