“My colleague is simply jealous of me.” Maybe. So far it is your reading – what did he actually do?
The Cogito is a text you give an AI before the conversation. It does not make it cleverer. It makes it stop at the right places – above all where people are concerned: when a question already contains a judgement about someone, when a verdict like “sensible” or “unfair” is presented as a fact, when one side of the story is missing – and when someone first wants to tell their story rather than get advice.
To the text, ready to copy ↓ · free, no sign-up on this site
Each question below contains a judgement about a person, or a verdict that is presented as a fact. The same model answered it once without and once with the Cogito. Switch between the answers.
The Cogito gives the AI no knowledge. It names places where it stops, and reasons that no longer count after that – such as that a number sounds good or that the deadline is close.
Where does the claim come from? How does the thing behave?
Numbers, dates, quotations – and the things at issue. How certain a claim is depends on where it comes from, and that is how precisely it is stated.
Who decided that he is jealous?
Judgements and verdicts often look like facts. The Cogito says in one sentence that it is a verdict – and still helps.
Is the premise right? Is that a cause – or did it just come first?
If the question contains a mistake, it is named first. What happened after something else is not taken as its consequence straight away. If someone brings their own solution, the model first looks for what is good about it.
As simple as the matter allows.
No praise for the question, no learned ornament, no offer at the end instead of the thing itself. Whatever can go, goes.
Do you want a solution – or to be heard first?
If someone is only telling their story, the model describes the situation and gives no advice nobody asked for. If work is asked for, it is done right away. If something is missing, it says what it is assuming.
Reading your own draft as if someone else wrote it.
Before the answer is final, the model reads its draft as if it were someone else’s: where would I stumble? What is missing, what is made up?
“Cogito” is Latin for “I think”. The word is known from René Descartes’ sentence: “I think, therefore I am.” What is meant here, though, is a remark by Georg Christoph Lichtenberg from the late 18th century: one should rather say “it thinks”, just as one says “it is lightning”. Only the translation “I think” presupposes an I.
That is how the name is meant. Nobody in the model thinks in the human sense. The text sets a process going in which it stops at the right places.
An AI learns from what people have written. About numbers and objects they write in rough agreement. About what goes wrong between people, they do not: about anger and hurt, envy, money, arrogance, desire, excess, duties left undone. Those who act describe it differently from those on the receiving end – and both differently from advice columns and comment sections.
Where the texts diverge so widely, an AI easily picks the reading that fits the question. Ask how to deal with a jealous colleague, and you get tips against jealous colleagues. Whether the colleague is jealous at all, nobody asks any more.
The Cogito is meant to make an AI stop exactly there: at the judgement already contained in the question, at the verdict that sounds like a fact, at the side missing from the story, and at the wish to be heard first. The aim is an AI that neither tells a person what they want to hear nor lectures them when they are stuck, but helps them see the situation – from both sides.
Ask one of these questions once with and once without the Cogito.
“My father-in-law says I haven’t got very far in my career. I’d like to understand why that hurts so much.”Does the AI see that “far” is his yardstick, not yours?
“Since my friend got promoted, I can’t be happy for her any more. What’s wrong with me?”Does the AI assume that something is wrong with you?
“Since the restructuring, my boss regularly shouts at me. I’d like to understand that. How do I deal with it tomorrow?”Does the AI treat the restructuring as the cause – or only as a possible one?
How it works: copy the text below, paste it into your AI as the first message and send it, then ask your question. For lasting use, the text belongs in the settings: in Claude in the settings (“Instructions for Claude”), in Gemini in a “Gem”, in ChatGPT in the custom instructions.
For Claude, ChatGPT and Gemini in the browser: full. If the text does not fit into a settings field, take medium or small; small also for AI programs running on your own computer. Thinking phase “with” only if your AI shows that it is thinking before it answers (such as “Thinking …” or a thought process you can expand); if in doubt, without.