The AI revolution has a customer experience problem.
On 10 November 2026, OpenAI opened Businesses in ChatGPT — in beta. Businesses can claim a profile, and once verified, update their details, add a product feed and see which topics they get cited for. The portal where you do that is called Business Hub.
I claimed mine in the first week. I have spent forty years in marketing, I have watched the internet arrive, then search, then digital advertising, then ecommerce, and now this. I build my consulting practice around helping businesses get found by AI systems. So a listing on the surface where my customers are starting to ask questions is not optional for me. I was going to be an early adopter whether or not it went smoothly.
It did not go smoothly. What follows is not a complaint. It is a case study in the gap between how good a technology is and how good the experience around it is — and that gap is the thing business owners keep paying for.
All I wanted to do was change one field
I registered Siamak Kalhor Consulting, went through the process, and got approved. Then I tried to change my business category to AI company.
Not build a platform. Not deploy an agent. Not integrate a database. Change one dropdown.
I found the profile editor, selected the category, clicked Save, and the window closed. I reopened it and the old category was still sitting there. I did it again. Same result. I have no way of knowing whether the change was queued for review, silently rejected, or simply lost, because nothing on the screen told me which.
Then the assistant became part of the problem
I did what anyone would do. I asked ChatGPT.
It misunderstood what I was asking for and built me a business management dashboard I had never requested. When I clarified that I meant OpenAI's own Business Hub, it started giving me navigation instructions — click this menu, open that panel — that did not match anything on my screen. I sent screenshots. It kept describing buttons that were not there, and at one point produced a generated image of an interface that does not exist.
It apologised each time. The apologies were not the problem. The problem was that it never once said I don't know what this interface looks like.
That sentence would have saved me an hour.
It is not only me
I want to be careful here, because one person's bad afternoon is an anecdote, not a finding. So I went and checked OpenAI's own developer forum before writing this.
There is a thread titled "Business Hub Persona verification failed twice, no reason shown." In it, users report identity verification failing with no stated reason and no way to retry. One says it failed five times with a national ID and a passport. The same failure appears from Ukraine, Indonesia, India and Brazil. Profiles lock in a failed state offering only "Add business" and "Provide feedback," so people create duplicates they then cannot delete. One user's support case was routed to Ads support, went unanswered for four days, and then received an automated inactivity warning.
The quote that stayed with me came from OpenAI support, as posted by one of those users: they could not yet confirm whether the business was eligible, or why the checks had failed. At the time of writing, no OpenAI staff member has replied in that thread.
So my category not saving may be its own bug, or it may be the same eligibility machinery failing quietly rather than loudly. I genuinely do not know which, and I am not going to pretend otherwise.
What is forgivable and what is not
Here is where I part company with most of the criticism I have seen.
This product was nine days old when I used it. It says beta on the label. Software that new having rough edges is not a scandal — it is how shipping works, and any founder who has launched anything knows it. If the only story here were "the save button is flaky in week one of a beta," I would not have written this.
The part that does not get a pass is different. A beta explains a broken field. It does not explain an assistant that confidently describes an interface it cannot see. Those are two separate failures and only one of them gets fixed by the next release.
Confidence without accuracy is the expensive failure
The assistant did not say it was unsure. It presented guesses in exactly the same tone it uses for things it knows. There is no signal in the output that distinguishes the two, which means the burden of knowing what is true sits entirely with me.
On a dropdown menu, that costs an hour.
Now move the same behaviour somewhere it matters. An assistant advising on an advertising budget. On which customers to stop calling. On what a contract clause means. On whether a number in a report is right. Same fluency, same confident tone, no flag on the sentence that happens to be invented. That is not a hypothetical risk for a small business — it is the ordinary way these tools get used once people trust them.
An AI assistant does not only need knowledge. It needs judgement about the edges of its knowledge, and it needs to stop repeating an approach that is visibly not working. Mine repeated the same kind of wrong answer four times while I told it the screen did not match.
What I would actually do
None of this is a reason to stay off the platform, and I want to be clear about that because the cynical read is easy and it is wrong.
- Claim your business anyway. It is free, it takes minutes, and it sits on the surface where your customers are increasingly asking which business to use. A rough beta is a weak reason to skip that.
- Screenshot what you entered. If a field does not persist, you want your own record of what you set and when. Do not rely on the system to remember on your behalf.
- Verify in a fresh session. Close it, reopen it, and check. An interface that closes a dialog after you click Save has told you nothing about whether it saved.
- Do not use the assistant as a manual for an interface it cannot see. This is the single most useful thing I took away. For navigation inside a product, go to the product's own help documentation. Ask the assistant to explain a concept, not to describe a screen.
- Do not build a process on it yet. By all means claim the listing. Do not promise a client results from a nine-day-old beta, and do not make it a dependency in anything that has to work.
The questions worth asking before you adopt any AI tool
I have been saying a version of this for most of my career, long before any of it involved AI. Businesses do not buy technology because it is impressive. They buy it because it solves a problem. Strategy matters more than technology, and outcomes matter more than features.
So before you adopt an AI system, ask the practical questions. Does it solve a real problem you actually have? Is it reliable enough to depend on? Can an ordinary employee use it without a specialist standing behind them? What happens when it goes wrong, and will anyone answer you? Does it recognise its own limitations, or does it guess with a straight face?
And the one that settles most arguments: does it save more time than it costs you?
A note to OpenAI, and to everyone else building this
I am not writing this to dismiss OpenAI. The underlying technology is genuinely remarkable and I use it every day. I am writing it because the gap between the model and the experience around the model is now the thing holding adoption back, and nobody seems to be measuring it.
If you want businesses to trust AI with operations that matter, the tools have to be dependable at the boring end — saving a field, explaining a failure, answering a support ticket, telling someone honestly that you are not sure.
Forty years in this business has taught me one thing that has not changed through any technology shift I have lived through. Technology gets attention. Reliability earns trust. And the experience around the product decides whether anyone stays.
That was true of the first websites I built. It is true of this.
Common questions
What is ChatGPT Business Hub?
Business Hub is the portal where a business claims and manages its presence inside ChatGPT. The product itself is called Businesses in ChatGPT, which OpenAI launched in beta on 10 November 2026. Once a business is verified it can update key details, add a product feed with pricing and availability, and see aggregated insights on mentions, website clicks and the topics it gets cited for. Eligible US ecommerce and online service businesses can also build a Business Agent.
Why won't my ChatGPT business profile changes save?
I cannot tell you the cause, and I want to be precise about that. In my case the category I selected did not persist between sessions, and I was not able to establish whether that was a bug, a pending review, or an eligibility limit applied silently. What I can tell you is that it is not only happening to me: OpenAI's own developer forum carries reports of verification failing with no reason given, profiles locking with no retry option, and duplicate profiles that cannot be deleted. OpenAI support told one of those users it could not yet confirm why the checks failed.
Should I still claim my business in ChatGPT?
Yes. Claim it. A rough beta is a weak reason to skip a free listing on a surface where your customers are already asking which business to use. Claim it, screenshot what you entered, and check it again in a week rather than assuming it saved. What I would not do is build a process on top of it, promise a client results from it, or treat anything in it as reliable yet.
Is Business Hub available outside the United States?
Partly, and this is the least clear part of the product. OpenAI says any eligible business in a market where ChatGPT is available can claim a profile, while Business Agents are limited to eligible US ecommerce and online service businesses. Forum reports of verification failures come from Ukraine, Indonesia, India and Brazil, and one user noted the help documentation pointing to existing US listings for local businesses. Whether non-US businesses are hitting an eligibility boundary rather than a bug has not been confirmed either way.
What is the real lesson for business owners adopting AI tools?
Separate the two failures, because they are not the same thing. Software nine days into a beta having bugs is ordinary and forgivable. An assistant confidently giving you navigation instructions for an interface it cannot see is a different problem, and it does not get fixed by the next release. Before you adopt any AI tool, ask whether it recognises the limits of what it knows. A system that says it is not sure is worth more than one that is fluent and wrong.
