In May 2026, researchers at ETH Zurich published a study showing that AI can predict Big Five personality traits from ChatGPT conversation history with significantly better-than-chance accuracy. The study used 62,000 conversations from 668 users. The participants consented to the research. The process was transparent and the data was handled carefully.
The study is not the privacy story.
The privacy story is what happens with personality inference at scale when the incentives run in a different direction.
What AI companies already know about your personality
OpenAI, Anthropic, and Google train their models on conversation data. The specifics of what is retained, for how long, and for what purposes vary by company and policy — and those policies have changed over time. But the structural fact is this: the conversation history you generate while using these tools contains personality signal.
The ETH Zurich research demonstrated this clearly. Openness and extraversion can be meaningfully predicted from conversation patterns. Conscientiousness shows signal in how requests are structured. Agreeableness appears in interpersonal framing. The researchers were careful to note the limits — inference is probabilistic and requires significant data — but the direction is clear.
AI companies have both the data and the computational capacity to run personality inference at scale. Whether any given company does this in practice depends on their policies and incentives. The incentive question is the one worth thinking about.
The difference between insight and profiling
There's a meaningful distinction between two ways personality data can be used.
Insight: You understand your own personality patterns better, so you can make better decisions, improve your relationships, and develop more deliberately. The data is used for your benefit. You control it. You can delete it.
Profiling: Someone else builds a model of your personality patterns and uses it to predict your behavior, target you with content, adjust what you see, or sell access to your profile. The data is used for their benefit. You may not know it's happening.
The same underlying information — personality inferred from conversation history — can serve either purpose. The difference is who controls it, who benefits from it, and whether you consented to it.
Most people haven't thought carefully about which mode they're operating in when they use AI tools. The assumption is usually that the conversation is private or at least neutral. The ETH Zurich research suggests it isn't neutral: it's rich with personality signal that a capable model can extract.
What the ad-supported model means for personality data
The companies with the most AI conversation data are also, in many cases, the companies with the most sophisticated behavioral targeting infrastructure. The intersection of personality inference and ad targeting is not a hypothetical concern — it's the logical extension of business models that have been operating in this direction for years.
Facebook's Cambridge Analytica episode demonstrated that personality data derived from behavioral patterns — in that case, Facebook likes — could be used to build targeted persuasion systems at scale. The same capability now exists with richer data: the full text of conversations covering how you think, what worries you, what you're working toward, and what you value.
This isn't a claim that any specific company is doing this. It's an observation about the structural situation: the data exists, the inference capability exists, and the incentive to use behavioral data for targeting exists in every ad-supported business model.
What private-by-design means in practice
"Private by default" is a design philosophy, not a compliance checkbox. The meaningful version of it looks like specific, verifiable commitments rather than vague reassurances.
When Memrov talks about privacy, these are the specific commitments:
Your raw data is deleted. The conversation export file you upload is used to generate your personality reading, then permanently deleted within seven days. What remains is the derived reading — the interpretation — not the source material. This is structurally different from retaining the raw data indefinitely "for service improvement."
Your reading isn't sold or shared. Memrov doesn't sell your personality data, use it for advertising, or share it with third parties for commercial purposes. Your reading exists in your account and can be deleted.
The analysis happens for you. Memrov builds your personality reading so you can understand yourself better. The purpose of the analysis is insight — yours — not profiling for someone else's benefit.
AWS with encryption. Memrov runs on HIPAA-eligible AWS infrastructure with AWS KMS encryption. Separate encryption keys for different data categories. Not because Memrov handles medical data, but because this is what rigorous data handling looks like.
These commitments exist because the alternative — personality inference at scale without user control — is a real outcome in a world where AI conversation data is routinely retained and analyzed.
The right question to ask about any personality product
The personality testing industry has never had to think seriously about data governance, because the data it collected — questionnaire responses — was low-fidelity and not particularly useful for behavioral targeting. That changes when the data source is AI conversation history.
When evaluating any product that reads your AI conversation history or builds a personality profile from your data, the right questions are:
What happens to the raw data? Is it deleted after your reading is generated? Retained? For how long?
Who benefits from the analysis? Is the product designed to give you insight, or to give someone else a profile of you?
What are the specific privacy commitments? Not a privacy policy link — specific, verifiable commitments about what is retained, what is deleted, and what is shared.
Is the product ad-supported? If so, the business model creates structural pressure to use behavioral data for targeting, regardless of stated intentions.
The ETH Zurich research made personality inference from AI conversation history a known, documented capability. The question now isn't whether it can be done — it clearly can. The question is who does it, for whose benefit, and with what degree of transparency and user control.
Memrov is built around a simple idea: your personality data should generate insight for you, not profiles of you for someone else. Take the free personality test →