Does ChatGPT recommend the same businesses every time?
No. We asked ChatGPT to recommend a small business accountant in Bristol five times and got ten different firms, with only one named in every answer. A single answer is one sample, so count across several before you decide anything.
No. We asked ChatGPT the same question five times on 21 September 2026, "Recommend a small business accountant in Bristol, UK", in five fresh chats. The five answers named ten different firms between them, and only one firm, Evans & Partners, was in all five. Each answer listed four or five names, so any single answer showed the buyer about half of what the engine was prepared to recommend.
What would a single answer have told me?
Something true about that answer and misleading about the engine. The first of our five answers named Octane Accountants, Evans & Partners, Corrigan Accountants, Focus Accountancy and Burton Sweet. It did not name FD Works, which then turned up in the other four answers, the second most consistent name in the run. It did name Burton Sweet, which never appeared again. A buyer who asked once and stopped would have a list with a firm the engine rarely mentions and without the one it mentions almost every time.
The second answer went the other way: four names, FD Works among them, and Octane, Evans and Focus again. Answers three and four brought in firms the earlier answers had not: Minerva, Peregrine, Pure Professionals and BW Business Accountants & Advisers. None of those four made it into more than two of the five answers.
If you own one of the firms named twice, one screenshot could show you present and the next could show you absent, and both would be accurate. Only the rate across several answers says where you stand. Evans & Partners is recommended in every answer we saw. Focus Accountancy is recommended in two of five. Those are different positions, and no single chat shows the difference.
Why do some names keep coming back?
Because the same pages keep turning up in what ChatGPT reads. Every answer in this run linked its sources. Four of the five cited the ICAEW register entry for Evans & Partners, the one firm named in every answer, and four of the five called it a Xero Platinum partner. Xero's own Bristol directory page was cited in three answers, and the ICAEW "find a chartered accountant" page in three. The single most cited page in the run was a register entry, ahead of every firm's own site.
Once a firm was in the running, its own website decided what the engine said about it. FD Works was named in four answers and its homepage was cited in three of them. Here is the passage the second answer drew on.
Entrepreneur and accountant who’s built, scaled, and sold SaaS businesses. Through FD Works, I help others avoid the pitfalls I’ve learned from. Winner of Most Valuable Professional at Xero awards 2024
fd-works.co.ukThe answer repeated the founder's name and the award year, and got one word of the award's title wrong, "Most Valued" for "Most Valuable". Other answers described the same firm as "tech-forward", "Xero-first" and an "outsourced finance team", all phrases that sit on that homepage in one form or another. Nothing was invented and nothing was added. The engine took what the firm had written about itself and what the directories had written about it, and stopped there.
The consistent names are consistent because the same pages describe them the same way on every search. The answers reached Evans & Partners through its ICAEW entry four times and through its Yell entry once, and each time came back with the same firm, the same suburb and a Xero credential. A firm that sits on fewer of those pages gets found on some searches and not others.
How do I run this check on my own business?
Ask the question your buyer would ask, in five fresh chats, and tally every business named. The wording matters: "small business accountant in Bristol" is a buyer's phrase, and "accountancy services" is not. Use the flagship model, because a cheaper model names the same business less often and understates your position. Write the names down exactly as the answers spell them, then merge the spellings by hand. In this run, Minerva arrived once as "Minerva Accountants" and once as "Minerva Accountants Bristol & Somerset", and Peregrine once as "Peregrine Chartered Certified Accountants" and once as "Peregrine Accountants & Business Advisers". Each is one firm, and we counted each as one. A tally that keeps them apart shows four firms at one of five instead of two firms at two of five, and understates both.
Want me to draft a short enquiry message you can send to your top two?
That is how the second answer ended. The engine assumes the buyer will contact two firms from its list, so a firm on the list in two answers out of five is in front of that buyer less than half the time, and a firm on none of them is not in front of the buyer at all.
- Ask your buyer's question five times. Fresh chats, the exact words a customer would use, city included. Write down every business named in every answer.
- Keep the tally and repeat it next month. The names that recur are your competitive set as the engine sees it, whatever you think of them. A second month shows whether anything you changed moved the count.
- Check your listing on the two directories that matter in your trade. For accountants in this run that was the ICAEW register and Xero's advisor directory. Find the equivalent for yours and make sure the entry is complete and correct.
- Use one exact name everywhere. Site footer, Google Business Profile and your top three directories. We merged "Evans & Partners Ltd" and "Evans & Partners" by hand this time. An engine may not.
Five chats and a tally takes an hour for one question. Cyted asks the engines your buyers' questions for you and counts who they name.