A prompt is circulating again across social media, and it promises something hard to resist: paste it into ChatGPT, and the model will analyze your conversation history to produce a detailed estimate of your IQ. Not just a number, but a full “forensic psychological analysis” complete with cognitive scores, named clinical instruments, personality vulnerabilities, and a predictive trajectory of your intellectual future.
The output looks impressive. It is also, in the way that matters most, fiction. Understanding why is a useful lesson in what large language models actually do.
What the prompt asks for
The viral prompt instructs ChatGPT to produce a structured dossier. It asks for a deep psychological profile covering behavioral strategies, blind spots, and internal conflicts. It asks for cognitive scores mapped onto real psychometric frameworks: WAIS-IV, WJ-IV, and the CHC model. It asks for a global IQ estimate, a domain-by-domain breakdown in table form, and a comparison against population averages. It asks for three “cognitive pillars” and a predictive trajectory spanning two to ten years, including an explicitly labeled “critical scenario.”
The formatting instructions are precise: hierarchical headings, tables, charts, bullet points, an operational report ready to use immediately.
ChatGPT will follow these instructions faithfully. That is the problem.
Why the output is generated, not measured
A large language model does not have an assessment engine. When you ask it for a “forensic psychological analysis with WAIS-IV scores,” it does not administer the WAIS-IV. It cannot. The WAIS-IV is a standardized, timed, supervised instrument delivered by a trained professional under controlled conditions, and none of those conditions exist inside a chat window.
What the model does instead is generate text that looks like the document you requested. It has been trained on a vast quantity of psychology writing, clinical reports, and test descriptions. It knows the vocabulary. It knows the structure. It knows that a document of this type contains a number somewhere around 100, often a little higher, presented with confidence. So it produces one.
The IQ figure it returns is not a measurement of you. It is the model’s prediction of what number would plausibly appear in a document of that genre, given the tone of your conversation. The cited instruments are name-dropped, not run. The tables are formatting, not data.
This is the core mechanic worth internalizing. An LLM is exceptionally good at reproducing the form of expert authority: the headings, the citations, the confident clinical register. The form is not evidence that any assessment happened underneath it.
Why the scientific costume makes it more dangerous, not less
An online horoscope is easy to dismiss because it looks like a horoscope. This prompt is harder to dismiss because the output looks like a clinical report. It has tables. It cites WAIS-IV. It assigns numbers. The scientific costume is doing persuasive work that the content has not earned.
Two things make the trend worth a warning rather than a shrug.
The first is the Barnum effect. The descriptions a model generates in this format are general enough to feel personal to almost anyone. A line like “you have a structured intellect but experience cognitive overload under pressure” applies to a large share of the population. Read as a personalized verdict, it feels uncannily accurate. It is not.
The second is the psychological profiling itself. The prompt explicitly asks the model to identify your vulnerabilities, your internal conflicts, and a “critical scenario” for your life over the next decade. An AI delivering a confident, authoritative-sounding account of someone’s psychological weaknesses and a dark future projection is not a harmless party trick. For anyone already anxious about their abilities or their future, a fabricated verdict wrapped in clinical language can land hard. The model has no basis for any of it, and it will not say so unless you ask.
A real cognitive or psychological assessment is a regulated professional process for a reason. The controlled conditions are the entire point, and a chatbot cannot reproduce a single one of them.
What ChatGPT can legitimately do here
This is not an argument that the model is useless for self-reflection. Asked plainly, ChatGPT can offer a genuinely interesting read on how you communicate: the patterns in how you build an argument, the way you tend to approach a problem in a given exchange, the rhetorical habits visible in your writing. Treated as a mirror for reflection rather than a verdict, that can be worthwhile.
The line is the claim of measurement. “Here is how you tend to frame problems in our conversation” is a reasonable observation. “Here is your IQ, scored against WAIS-IV, with a critical ten-year scenario” is a fabrication in a lab coat.
If you run the prompt out of curiosity, run it knowing that. The number is not yours. It was never measured. It was written, the same way the model writes everything else, by predicting what plausibly comes next.
