When AI Reads Your Personality: What ChatGPT Revealed About My Team
Last quarter, I ran an experiment. I fed my team’s Slack messages, email drafts, and meeting notes into ChatGPT and asked it to generate MBTI profiles for each person. The results were fascinating — and deeply flawed. Three people who had tested as INTP for years came back as ESTJ. One quiet developer was labeled “highly extroverted.” The AI was confident, but was it correct?
This isn’t just a curiosity. Managers and HR professionals are increasingly wondering whether AI tools can replace traditional personality assessments. A recent study in Frontiers in Computational Neuroscience put this question to the test, using large language models to analyze text and predict personality types. The findings reveal both promise and serious limitations.
How AI “Reads” Personality
Large language models work by analyzing patterns in text — word choice, sentence structure, emotional tone, and topic preferences. When given enough writing samples, these models can identify traits that correlate with personality frameworks like the Big Five or MBTI.
The 2026 research showed that LLMs could predict MBTI types from written text with above-chance accuracy. That’s genuinely impressive. But here’s the catch: the same study found systematic biases. The models tended to over-predict certain types (especially “judging” over “perceiving”) and showed overconfidence in their assessments. They created polarized predictions that don’t reflect real population distributions.
For my team, this meant the AI saw our formal Slack communication — structured, task-focused, deadlines-oriented — and concluded we were all high on conscientiousness. It couldn’t account for the context: corporate communication norms flatten personality expression.
What AI Gets Right About Personality
Despite the bias issues, AI-driven personality analysis has genuine strengths:
- Scalability: Analyzing hundreds of team members is impractical with traditional tests but trivial with text analysis
- Unobtrusiveness: No one needs to fill out a 60-question survey; the analysis happens passively
- Behavioral sampling: Instead of self-reported preferences, AI looks at actual language use — what people do, not what they say they do
A growing number of platforms are experimenting with these approaches. If you want to understand where your own personality sits across scientifically validated dimensions, tools like personalitree.com offer free Big Five and 16-type assessments that give you a grounded starting point before you jump into AI experiments.
The Biases You Need to Know About
Before you use AI to evaluate your team — or yourself — understand these limitations:
Context Blindness
People write differently in a work email versus a group chat versus a journal entry. AI typically trains on whatever text is available — often formal work communication — and misses the full spectrum of someone’s personality expression.
The Labeling Trap
MBTI’s binary forced-choice design means 50% of people get a different type when retaking the test just five weeks later. AI doesn’t fix this; it inherits the same flawed framework. If your AI-generated type doesn’t feel right, it might not be wrong — just reductive.
Overconfidence Illusion
The study found that AI models present personality predictions with high confidence levels even when accuracy is modest. This creates a dangerous dynamic: managers trust an authoritative-sounding AI output more than their own human judgment.
“The AI told me our lead designer was an INTJ. She’s one of the most collaborative, emotionally attuned people I know. I almost reshuffled our team structure based on that reading.” — Engineering Manager, anonymous feedback from my experiment
Better Ways to Use AI for Personality Insights
AI personality analysis isn’t useless — it just needs the right framing:
- Use it as a conversation starter, not a verdict. Share AI-generated profiles with team members and ask: “Does this resonate? What’s missing?”
- Combine frameworks. The Big Five model captures nuance that MBTI misses, including facet-level detail like “anxiety” versus “vulnerability” within neuroticism.
- Gather more data. The more diverse text samples you feed the AI (personal writing, brainstorming notes, social chat), the richer the profile.
- Validate against self-report. Have team members take a proper assessment and compare results with the AI’s analysis.
What Actually Matters
Personality frameworks — whether assessed by a human or an AI — are maps, not territories. They help you navigate differences in how people think, communicate, and recharge. But they become harmful when you mistake the map for the person.
My biggest takeaway from the experiment wasn’t about AI accuracy. It was about how quickly we want a single label to explain someone’s complexity. The developer who scored ESTJ from Slack messages was the same person who runs a D&D campaign, paints watercolors, and volunteers at an animal shelter on weekends. No four-letter code — and no AI model — captures that.
If you’re curious about where you fall on different personality dimensions, start with a reputable self-assessment rather than an AI guess. personalitree.com provides free, research-backed tests that give you a clearer picture than feeding your chat history to an LLM. Take a free test, explore your personality type, and see how it aligns — or doesn’t — with what AI might say about you.