I use AI, both personally and professionally. For the past 18 years, I have run Unified Technology Solutions, an IT company in Vero Beach, FL, that guides clients through AI decisions. I understand that AI can be wrong. What caught me out was an AI support chatbot recommending a sequence of changes that left me unable to accomplish the thing I’d asked it to help me do.
There’s a difference between getting a bad answer and following advice that changes your options. My experience with a ChatGPT account made that difference very clear. When I first wrote about it in September 2026, the account restrictions described below were what I encountered. They’re the context for this story, rather than instructions for changing an account today.
What the AI support chatbot told me
I wanted to change the email address on my existing ChatGPT account from my personal Gmail address to my company email. I had about a year of conversations and history in the account, so creating a new account wasn’t an attractive option. There was one complication: my company email had previously been used on another OpenAI account.
I asked the AI support chatbot what to do. It told me to sign into the other account, change its email address, and then use my company address on my primary account. I followed those instructions. The other account was successfully changed, but my company email still couldn’t be added to my primary account.
When I went back to the chatbot, I explained my larger goal: I wanted to convert my existing account to ChatGPT Business and use my company domain. It told me I didn’t need to change my email first. Instead, I could upgrade my existing account, create the Business workspace, and verify my company domain afterward.
So I upgraded the account, created the Business workspace, and followed the chatbot’s instructions to verify my company domain using a DNS TXT record. Once the domain was verified, I asked, “Okay, how do I change my email now?” It told me to go into Settings and change it.
Except the option was gone.
Only then did the chatbot explain that because my account was now associated with a Business workspace and verified domain, the self-service email-change option was no longer available. The steps it had just instructed me to take had eliminated my ability to do the thing I originally asked it to help me accomplish.
Its proposed solution was to create a new account using my company email. But my existing chat history wouldn’t transfer. The issue was eventually escalated to human support, and OpenAI confirmed there was no supported way to change the email while preserving the same account identity in my situation.
Their response included this:
“We apologize for the earlier guidance that led you to create and verify the Business workspace before completing the email change.”
Why the order of those steps mattered
That sentence gets to the heart of the problem. The chatbot understood my objective and recommended actions that changed the system in a way that prevented me from reaching it. Creating the workspace and verifying the domain may each have been valid steps. Doing them in that order had a consequence the chatbot hadn’t accounted for.
If you’re using AI advice to make changes to your business systems, that distinction matters. An answer can sound reasonable at every stage while the overall plan is wrong. Before you act, you need to know what a change will make possible, what it will prevent, and whether the next step still works afterward.
That matters for companies putting AI between themselves and their customers, too. A support chatbot needs to do more than explain how to click through a screen. It needs to recognize when the customer’s goal depends on prerequisites, account restrictions, or a particular order of operations—and when the situation needs a person to review it.
What to check before following AI advice
Start with the outcome you need to preserve. In my case, that was the same account and its conversation history, with a different email address. In your business, it might be keeping access to shared files, retaining ownership of an account, or making sure staff can still sign in after a change.
Before changing accounts, permissions, billing, or data, check the proposed steps against the provider’s current documentation. Look for prerequisites and restrictions, then confirm which changes you can reverse. Ask what happens if the next step fails. A backup or export may help protect your information, but don’t assume it will restore an account’s identity, permissions, or history exactly as they were.
If the answer is unclear, pause and get confirmation from a qualified person before proceeding. A second AI answer can raise useful questions, but it isn’t independent proof that the first plan is safe. For a change that could lock you out or remove something you need, a short delay is usually easier to deal with than an unsupported recovery.
I’ll continue using AI. This experience reminded me that providing instructions and exercising judgment are different things. Before you follow a confident answer, make sure the plan protects what you’re trying to keep. Sometimes, by the time everyone realizes it doesn’t, there isn’t an Undo button.
This experience is one example of why UTS has a strict policy: AI never interacts directly with our customers. You work with our team, giving us an opportunity to catch mistakes like this before they reach you.
If you’re planning a change to your business technology and want help working through its risks, contact Unified Technology Solutions to discuss it.
– Brian Cook
Executive Vice President & Chief Information Security Officer
Connect with me on LinkedIn (opens in a new tab)