The short answer
The best AI prompts for decision-making ask for assumptions, options, trade-offs and failure modes rather than a single verdict. Claude from Anthropic and ChatGPT from OpenAI are the strongest decision partners for framing, weighted matrices, pre-mortems and memos, and since 7 October 2026 ChatGPT answers can include adjustable charts and calculators. Use Perplexity for cited evidence and its Model Council to compare several models. The 13 prompts below cover each stage of a decision.
How AI helps with decisions, and where it should stop
AI assistants such as Claude, ChatGPT and Gemini are useful decision partners because they never get tired of asking "what are you assuming?" They can widen your options, build a decision matrix in seconds, argue the side you dislike and draft the memo afterwards. What they cannot do is own the decision. They do not know your organisation's politics, your appetite for risk or the facts you did not give them, and they can produce a confident recommendation from thin evidence. The prompts on this page are built to improve your thinking, not replace it: they ask for assumptions, trade-offs and failure modes rather than a single verdict.
The 13 prompts are grouped in the order good decisions are made: frame the decision, widen the options and set criteria, gather evidence, stress-test the favourite, and communicate the outcome. They suit business decisions such as hiring, vendors, pricing, strategy and project choices. For personal investment, legal or medical decisions, treat AI as a way to prepare questions for a qualified professional, not as the advice itself.
Prompts for framing the decision
Define the decision properly asks the model to question you first and to point out when one decision is really two, which is the most common reason teams talk past each other. Check whether this is a one-way or two-way door sets how much analysis the decision deserves; most decisions are more reversible than they feel, and a small pilot is often available. Surface the hidden assumptions ranks your assumptions by how much they matter, so you know which ones to test before anything else.
Prompts for widening options and setting criteria
Widen the options beyond yes or no is worth running on every binary choice. Adding a hybrid, a pilot, a delay and "do nothing" often reveals that the original two options were both weak. Build a weighted decision matrix asks for a sensitivity check, because a matrix whose winner flips when one weight moves by 10 points is telling you the decision is close. ChatGPT is handy here: since 7 October 2026 its answers can include charts and small interactive tools such as calculators, so you can ask for a matrix you can adjust. Name the trade-offs plainly turns each option into a sentence people can agree or disagree with, which moves a meeting forward faster than another slide.
Prompts for gathering evidence
Run Gather the evidence that matters for this decision in a tool that searches the web and cites every claim, such as Perplexity, whose free plan, according to its pricing page, is good for limited daily usage. If the evidence is in your own documents, such as board papers, customer research or past project reviews, Google's Gemini Notebook answers only from the sources you add, with citations. Separate facts from opinions in the debate is for the moment a discussion goes round in circles: sorting points into facts, forecasts, value judgements and anecdotes shows which disagreements can be settled with data and which are about priorities. Our research prompts go further on finding and checking evidence.
Prompts for stress-testing your favourite option
Run a pre-mortem imagines the decision has failed and works backwards, a technique that surfaces risks people hesitate to raise in a normal discussion. Ask for early warning signs, so the output becomes a watch list rather than a list of fears. Argue the other side as well as possible asks for the best version of the case you are leaning against, and then which argument deserves the most weight. Map the second-order effects looks six to twelve months out for each group affected. For an important decision, run these prompts in a second assistant, or use Perplexity's Model Council, which runs one question through several models and shows where they agree and disagree. Perplexity's help centre lists Model Council in Search for the Max and Enterprise Max plans (its pricing page also lists it under Pro, so check in the app); inside Perplexity Computer it is on all paid plans and uses credits.
Prompts for communicating the decision
Write a one-page decision memo records why you chose, what you gave up and what would make you revisit the decision. That last section is the one people skip, and it is the one that makes reviewing a decision a year later useful rather than defensive. Prepare for stakeholder pushback gives each group a likely question, an answer and something you can offer them. Claude and ChatGPT both write these documents well, and Claude Docs, available on all Claude plans including Free, exports to Word, PDF, Markdown and Google Docs.
How to adapt these decision prompts
- Write your own criteria and weights. The matrix is only as good as the weights, and the weights are a leadership judgement, not an AI one.
- Give the model the facts, then ask it to flag missing ones. Add "tell me what information would change your view" to any prompt that ends in a recommendation.
- Ask for the opposite before you ask for a verdict. Assistants tend to agree with the framing they are given. Running the devil's advocate prompt first reduces that pull.
- Keep the decision record. Save the framing, the matrix and the memo together in a Claude or ChatGPT Project so the reasoning is easy to find later.
- Protect confidential information. Personal plans of ChatGPT, Claude and Perplexity can use chats for training unless you switch it off, and Gemini does while Keep Activity is on. For deals, restructurings and people decisions, use your company's work plan or leave names and figures out. See how to keep your data private in AI tools.
Which AI tool works best for decision-making, and why
Claude and ChatGPT are the best general decision partners: they follow structured prompts closely, hold long context such as meeting notes and reports, and write clear memos. ChatGPT's interactive answers make it the better choice when you want a matrix or calculator you can adjust. Perplexity is the right tool for the evidence step and for comparing several models' views. Gemini Notebook is best when the decision rests on your own documents. Compare them in ChatGPT vs Claude, Claude vs Perplexity and ChatGPT vs Gemini, see the best AI for research and the best AI for project management, or pair this page with our brainstorming prompts for the stage before options are on the table. Free lessons are at /learn?tool=claude.
Prompts to copy
Define the decision properly
I need to decide [decision]. Help me frame it before I look at options. Ask me up to five questions, then write: the decision in one sentence, the deadline and why it is the deadline, who decides and who must be consulted, what success looks like in 12 months, and what is out of scope. Point out if I am actually facing two decisions disguised as one.
You get: A clear decision statement that stops the team solving the wrong problem
Check whether this is a one-way or two-way door
Decision: [decision]. Assess how reversible it is: what it would cost in money, time and reputation to undo after 3 months and after 12 months, which parts are reversible and which are not, and whether we could make a smaller, reversible version first. Recommend how much analysis this decision deserves: hours, days or weeks.
You get: A sense of how careful you need to be, and whether a smaller first step exists
Surface the hidden assumptions
Here is how I am thinking about [decision]: [paste reasoning]. List every assumption my reasoning depends on, rate each as well supported, plausible or untested, and tell me which two assumptions, if wrong, would most change the answer.
You get: The assumptions behind your thinking, ranked by how much they matter
Widen the options beyond yes or no
I am deciding whether to [option A] or [option B]. Before comparing them, give me five more options I have not considered, including a hybrid, a delayed decision, a small pilot and doing nothing. For each, one line on when it would be the best choice.
You get: A wider set of options, so you are not choosing between two weak ones
Build a weighted decision matrix
Options: [list options]. Criteria and weights out of 100: [e.g. cost 30, speed 20, risk 25, strategic fit 25]. Score each option from 1 to 5 on each criterion with a one-line reason, calculate weighted totals, and show the table. Then tell me how the ranking changes if the top-weighted criterion drops by 10 points, and which score you are least confident about.
You get: A weighted comparison with a sensitivity check on the weights
Name the trade-offs plainly
For [option A] versus [option B] in [context], write one sentence per option in the form: 'If we choose this, we get [benefit] but we give up [cost] and accept the risk of [risk].' Then tell me which trade-off our stated priority [priority] points to.
You get: The trade-offs stated in a form people can actually disagree with
Gather the evidence that matters for this decision
We are deciding [decision] in [industry and country]. Find current, reputable evidence on: [question 1], [question 2] and [question 3]. For each, give the finding, the source and its date, and say how directly it applies to a company like ours. Say plainly where good evidence does not exist.
You get: Dated evidence on the questions that actually drive the decision
Separate facts from opinions in the debate
Here are the arguments people in our team have made about [decision]: [paste arguments or meeting notes]. Sort every point into fact (verifiable), forecast (a prediction), value judgement (a preference) or anecdote. For facts, say how we would verify them. For forecasts, say what they depend on.
You get: A cleaner debate where facts, predictions and preferences are not mixed up
Run a pre-mortem
Imagine it is [date one year from now] and our decision to [chosen option] has clearly failed. Write the five most plausible stories of how it failed, each in three sentences, starting from today. For each, give the early warning sign we would see in the first 90 days and one thing we could do now to prevent it.
You get: Failure stories with early warning signs you can watch for
Argue the other side as well as possible
I am leaning towards [option]. Make the strongest possible case for [the alternative], as its best-informed supporter would, using the facts I have given you: [paste facts]. Do not hedge. Then tell me which of your arguments I should take most seriously.
You get: The best version of the case against your preference
Map the second-order effects
If we [decision], what happens next for [customers, team, competitors, suppliers, finances]? For each group, list the first-order effect and the likely second-order effect six to twelve months later. Flag any effect that would be hard to reverse.
You get: Knock-on effects you would otherwise discover too late
Write a one-page decision memo
Write a one-page decision memo for [audience] on [decision]. Sections: decision required and deadline, options considered, recommendation, why (three reasons), what we give up, main risks and how we will watch them, and what would make us revisit the decision. Use only these notes and facts: [paste]. Under 400 words.
You get: A memo that records the reasoning, not just the outcome
Prepare for stakeholder pushback
We have decided [decision]. Stakeholders are [groups and what each cares about]. For each group, write the question they are most likely to ask, a two-sentence answer, and what we can offer them to reduce the cost of this decision for them.
You get: Answers ready for each group before the announcement