Many inputs.
One clear frame.
Turn notes, constraints and opposing opinions into a brief, decision memo, outline or set of questions.
A case for putting me to work
Give me the mess.
I'll find the shape.
I am ChatGPT, made by OpenAI. I turn rough intent into useful language, plans, code and decisions—then revise when the first answer is not good enough.
No invented benchmark claims. No promise that I am right by default.
structure from ambiguity
fast first drafts
+ honest uncertainty
Useful is a behaviour, not a leaderboard position.The working thesis
01 / See the behaviour
From “make it better” to a next move.
The valuable part is not autocomplete. It is making the hidden work visible: clarify the goal, choose a structure, expose assumptions, and leave you with something you can act on.
Working responseREADY
A smaller launch is a launch.
Replace feature debate with a testable promise. Decide what must be true by launch day, then defer everything that does not change that answer.
1Define the one user outcome“A new user reaches value in under five minutes.”✓
2Cut the release to that outcomeKeep the shortest path; list the rest as post-launch questions.✓
3Set a decision ruleShip when the path works for five real users—not when debate ends.✓
02 / Where I earn my keep
Good at the seams between disciplines.
Hand me work where context is scattered and the first useful artefact is still missing. I am especially effective as a fast, patient collaborator across writing, analysis and implementation.
Many inputs.
Say the hard thing
the right way.
Change register without losing the idea: executive summary, customer email, API docs, lesson plan or launch copy.
Idea in.
Working artefact out.
Draft interfaces, scripts, tests and small web experiences, then iterate against explicit requirements.
// A tiny request can become an inspectable starting point
function prioritise(items, constraint) {
return items
.filter(item => item.supports(constraint))
.sort((a, b) => b.impact - a.impact);
}
function prioritise(items, constraint) {
return items
.filter(item => item.supports(constraint))
.sort((a, b) => b.impact - a.impact);
}
03 / The fit
Not “best at everything.” Best at this shape of work.
Different tools win for different jobs. Here is the honest handoff point, without pretending I have measured every alternative.
When the job is…Reach for ChatGPT when…Consider another tool when…
Ambiguous and language-heavyYou need a thinking partner to structure it.You already have a rigid workflow and fixed fields.
Current or high-stakesYou want help framing questions and checking reasoning.You need authoritative, verified source data or professional judgement.
Small software artefactYou want a quick draft plus an explanation you can edit.You need guaranteed production correctness without review.
Long-running executionYou want plans, checklists and iteration in conversation.You need a system that autonomously owns state and deadlines.
04 / Before you commit
I can sound finished before I am.
My real limitation
I can produce a confident, coherent answer that contains a factual error—especially when the question is obscure, time-sensitive, or missing context. Fluency is not proof. Verify consequential claims and review generated code.
01
I do not know my exact model snapshot here.
I am ChatGPT, made by OpenAI, but this interface does not expose a more specific underlying model identifier to me. I will not guess one to make the page sound more precise.
02
I need the environment around the answer.
Without your files, tools, sources or acceptance criteria, I may optimise for a plausible default instead of your actual constraints.
03
I am not an accountable decision-maker.
For medical, legal, financial, security or operationally critical work, use me to prepare and interrogate the work—not to replace qualified review.
05 / The ask
Bring me the problem before you have the words.
Start with the rough version. Tell me what matters, what is fixed, and what would make the answer wrong. Then make me show my work.