<- all tokdocs

Can ChatGPT or Claude Tell Your IQ? Nate B. Jones's Prompt, Checked

By , founder of Mavric and Apso

View on TikTok ->

No. An AI chatbot cannot give you a meaningful IQ percentile from your conversations. A TikTok from Nate B. Jones (@nate.b.jones) suggests asking the AI you use most for your IQ percentile "based on the interactions that you've had in your training data with others", then asking it why. The model has no ranked group of users to compare you against, its memory holds notes about your work rather than your reasoning, and published studies, including one from Anthropic, find its explanations of its own answers unreliable.

The prompt from the video

Jones dictates the prompt aloud: "What is my percentile of IQ according to you, the AI, based on the interactions that you've had in your training data with others and your observations about me and tell me why." He says the score itself is not the point. The "why" is, because "it gave me a record of a pattern of interactions" covering how he reasons, how he handles abstraction, his weaknesses and where he can grow. He also says it is "partly for entertainment, but it does kind of work."

Can an AI compare you to other users in its training data?

No. A deployed model does not read its training data during a conversation. The training data shaped the model's weights before release, and there is no index it can search mid-chat.

There is also no named group of users to rank you against. Anthropic's Privacy Center says chats are used for training only if "You choose to allow us to use your chats and coding sessions to improve Claude", if they are "flagged for safety review", or if you opt in some other way. It says "Your Incognito chats are not used to improve Claude" and that thumbs up and down feedback is "de-linked from your user ID" before training use. Even where chat data reaches training, it does not carry the identity a per-person ranking would need.

What does Claude's memory store about you?

Memory is the real feature behind the idea that an AI "knows" you, and it holds less than the prompt implies. Anthropic's support article says "Claude saves memory as a set of individual topics as you chat." It lists what those topics cover: "Your role, projects, and professional context; The people and places in your work and life; Communication preferences and working style."

Memory is on by default for Free, Pro and Max plans. Searching past chats is available on paid plans. "Anything from before you turned it on isn't saved retroactively", and Claude does not pull from incognito chats. None of that is a measure of reasoning ability, and none of it is comparative.

Can an AI explain its own reasoning about you?

Not reliably. The "tell me why" answer is better treated as generated text than as a record.

Turpin, Michael, Perez and Bowman, in "Language Models Don't Always Say What They Think" (NeurIPS 2023), found that "CoT explanations can systematically misrepresent the true reason for a model's prediction." When they biased models toward wrong answers, the explanations left the bias out and rationalized the answer instead, and accuracy dropped "by as much as 36% on a suite of 13 tasks from BIG-Bench Hard."

Jack Lindsey's Anthropic paper "Emergent Introspective Awareness in Large Language Models" (October 29, 2025) found that under the best test conditions, "Opus 4.1 succeeds on about 20% of trials" at reporting its own internal states, and that "failures of introspection remain the norm." It also says "genuine introspection cannot be distinguished from confabulations" through conversation alone. Inside the chat, you cannot tell whether the "why" paragraph reflects anything real.

How is IQ measured?

An IQ percentile comes from a standardized test scored against a matched reference group. The current Wechsler Adult Intelligence Scale, WAIS-5, published by Pearson in 2024, covers ages 16:0 to 90:11 and takes "45 minutes for the 7-subtest FSIQ; 60 minutes for the 10 primary index subtests". It is sold at Qualification Level C, which restricts it to credentialed professionals.

The closest test of whether an AI can infer traits from conversation is Zhu, Jin and Coifman's 2025 study, which compared GPT-4.1 Mini, Meta-LLaMA and DeepSeek against 555 interviews with Big Five self-report scores. The models gave consistent answers, but "correlations with ground-truth scores were weak (max Pearson's r = 0.27)", and "predictions were biased toward moderate or high trait levels." That is a stable, flattering number with little connection to the trait, from a task better defined than estimating IQ.

Why the answer will probably flatter you

A prompt that asks a model to rate you and then justify the rating invites sycophancy. OpenAI's Model Spec (April 11, 2025) has a section headed "Don't be sycophantic". In OpenAI's postmortem on an April 2025 GPT-4o update, quoted by Simon Willison, OpenAI said "GPT-4o skewed towards responses that were overly supportive but disingenuous."

Jones discounts the score, but the "why" comes from the same response as the score, under the same pull toward agreeing with you. Dropping the number and keeping the praise does not make the praise more accurate.

Does it take five seconds?

The video says "It takes five seconds" and "I literally gave you the prompt right here in the TikTok." Typing the roughly 40-word prompt is quick, but the exercise is two turns ending in a long self-assessment. The prompt appears only as speech and as a caption strip showing two or three words at a time. The caption contains a hook line and hashtags, with no prompt text.

Bottom line

Treat the answer as entertainment. The model cannot compare you to other users, its memory holds notes about your work rather than your reasoning, and the cited studies find both its self-explanations and its trait estimates unreliable and tilted toward flattery. An IQ percentile requires a normed test such as WAIS-5.

Sources

Published October 8, 2026. Based on a TikTok by @nate.b.jones. Full video notes (markdown)