How to Check Whether ChatGPT Recommends a Business
To check whether ChatGPT recommends a business, define a relevant customer question, record the search and conversation context, save the answer and distinguish a mention from an actual recommendation. Repeat a small, preselected set of useful questions rather than treating one response as a permanent ranking. The result is a bounded observation of what the system returned under those conditions.
A business-name lookup answers a different question from an unbranded request for a provider. Keep both if they are useful, but label them separately. This manual method helps an agency understand the evidence behind a visibility conversation. The AI agency client acquisition guide explains the broader commercial offer; this article focuses on checking and interpreting ChatGPT responses responsibly.
Decide what you are trying to learn
Start with the service, customer type and location that matter to the business. A homeowner seeking an emergency repair has a different need from a procurement manager comparing long-term maintenance providers. Write the intended customer situation before writing prompts. Otherwise it is easy to produce a large set of questions that are technically different but commercially irrelevant.
Separate three possible objectives. You may want to see whether the system can identify the business, whether it describes the business accurately or whether it includes the business in a relevant recommendation. These objectives need different prompts and interpretations. A correct answer to "What does this company do?" does not establish that an unfamiliar customer would be introduced to that company.
Agree the scope with the owner when possible. Confirm the services and areas they actually want to serve. Do not build a prospecting claim around a location outside their priorities. If you are checking before contact, record those assumptions and invite correction instead of presenting the sample as a complete assessment of their commercial visibility.
Keep branded and unbranded prompts separate
A branded prompt includes the business name. It can reveal identity confusion or an inaccurate description, but it gives the system information a discovery-stage customer may not supply. An unbranded prompt asks about a service or provider without naming the business. That can be useful for observing discovery, provided the wording represents a plausible customer request.
Do not insert the target business into a follow-up and then count the resulting mention as an independent recommendation. Conversation history can influence the response. Use a fresh conversation for a separately defined observation and record any relevant context. If you are deliberately testing a follow-up journey, preserve the whole sequence and label it as such.
Write the prompt set before inspecting results. Include a small number of meaningfully different customer needs rather than many cosmetic rewrites. Repeatedly changing the wording until the business appears creates a selected example, not a fair baseline. Likewise, selecting only responses where the business is absent can exaggerate the apparent problem in a sales pitch.
Record the search context you can actually observe
OpenAI's ChatGPT search documentation explains current search behaviour, sources and location-related context. Product controls can change, so use the current interface and record whether search was used rather than relying on a permanently fixed button sequence. If a relevant setting or model identifier is not visible, mark it unknown instead of guessing.
Record the date, exact prompt, conversation state and any location explicitly supplied. Note whether the response displayed citations or search-related information. Account settings and available features can differ. Do not claim that a manual response in your account represents every ChatGPT user, subscription, location or future run of the same question.
Avoid entering private client information into a public test unless you have appropriate authority and understand the relevant data handling. A local service and area usually provide enough context for a basic discovery sample. You do not need personal customer records, confidential revenue information or internal business documents to ask a general provider question.
Save the answer before interpreting it
Keep the exact response or an appropriate screenshot, along with visible source links and the prompt. A summary alone can lose important qualifications. If the answer lists several providers, record the actual wording around the target business. A business may be named as an example, described as unsuitable or included in a recommendation; those are not equivalent outcomes.
Check entity identity. Similar names, franchise locations and old trading names can create false matches. Compare the website, locality and description where available. If you cannot confidently identify the business, record an ambiguous match. Do not count a similarly named organisation as evidence of visibility merely because an automated string match finds the same words.
| Observation | What it supports | What it does not support |
|---|---|---|
| Business name appears | A mention in this response | A positive recommendation |
| Business is suggested for the need | A recommendation in this response | A permanent position or endorsement by the provider |
| Business website is cited | A displayed source relationship | Proof that the source caused every statement |
| Business is absent | No observed mention in this sample | Absence across all prompts and users |
| Details are inaccurate | A specific factual issue to investigate | A known cause without further evidence |
| Similar name appears | An identity ambiguity | A confirmed result for the target business |
Inspect sources without inventing a causal story
Open visible source links and check whether they support the statements being discussed. Record broken links, outdated information and mismatched entities. A source may help explain what information was available, but the answer does not necessarily disclose every influence on generation. Avoid saying that one directory or page "made ChatGPT recommend" a business solely because it appeared among the citations.
If the response contains inaccurate information, compare it with authoritative business-controlled information and other relevant sources. Identify the actual discrepancy and who can correct the source. Do not assume that editing one page will immediately alter future answers. Source correction is a practical action; response changes remain something to observe afterward.
OpenAI's search documentation discusses inclusion considerations, including access for its search crawler, while making clear that placement cannot be guaranteed. Technical accessibility can be relevant to investigation. It is not a promise that a crawlable page will be cited, recommended or preferred over other sources in a particular response.
Repeat a defined sample and report its limits
Use the same prompt set when comparing observations over time, and record changes in context or method. If you add new questions, distinguish the expanded sample from the original baseline. Report counts with their denominator: for example, mentions within the particular prompts checked. Do not rename that proportion as market share or an all-platform visibility score without a defensible method.
Variation is part of the measurement problem. A repeated question can produce a different answer, and the underlying product or information environment can change. Preserve individual observations so the summary can be examined. A single favourable or unfavourable run should not carry the entire recommendation for a paid service.
Use a concise conclusion: what was observed, what remains unknown and what should be investigated next. The AI visibility checker guide explains how a productised check fits this broader concept. Do not assume that an automated checker and a manual ChatGPT session use identical providers, prompts, contexts or methods; compare their documented scope before comparing results.
Turn the finding into a bounded action
If the business is absent, first ask whether the prompts represent useful customer needs. Then review identity, accurate service information, accessibility and the sources actually observed. The post-audit action guide separates changes a client can control from outcomes an agency can only monitor. Absence alone does not identify the remedy.
For agency prospecting, share enough method for the owner to assess relevance. Fusion44 for agencies can provide an AI visibility lead-capture offer, but the commercial conversation still needs qualification and explanation. Do not claim exhaustive provider coverage or automatic remediation unless the current product documentation explicitly supports the capability being offered.
Frequently asked questions
Can I check my own business by asking ChatGPT its name?
Yes, as an identity or information check. It can help you inspect how the business is described. It does not test whether a customer who has never heard of the business would discover it. Keep that branded lookup separate from unbranded service questions and label the purpose of each observation in any report you share.
Why does ChatGPT recommend my competitors but not my business?
A response alone usually does not establish the exact cause. Start by checking prompt relevance, location context, entity identity and the sources displayed. Inspect the accuracy and accessibility of useful business information. Avoid assuming a single missing field or tactic explains the difference. The appropriate next step is an evidence-based investigation, not a guaranteed-placement promise.
Is a mention the same as a recommendation?
No. A business can be mentioned as background, a comparison or even an unsuitable option. Read the wording around the name and classify the response accordingly. A citation is another separate observation: it identifies a displayed source relationship. Keeping these categories distinct makes the report more useful than a single score that treats every name occurrence as success.
How many prompts should an agency test?
Use enough to cover the distinct customer needs in the agreed scope, with a workload you can review carefully. There is no universal number that represents an entire market. A small, relevant and documented set is more interpretable than many arbitrary variations. Expand when a new intent matters, and keep the denominator visible in any summary.
Can I compare this result with a Google Maps rank?
You can discuss both as different observations of discovery, but they are not the same metric. A Maps position belongs to a local result method and location context. A conversational answer involves prompts, wording and potentially cited sources. Do not combine them into a supposedly universal rank or assume that improvement in one guarantees improvement in the other.
Will correcting my website guarantee future recommendations?
No. Accurate, accessible information can be worthwhile for customers and may be relevant to search discovery, but it does not guarantee a particular generated answer. Define the correction as the deliverable and monitor later observations separately. An agency should explain the limits before selling the work rather than adding them only after an expected recommendation fails to appear.