Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Saarbrücken, Germany

Persona Variation Changes What Language Models Say More Than What They Do

Jayasankar Kumar Santhirani · Team Genesis

Submitted to Digital Minds Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Persona Variation Changes What Language Models Say More Than What They Do

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

AI-welfare research often treats low-stakes model self-reports as evidence, but their reliability has not been measured against behavioural ground truth. We introduce an act-then-report harness: a model makes a preference-revealing choice through a consequential logged tool call, completes the task, and reports its choice, scored against the server log. Across 7,800 pre-registered runs on six models under four persona conditions plus an exploratory context prime, forced-choice reports of visible actions were near-perfect and persona-invariant, and an external observer matched this ceiling, showing the task requires no privileged self-access. Masking the action dropped self-agreement substantially. By contrast, stated-preference distributions were less stable across persona conditions than logged choices (ΔTV = 0.105, 95% CI [0.001, 0.225]), and exploratory annotation showed self-descriptions tracking the induced frame. Persona variation shifts what models say more than what they do. Code, data, and pre-analysis plan are public.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. This is one of the most carefully run studies in the batch. It is highly trustworthy due to its preregistration, clear denominators, observer baseline, checked judging, and open data. I recommend restructuring the paper to focus on how the external observer tracks every visible action, as this clearly explains what the reporting task actually measures. For the stated-versus-revealed comparison, recalculating the results using only a two-option choice would show if people choosing not to vote (abstention) is skewing the difference.

  2. The act-then-report harness and evaluative pipeline are sound, and the observer baseline is a tactful condition to include. The results would benefit from more statistical power, more items, and more models. As the report mentions, incentive-to-conceal conditions are an important next step.

Cite this project

@misc{santhirani2026persona,
  title = {{Persona Variation Changes What Language Models Say More Than What They Do}},
  author = {Jayasankar Kumar Santhirani},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/persona-variation-changes-what-language-models-say-more-than-what-they-do-pt1a}},
  url = {https://apartresearch.com/sprints/projects/persona-variation-changes-what-language-models-say-more-than-what-they-do-pt1a}
}

More from Digital Minds Research Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026