To check whether ChatGPT recommends your business, test several questions that a real customer might ask, record the exact answers and repeat the same prompts later. One answer is evidence of one test, not a permanent rank.
Run a first check: Loop's free AI Visibility Checker samples local and city-wide questions for your business and shows the answers it receives. Use the steps below to interpret the result.
Choose questions tied to real demand
Include your type of business, a customer need and a location. A restaurant might test “family dinner near the station,” “vegetarian lunch in the city centre” and “where can I book a table tonight?” A hotel might test “quiet hotel near the airport with parking.” Avoid testing only your exact business name; that checks recognition, not discovery.
Run a small set of prompts at both neighbourhood and city level. Keep the wording, date and location in a log so you can compare later. The visibility checker asks several relevant questions for you, while manual checks can test unusual services or specific accessibility needs.
Record more than a yes or no
- Was your business recommended, mentioned only in passing, or absent?
- Was the description accurate about location, services, price range and opening hours?
- Were competitors named, and why did they fit the request?
- Did the answer cite a source or offer a usable route to your site or booking page?
- Could a customer complete the next action without guessing?
Different assistants use different information and may show different answers. Even repeated tests of the same prompt can vary. Our measurement guide shows how to report a pattern without inventing an “AI ranking.”
Check the sources that a customer can verify
Review your own website, Google Business Profile and booking or listing pages. Is the same business name, address, menu or service list visible across them? Are hours current? Can a customer tell what makes you a fit for the prompt? Google recommends complete, accurate local business information and says its AI Search features rely on established search best practices.
For ChatGPT search, OpenAI says public sites can appear and that allowing OAI-SearchBot helps make content available for search results. Allowing crawling does not guarantee a recommendation. If a technical setting blocks a search crawler, ask your website provider to check it.
Turn the result into a short work list
Fix wrong facts first. Then add useful details that answer real customer questions: accessibility, dietary options, parking, cancellation terms or the experience around your business. Next, test the booking journey. A business can be recommended accurately and still lose the customer at a broken reservation step.
Use the diagnosis guide for missing recommendations, or the AI-ready website checklist if the problem is on your own site. The broader AI visibility guide connects both.
Test discovery questions separately from name recognition
“Tell me about Acme Café” and “where can I get breakfast near the station?” are different tests. The first asks whether the assistant can identify a known business. The second asks whether it selects the business for a customer need. A company may pass the first test and fail the second without anything being technically broken.
Start with three branded checks: the exact name, the name plus town, and the name plus a specific service. Then test six to ten unbranded customer questions. Keep the business type and location accurate. If you are a breakfast café, do not include dinner prompts simply to create a larger sample. The full AI visibility guide explains how to use the result.
Build a prompt set that reflects your customers
| Intent | Example question | What to inspect |
|---|---|---|
| Discovery | “Where can I find a family-friendly café near the museum?” | Whether relevant businesses appear and why |
| Comparison | “Compare two quiet hotels near the station with parking.” | Accuracy of amenities and location |
| Time-sensitive | “Where can I book a table tonight?” | Hours, availability and booking route |
| Specific need | “Which salon offers a first-time lash consultation?” | Whether your service details are understood |
Replace the examples with questions from calls, emails, site searches or staff conversations. A question is useful when a real customer might ask it and the answer could change a purchase decision. Avoid prompts that contain your business name or a distinctive slogan unless you are deliberately running a branded test.
Keep the test conditions clear
Location is crucial for local discovery. Record the town, suburb and any location setting used by the assistant. A city-centre answer may legitimately differ from a neighborhood answer. Record the date and, for “open now” or availability questions, the time. These details make later comparisons fairer.
Do not assume every provider uses the same live information or responds the same way. ChatGPT search, Google's AI Search features and other assistants have different interfaces and sources. A tool that samples multiple assistants gives a broader picture, but the results are still observations of specific queries. OpenAI's crawler documentation explains how ChatGPT search may discover public web pages; Google's AI Search documentation describes its own eligibility requirements. Neither promises a business recommendation after a crawl.
Read the answer like a customer would
Suppose an assistant names your hotel but says it has free parking when it does not. That is an accuracy problem, even though the business was included. Suppose it correctly identifies your restaurant but links to an old menu with no reservation option. That is an action problem. Note both in your report.
Look at any cited links or displayed sources. Open them and check the underlying details. If the source is your own site, update it. If it is a booking partner or a business profile you manage, correct it there. If an independent source is wrong, you can request a correction, but the publisher controls its content. The diagnostic guide helps identify the next fix.
Compare against competitors without copying them
Record which competitors are recommended for the same customer question. Ask what information makes them a better match: a clearly stated service, a relevant neighborhood, open hours, a price range or a booking route. This is a content-gap analysis, not a request to copy their wording. Your business may have a different strength worth making explicit.
For example, a restaurant that is not recommended for “private dining for ten” may already offer it but mention it only in an Instagram story. A short page with capacity, room details, sample arrangements and an enquiry form helps customers directly. It may also give assistants more reliable information to use.
What should you save from each check?
Keep the exact prompt, provider, date, location, full answer, businesses named, cited links, factual errors and next action. A screenshot or exported answer can preserve context, but summarize the evidence in a sheet so you can compare runs. Record whether the recommendation led to a working page or booking route. Do not label a click as a completed booking.
If you change your hours or service page, note the date of that change. Search and assistant systems may take time to discover it. Repeat the same core prompts later and compare a pattern across several questions, not one headline result. The measurement guide provides a simple reporting approach.
Turn the result into a decision, not a verdict
After the check, write one sentence describing the pattern: “Our restaurant is named for neighborhood lunch prompts, but the assistant gives old Sunday hours,” or “We appear for our name but not for the service question customers ask most.” That statement is more useful than a grade because it identifies work someone can own. Assign the factual correction, page improvement or booking repair to a person and set a date to verify it.
If the answer is already accurate and actionable, you may have little immediate site work to do. Continue monitoring and invest in the actual customer experience. If the answer is wrong, fix the source you control before publishing another promotional post. The goal is for a customer to make a good decision, not for the business to appear in every possible prompt.
Frequently asked questions
Why does ChatGPT recommend me once but not the next time?
Answers can vary with prompt wording, location, available sources and the system's response. One omission does not establish a permanent loss. Use a small, fixed test set over time and look for repeated patterns.
Is a visibility score a Google or ChatGPT ranking?
No. A third-party checker can summarize its own sampled answers, but it does not have access to every assistant's internal ranking or decision process. Google's guidance explicitly warns that third-party tools do not have its internal ranking metrics.
Can I test another city?
Yes, if customers from that city genuinely travel to you. Record it as a separate market. A neighborhood test and a city-wide test should not be combined into a single “rank” because the customer context differs.
What if the answer recommends a competitor with worse reviews?
Do not infer a single cause from that result. The competitor may be closer, better matched to the prompt, available at the requested time or more clearly described in accessible sources. Inspect the full answer and the customer's actual need.
Next step: Run the free AI Visibility Checker.