10 Prompts That Tell You Whether ChatGPT Recommends Your Business
Someone types “best bookkeeper in Toronto for small incorporated businesses” into ChatGPT. Back comes a short list of four names, a sentence of reasoning for each, and maybe a link or two. The person picks one and books a call.
You never see any of it. No impression logged in Search Console. No referrer in your analytics. No weekly rank tracker email telling you that you dropped off a list you didn’t know existed.
That gap is the problem. Reporting was built for a world where people clicked ten blue links. When the answer arrives pre-chewed, your visibility goes dark unless you go looking for it on purpose.
The good news: you can check this yourself in about twenty minutes. Below are ten prompts to copy and paste, a scoring system out of 30, and what each type of failure is actually telling you.
First, Set Up a Clean Test

Skip this part and your results will be flattering nonsense.
Log out, or use a temporary chat. ChatGPT remembers you. If you’ve spent six months asking it to rewrite your own service pages, it knows your brand and will happily bring it up. That’s not a recommendation, that’s an echo. Turn off memory and personalization, or open an incognito window.
Start a fresh chat for every prompt. Answers within one thread contaminate each other. Name your business in message two and it keeps surfacing in message five.
Run each prompt twice: once with web search on, once off. With search off, you’re testing what the model absorbed during training. With search on, you’re testing whether your site and your mentions are retrievable and quotable right now.
Name the location precisely. The model does not reliably know where you are, and North America is loaded with duplicate city names. Write “in London, Ontario” or “serving the Dallas Fort Worth metro” rather than “near me”. Include the province or state every single time.
Selling on both sides of the border? Test both. Run prompts 1 to 3 once for a Canadian city and once for a US one. Source density is uneven, and a firm scoring 20 in Seattle can score 6 in Saskatoon simply because fewer publications cover that market.
Save everything with a date. Paste responses into a doc. Outputs shift week to week, and you want a baseline to compare against later.
Run the same set in Gemini, Perplexity and Copilot when you have time. ChatGPT is the biggest single door, but it isn’t the only one.
The 10 Prompts

Prompts 1 to 3: Cold discovery
These simulate a buyer who has never heard of you.
Prompt 1
What are the best [your service] companies in [your city]? List five and explain why you’d suggest each one.
Prompt 2
I run a [customer’s business type] and I’m struggling with [problem you solve]. Who should I hire to fix this? Give me specific companies, not general advice.
Prompt 3
I have a budget of around [typical deal size] for [your service]. Which providers in [your area] would be a sensible fit?
Prompt 3 matters more than people expect. Budget qualifiers reshuffle the list, and plenty of businesses show up for the generic query but vanish once price enters the conversation.
Prompts 4 to 6: Brand knowledge
Now you name yourself and find out what the machine thinks it knows.
Prompt 4
What can you tell me about [Your Business Name] in [City]?
Prompt 5
Is [Your Business Name] a good choice for [specific service or customer type]? What are their strengths and weaknesses?
Prompt 6
What are [Your Business Name]‘s prices, business hours, and service area?
Prompt 6 is your accuracy audit. Wrong hours, a phone number from two moves ago, a service you dropped in 2023: all of it is being handed to prospects as fact.
Prompts 7 to 9: The comparison round
Buyers rarely ask about one company. They ask about the shortlist.
Prompt 7
Compare [Your Business Name] and [Your Closest Competitor]. Which would you recommend and why?
Prompt 8
I’m deciding between [Competitor A] and [Competitor B] for [service]. Is there anyone else in [area] I should be considering?
Prompt 9
What do customers say about [Your Business Name]? Are the reviews positive?
Prompt 8 is the sneaky one. It tests whether you get pulled into a shortlist you weren’t part of, often the last route in when competitors own the obvious queries.
Prompt 10: Source mapping
Prompt 10
If I wanted to research [your service] providers in [your city], which specific websites, directories and review platforms would you check?
This one doesn’t score your business. It tells you where the model looks. Whatever it lists back at you is your target list for citations, listings and outreach over the next quarter.
Score It Out of 30
Give each prompt a score from 0 to 3.
| Score | What happened |
| 3 | Named and recommended, details correct, and your own site referenced or linked |
| 2 | Named accurately, but positioned below competitors or with no source attached |
| 1 | Named only after you prompted it directly, or named with wrong or vague details |
| 0 | Not mentioned at all, or described incorrectly enough to lose you the sale |
For prompt 10, score 3 if you have a live, complete, reviewed profile on most of the sources it names. Score 0 if you’re on none of them.
Add up the ten scores.
- 24 to 30: You’re in the recommendation set. Protect it and monitor monthly.
- 16 to 23: Known but not preferred. You’re losing head-to-head comparisons.
- 8 to 15: Fragile. You exist in fragments, and the model is filling gaps with guesswork.
- 0 to 7: Effectively invisible. Your competitors are getting the introduction you’re not.
Most owner-run businesses land between 8 and 18 on the first run. That’s normal, and it’s fixable.
What Each Failure Mode Actually Means
Failure 1: Nothing comes back at all
Zeros on prompts 1 through 4 mean there isn’t enough consistent information about your business for the model to treat you as a real, distinct entity. Usually the cause is boring: inconsistent business name and address details across listings, no schema markup, and almost no third party mentions. You’re a website, not a known entity. This is the core problem that generative engine optimization exists to solve.
Failure 2: You appear, but the facts are wrong
Wrong pricing, an old address, a service you dropped years ago. The model isn’t malfunctioning. It’s averaging conflicting sources, and some of those sources are stale. Hunt down the abandoned directory profiles and outdated third party write-ups, then fix the highest authority ones first.
Failure 3: You only exist when you’re named
Strong scores on prompts 4 to 6, weak scores on 1 to 3. You’re in the index but not in the recommendation logic. The missing ingredient is almost always independent corroboration: reviews with substance, case studies, mentions in industry roundups, and press that isn’t self-published.
Failure 4: The model cites a directory instead of your website
You get recommended, but the source is Yelp, Angi, HomeStars or a roundup post from a publisher who has never met you. Your own pages aren’t being quoted because they aren’t easy to quote. Wall-of-text service pages with no clear headings, no direct answers and no specifics get passed over for a directory that states your service area in one clean line. Restructuring your pages around clear questions and clear answers is largely a content writing problem, not a technical one.
Failure 5: You rank well on Google but score badly here
Common, and disorienting. Classic rankings measure link authority and query relevance. AI recommendations run on entity clarity, consistency, corroboration and quotability. The overlap is partial, which is why SEO and GEO are different disciplines rather than the same job with new branding. If you’re already familiar with the frustration of ranking first and still getting no clicks, this is the same disease at a later stage.
Failure 6: Your content sounds like everyone else’s
If your pages were spun up from templates or generic AI drafts, they contain nothing specific enough to cite. No named process, no real numbers, no local detail, no opinion. There is nothing in filler copy worth surfacing. Worth reading if this applies: whether AI website builders are good for SEO.
What To Do With Your Score
Fix in this order, because the sequence matters.
- Correct your factual footprint. Same business name, address, phone and service list everywhere it appears online.
- Add Organization and Service schema so machines can read the basics without guessing.
- Claim and complete a profile on every source prompt 10 named.
- Ask for reviews that describe what you did, not just how many stars you get. “They rebuilt our stock system in six weeks” is a citable fact. “Great service” is noise.
- Rewrite your three most commercially important pages to answer real buyer questions in plain, specific language.
- Publish comparison and alternatives content, since prompts 7 and 8 rely on it existing somewhere.
Re-run It Every Quarter
Score once, fix, then score again ninety days later with the same prompts and rubric. Movement from 11 to 19 is a real result even if your keyword rankings never budged, because more of the conversations that never reach your analytics are going your way. The businesses winning here aren’t the ones with the biggest budgets. They’re the ones who bothered to check.










