FIELD NOTES / AI, SEARCH & DISCOVERY
Can ChatGPT Actually Find Your Business?
When someone asks ChatGPT for a plumber, an accountant or a place to order a cake, can it find useful, accurate information about your business? The answer is more complicated than typing in your company name.
NU SIGNATURE · 12 OCTOBER 2026 · 10 MIN READ
Try the easiest possible test. Open ChatGPT, type in the name of your business and ask what it does. If the answer is broadly correct, that feels reassuring.
It is also the least interesting version of the question.
A new customer who already knows your name has largely solved the discovery problem. The commercially useful test is what happens when someone describes the problem they need solved, where they need it solved and any practical constraints, without mentioning your business at all. That is closer to the moment where search, maps, local listings, reviews and AI-assisted discovery begin to compete for attention.
Finding you is not one test
For a local or service-led business, there are at least four different questions hiding inside “Can ChatGPT find us?” They should be tested separately.
01 / IDENTIFICATION
Can it identify the right business?
If somebody asks for your company by name, can the answer resolve the correct official website, location and business rather than a namesake, directory listing or stale profile? This is a basic entity problem. A correct match matters, but it says little about whether you will appear for an unbranded customer need.
02 / DISCOVERY
Can it surface you when your name is not in the prompt?
This is the harder and more useful question. A prospective customer might ask for a plumber serving George, a payroll provider in Edgemead or a bakery that can handle a celebration cake. If your business genuinely fits the request, is there enough clear public information for a search-assisted answer to consider it?
03 / ACCURACY
Can it describe you without inventing the details?
Being mentioned is not enough. Service areas, opening hours, exclusions, customer fit and contact details need to survive the journey into the answer. A confident recommendation based on the wrong information is not useful visibility.
04 / NEXT STEP
Can it show the customer what to do next?
A useful answer should lead somewhere dependable: the official website, the right phone number, a quotation route, a booking page or another genuine action. Discovery and transaction are separate things. An ordinary contact link is not proof that a business can be booked or paid inside an AI conversation.
RESEARCH NOTE / 5 OCTOBER 2026
We designed the test. We did not manufacture the result.
Our research protocol uses five unbranded service-and-place prompts and four named-business checks, repeated three times in fresh chats across two days. That produces 27 planned consumer responses, with the exact prompt, account state, search behaviour, citations and retrieval date recorded for each run.
A logged-out feasibility attempt on 5 October was stopped by ChatGPT's sign-in requirement before an answer was produced. We did not treat that blocker as a failed business result, and we did not rerun the test until a more interesting answer appeared. So this article does not claim a recommendation rate, ranking pattern or “we tested 27 businesses” result. It publishes the parts we can substantiate now: the discovery framework, official platform guidance and dated public-page examples.
Search access and model training are different controls
One of the easiest mistakes in AI-search discussions is to treat every OpenAI crawler as though it does the same job. OpenAI separates OAI-SearchBot, used for search discovery, from GPTBot, which relates to model training. ChatGPT-User is associated with user-triggered visits to pages.
That distinction matters. A business can make a deliberate choice about search access without making the same choice about training. Allowing search crawling can make content eligible to be discovered; it does not guarantee that the business will be selected for a particular customer question.
The same restraint matters elsewhere. Google's current guidance for its generative search features still points back to familiar SEO foundations: accessible pages, useful information and structured data that accurately reflects visible content. It does not ask businesses to create a magical AI text file or special schema solely for AI results. There is no disclosed recipe that turns one technical tweak into a recommendation.
The useful details are often very ordinary
To make the problem concrete, we reviewed four South African public-business examples across different sectors. They were selected for editorial variety, not as a representative national sample or a league table. The observations below describe what their public pages stated when reviewed on 5 October 2026.
A location is not the same thing as a service area
AGH Plumbers' public contact information distinguished its Mayville base from the wider Durban region it serves, and separated normal office telephone numbers from an emergency contact route. That is useful because an address, a coverage area and an urgent-response channel are three different facts. A system trying to answer “Who can help with a burst pipe in Durban?” needs those distinctions to be explicit. We did not test whether published emergency availability was fulfilled.
Hours, closures and the job itself all matter
Westees Bakery & Cafe published a North End address, weekday and Saturday hours, Sunday and public-holiday closure information, and several contact options. Its public content also described celebration cakes and catering. A generic search for a bakery and a request for event catering are not the same customer problem. The more precisely the business explains what it does, when it is available and how to ask, the less a search-assisted answer has to guess.
Customer fit can be more valuable than a generic service label
Pretorius & Associates described accounting, bookkeeping, payroll and tax services, while explicitly referring to small-to-medium entities and sole proprietors. It also located the practice in Edgemead. That extra specificity gives a source-grounded answer something useful to work with when the prompt is “I run a small business nearby and need monthly bookkeeping and payroll support.” We did not independently verify professional qualifications, capacity or response times.
Sometimes the correct result is not to recommend you
BookFish is the useful negative control in the research. Its public page described an online seller of new and used books and listed areas of interest, but explicitly excluded textbooks. If someone asks where to sell second-hand school textbooks, recommending the business would conflict with its own published information. That is an important reminder: visibility is not the only goal. Accuracy includes knowing when a business is not a fit.
These examples do not prove that ChatGPT recommends the businesses, how often it might do so, or that any particular technical feature caused visibility. They show something more basic and more defensible: the public web contains facts that a sourced answer can either use correctly, contradict or fail to resolve.
The boring work is still the useful work
There is a temptation to make AI discovery sound like a new technical discipline with a new collection of hacks. Most businesses should resist that temptation.
Google's local guidance still centres relevance, distance and prominence, and recommends complete, accurate business information. That does not mean a strong Google Business Profile automatically causes a ChatGPT recommendation. It does mean that consistent, specific local information remains useful wherever customers and systems are trying to understand what a business is, where it operates and whether it fits the request.
The priority is not to write for robots. It is to remove avoidable ambiguity for people and machines at the same time.
PRACTICAL PRIORITIES
Make the business easier to verify, not easier to hype.
1. State your real service area clearly. Do not make a visitor infer coverage from the office address.
2. Describe specific services and customer fit. “Professional solutions” gives nobody much to work with.
3. Keep opening hours, exclusions, contact details and practical limits current wherever customers may check them.
4. Maintain authoritative local profiles so the website, maps and major listings agree on the basic facts.
5. Review genuine crawl barriers. A page that a search system cannot access cannot be cited from that page.
6. Use appropriate structured data because it accurately describes the visible page, not because somebody promised it would unlock an AI ranking.
A simple way to test your own business
You do not need a giant benchmark to learn something useful. Start with one controlled diagnostic and record what happened.
1. Run the named-business check. Ask for the official website, what the business offers, where it operates and the official contact route.
2. Run the unbranded need. Describe a real customer problem, the city or service area and an important constraint without mentioning your company name.
3. Check whether live web search was actually used. Open the citations. A fluent answer without inspectable sources is not evidence of current hours, availability or coverage.
4. Compare each concrete claim with the official page or listing. Mark it supported, contradicted or unresolved rather than simply “good” or “bad”.
5. Repeat the same wording in fresh chats before drawing conclusions. Record every run, including omissions and failures. Do not rerun until the answer tells the story you wanted.
If you take the test seriously, keep the date and time, ChatGPT product and displayed model, account state, search mode, exact prompt, answer and cited URLs. Order can be recorded as observed, but it should not be presented as a permanent ranking. One omission is not proof that the business is “absent from ChatGPT”.
Visibility, visits and leads are three different things
There is another measurement trap waiting here. A visit from ChatGPT is not the number of times your business appeared in an answer. An appearance is not an enquiry. An enquiry is not a sale.
If you improve your service pages, business information or local profiles, keep a dated before-and-after record and repeat the same prompts under similar conditions. Where analytics permits it, review referral visits. Separately track meaningful enquiries and actual sales. Do not collapse those stages into one impressive-looking AI visibility number.
The same rule applies to technical work. Schema, crawler access and emerging agent interfaces can all be useful in the right context. None should be presented as a guaranteed recommendation mechanism or a substitute for a business that is clearly described and genuinely relevant to the customer's need.
So, can ChatGPT actually find your business?
The honest answer is that “find” is doing too much work in that sentence.
ChatGPT may be able to identify a business by name yet never surface it for an unbranded need. It may mention the right company but repeat an outdated detail. It may understand the service perfectly and still give the customer no useful route to act. Those are different problems, and they need different fixes.
When someone asks for exactly what you sell in exactly the place you serve, the goal is not to make an AI system like you. It is to make the public facts of your business clear, current, specific and easy to act on. That is good search work. It is good website work. Increasingly, it is also good AI-discovery work.
A CONNECTED VIEW
Can a customer understand enough to choose you?
Nu Signature reviews the complete digital journey: what customers can discover, what the website explains, where confidence breaks down and whether the next step is genuinely clear. The aim is not an AI score. It is a better path from discovery to enquiry.
Explore the Customer Journey ReviewSources and research notes
Research reviewed 5 October 2026. The public-business examples are dated observations of published information, not endorsements or service-performance tests. The planned consumer ChatGPT experiment had not produced a completed response at the time of writing, so no recommendation-rate, ranking or conversion claims are made.
Google Search Central: AI features and your website
Google Business Profile Help: local ranking
OpenAI Plugins: Local Services Get Quote conversion specification
OpenAI Commerce: Agentic Commerce key concepts
Public example: AGH Plumbers contact page
Public example: Westees Bakery & Cafe contact page