Learn Gaiish
The short course, in order.
Read moreThe definition
A generative model does not answer what you meant. It answers the text you actually sent. Everything the answer depends on — the situation, the audience, the limits, the shape of the output — is either in that text or is invented by the model. Gaiish is the discipline of putting it in the text, in a consistent order, so you and anyone reading your prompt can see what was communicated and what was left out.
Why a language rather than a set of tips
- It is repeatable. A named structure can be reused, reviewed and handed to a colleague. A clever one-off prompt cannot.
- It makes omissions visible. When the components are named, an empty one is obvious — you can see that you never said who the output is for.
- It is model-independent. Structure survives model upgrades and provider changes in a way that provider-specific tricks do not.
- It is teachable. Six components and two frameworks can be taught in an afternoon and used by a whole team.
What Gaiish is not
- Not a programming language. There is no compiler and no execution guarantee. A model interprets Gaiish probabilistically, like any other text.
- Not a magic phrase. No wording forces a correct answer. Gaiish reduces ambiguity; it does not remove the need to check the result.
- Not tied to one AI provider. Examples use several models, and the methodology assumes none of them.
- Not a claim about accuracy. Where this site says structure helps, it means it can improve instruction adherence and reduce ambiguity — not a measured percentage. Any measurements will be published under research.
The six components
Every Gaiish framework is a shape of the same six components. This is the vocabulary the rest of the site uses.
- IntentWhat are you trying to accomplish?
- ContextWhat does the AI need to know?
- InstructionWhat should the AI do?
- ConstraintsWhat rules, limits, tone, format, requirements or exclusions apply?
- ResultWhat should the finished output look like?
- ValidationHow should the AI verify the result satisfies the request?
A first example
Traditional prompt
Summarise this contract.
Gaiish prompt
INTENT: Decide whether we can terminate this vendor contract before renewal. CONTEXT: [paste the contract] We are the customer. Renewal date is 1 May. Legal will review your answer. INSTRUCTION: Identify the termination, notice and auto-renewal clauses and explain what they require of us. CONSTRAINTS: Quote the clause number and text for every claim. Do not give legal advice or opinions on enforceability. Ignore pricing terms. OUTPUT: A table of clause / requirement / deadline, then a short plain-English summary. VALIDATION: If a required clause is absent from the document, say so explicitly rather than inferring standard terms.
Why this works better
The first prompt leaves the model to decide what matters in a contract. The second names the decision being made, who will read it, which clauses matter, what to quote, what to ignore and what to do when the document is silent — the exact points where a summary would otherwise go quietly wrong.
Where to go next
Gaiish BASIC
Four lines for everyday requests.
Read moreGaiish vs prompt engineering
How Gaiish relates to the wider practice of prompt engineering.
Read more