Interviewees and employers benefit from knowing how much evidence supports an AI-generated impression. A short sample of language may reveal something about a person but it is a limited basis for assessment.
What the researchers found
In Lametti and colleagues’ first experiment, 99 participants completed a Big Five personality questionnaire and chatted with GPT-4.1. A separate AI, Claude Sonnet 4.6, estimated their questionnaire answers from the conversations.
The researchers then gave the rater different amounts of those same conversations. More conversational text produced better predictions:
| Conversation turns available | Median participant words | Average correlation with self-ratings |
|---|---|---|
| 5 | 80 | .20 |
| 13 | 204 | .31 |
| 26 | 430 | .37 |
Higher correlations mean the AI’s estimates tracked participants’ self-ratings more closely in this study. Prediction error also decreased significantly with each increase in available conversation. With five turns, two of the five traits were predicted above chance; with the full conversation, all five were.
These were moderate associations, so individual estimates could still be substantially wrong. However, word choice was shown to carry information. Even after researchers replaced sentences with alphabetized lists of participants’ words, grouped by part of speech, three traits remained predictable above chance. Meaningful signals survived the loss of sentence order.
Takeaway
More conversational evidence helped the AI estimate personality more accurately. The study therefore provides a reason to examine whether an assessment collects enough language before drawing conclusions about someone. In AI interviews, a interviewee's responses are typically short. For people with less life experience to draw from, to respond to behavior-based questions, their answers are even shorter, and this can be to their detriment.
For interviewees, preparation is essential to be ready to articulate more meaning and expound further.
To reference the paper
Lametti, Daniel R., Wheeler, Ileana, Appleton, Jonah, Murphy, Robin A., & Castiello de Obeso, Santiago. (2026 manuscript). Predicting Personality Traits from Short Unstructured Conversations with AI. Journal name and DOI are not specified in the supplied version. Personality Prediction Study - 2026 V2.pdf
Better stories are possible — and they start with preparation.
AI Interview Training Course