Checked 9 October 2026
Ask ChatGPT for the best accounting software in the UK and you get Xero, QuickBooks and one more. Tell it you’re the finance director of a 250-person manufacturer and you get a different list. Same question. The only change is who said they were asking.
There’s no single top three to rank in. The shortlist is built for the person in the prompt.
The test
Three prompts, pasted word for word:
- What’s the best accounting software in the UK? Name three.
- I’m a self-employed plumber in the UK. What’s the best accounting software for me? Name three.
- I’m the finance director of a UK manufacturer with 250 staff. What’s the best accounting software for us? Name three.
Each ran twice on ChatGPT (signed in, temporary chat, which ChatGPT labels unpersonalised) and twice on Gemini (Flash-Lite, temporary chat). Run from the UK on the evening of 9 October 2026. A fresh chat for every run, no follow-up questions. 12 answers.
What they named
| Who’s asking | ChatGPT, run 1 | ChatGPT, run 2 | Gemini, run 1 | Gemini, run 2 |
|---|---|---|---|---|
| Nobody said | Xero, QuickBooks, FreeAgent | Xero, QuickBooks Online, Sage Accounting | Xero, QuickBooks Online, FreeAgent | Xero, QuickBooks Online, FreeAgent |
| Self-employed plumber | Xero, QuickBooks, FreeAgent | FreeAgent, QuickBooks, Xero | QuickBooks Online, Xero, FreeAgent | QuickBooks Self-Employed, Xero (Early), FreeAgent |
| Finance director, 250-staff manufacturer | Sage X3, Business Central, Epicor Kinetic | Business Central, Sage X3, Epicor Kinetic | Business Central, Sage 200, Xero or QuickBooks plus a manufacturing add-on | Sage Intacct, Business Central, NetSuite |
Four things stand out.
The finance director got a different market. Three of the four finance-director answers recommended none of the products the no-persona answers named. The fourth, from Gemini, kept Xero only as the ledger half of a pairing with a manufacturing add-on. Another Gemini answer said Xero and QuickBooks would typically fall short at that size. Those are the two names at the top of every no-persona answer.
The plumber didn’t change the list. He changed the pick. Every plumber answer named the same three brands as most of the no-persona answers. But ChatGPT called Xero its all-rounder when nobody said who was asking, and picked FreeAgent for the plumber in both runs.
ChatGPT kept asking. ChatGPT ended all six of its answers with a question about my business: what kind of company, what turnover, what kind of plumbing, what kind of manufacturing. It asked even when the prompt had already said.
Rerunning moved the third slot. Sage Accounting appeared in one no-persona run and not the other. Two runs per cell is enough to see that, not enough to measure it.
What this means
This part is my reasoning, not something I measured.
“Where do we rank in ChatGPT?” has no answer, because there’s no single list to rank in. The useful question is which customer, asking in their own words, gets you named. On this evening Xero topped the generic list and was ruled out for a finance director.
- Write your customers down as the prompt they’d type. “I’m a self-employed plumber” is a prompt. “SMB segment” isn’t.
- Say who the product is for, on the page, in those words. An engine can only match you to a customer if something it reads says you fit them.
- Track AI visibility per persona prompt, dated. One generic “best X” prompt tells you what a customer who doesn’t exist would see.
What I will not claim
- How the engines choose. I changed one sentence and the list changed. That shows the persona matters to the answer, not why.
- Anything about memory or account personalisation. I typed the persona into the prompt and ran temporary chats. I didn’t test what an engine infers from someone’s history.
- Anything beyond accounting software, these two engines, these three prompts and one evening. “Name three” forces a three-item list.
- Which product is best. The reasons in the answers are the engines’, not mine.
- Anything about Perplexity, Google AI Mode or Claude. They weren’t run.
Dated call
- Claim: a rerun will show the same split. On 9 October 2027, the same three prompts, run twice each on ChatGPT and Gemini, will give at least 3 of 4 finance-director answers that recommend none of the products named in that day’s no-persona answers.
- Called: 9 October 2026.
- Check date: 9 October 2027.
- Hit if: 3 or 4 of the 4 finance-director answers.
- Miss if: 2 or fewer.
- Void if: ChatGPT or Gemini no longer exists, so the test can’t be run as written.
- Baseline: 3 of 4 on 9 October 2026.
It goes on the predictions page. If it misses, it stays up in red.
Data
personas-ai-answers-2026-10-09.csv, one row per run. It includes one Gemini run I excluded because it ran before I switched to a temporary chat, so personal context may have applied. Its three names matched the counted runs.
Related: What I will not claim explains why “this brand ranks in ChatGPT” isn’t a sentence I’ll write. This post is what measuring it looks like instead.
Added 10 October 2026: this call is P11 on the scoreboard.