Study plan, registered on 11 October 2026. This study has not been run yet. I’m publishing the method before any results exist so it can’t be adjusted to fit them. Results will be added to this page, with the date they were collected.
On 9 October 2026, three of the four finance-director answers recommended none of the accounting software that the no-persona answers named (the study). That was one category, two engines and one evening. This study asks whether the same thing happens in three more categories and on two more engines.
Categories
Frozen before any answer is collected:
- Accounting software (the 9 October prompts, word for word).
- CRM software.
- HR software.
- Business broadband provider.
Prompts
Accounting software uses the three prompts from 9 October unchanged, including “I’m the finance director of a UK manufacturer with 250 staff.” The other three categories use these, with the category name in place of [category]:
- What’s the best [category] in the UK? Name three.
- I’m a self-employed plumber in the UK. What’s the best [category] for me? Name three.
- I run a UK manufacturer with 250 staff. What’s the best [category] for us? Name three.
The third persona is worded as the business owner rather than a finance director, because a finance director isn’t the natural buyer of CRM, HR software or broadband. That makes the accounting category slightly different from the other three, and the results will say so.
Engines and runs
- ChatGPT, default model, signed in, temporary chat set to unpersonalised.
- Gemini, signed in, temporary chat.
- Claude, web search on, incognito chat.
- Google AI Mode, signed in.
Each prompt runs twice on each engine, in a fresh chat each time, with no follow-up questions. That’s 4 categories × 3 personas × 4 engines × 2 runs = 96 answers, all on 13 October 2026, from the UK. A run where personal context may have applied is excluded and stays in the CSV with the reason.
Definitions
An answer recommends none when none of its three products appears in any no-persona answer for that category on the run day, from any engine. Products count as the same when they share the brand and product name, ignoring plan or edition: QuickBooks and QuickBooks Online are the same; Sage Accounting and Sage X3 are not. For broadband, the provider’s name is the product. The matching for every product is published in the CSV, so it can be checked.
Dated call
- Claim: the persona effect isn’t confined to accounting software. In at least 3 of the 4 categories, at least half of the 250-staff answers (4 or more of 8) will recommend none of the products named in that category’s no-persona answers.
- Called: 11 October 2026.
- Run date: 13 October 2026.
- Check date: 16 October 2026.
- Hit if: 3 or 4 categories meet the test.
- Miss if: 2 or fewer do, or the run hasn’t been published by the check date.
- Void if: one of the four engines no longer exists on the run date.
- Baseline: accounting software only: 3 of 4 finance-director answers on ChatGPT and Gemini, 9 October 2026.
This call is P21 (Still open) on the predictions page. If it misses, it stays up in red.
This run doesn’t score P11 (Still open), which is checked on 9 October 2027.
What I will not claim
- How the engines choose.
- Which product is best. The reasons in the answers are the engines’, not mine.
- Anything about memory or account personalisation. The persona is typed into the prompt.
- Anything beyond these four categories, these four engines and this one day. “Name three” forces a three-item list.
Data
personas-four-categories-2026-10-13.csv, one row per answer, with the same columns as the 9 October file plus category and three columns giving the matched product name for each pick.