How to Make a Decision Matrix With AI (5 Steps + Prompts)

Written by

You have three options on the table, a deadline on Friday, and a team that keeps saying “they’re all fine, really.” Maybe it is which CRM to buy, which freelancer to hire, or which of three projects gets the budget. You have gone back and forth so often that you no longer trust your own gut.

Here is how to make a decision matrix with AI: tell ChatGPT or Claude what you are deciding and what matters to you, let it suggest the criteria and weights, then fix those weights yourself before it sees your options. In a second prompt, ask it to score each option from 1 to 5 with a one-line reason per score. Copy the table into a spreadsheet, let a formula do the math, and ask the AI which single change would flip the winner. A first matrix takes about 20 minutes, and you end up with a choice you can explain to anyone.

Below: the five steps, prompts you can copy, a worked example, the spreadsheet formula, and the two traps that make AI matrices lie to you.

No password needed – your reset opens instantly.

No password. Just your email, and you’re in.

What a decision matrix is, and why AI makes it easier

A decision matrix is a table that compares options against a shared set of criteria, such as cost, quality or risk (Asana, decision matrix guide). Each criterion gets a weight for how much it matters, each option gets a score per criterion, and the weighted total shows which option fits your priorities best.

The idea is old and simple. What makes people skip it is the setup: thinking of the right criteria, agreeing on weights, writing a reason next to every score. That is the part an AI assistant speeds up. It will happily suggest criteria you forgot, phrase them so they can be scored, and keep the reasons short.

What it can’t do is know what matters to you. A weight is a business judgment. If you let the AI set it quietly, it has made the most important call in the whole table. So the rule for this guide: the AI drafts, you decide the weights, a spreadsheet does the math.

What you need

A chat assistant: ChatGPT, Claude or Gemini. The free plans are enough. New to them? Start with our ChatGPT tutorial for beginners or the Claude tutorial for beginners.
A spreadsheet: Excel or Google Sheets. You only need one formula, and we give it to you below.
Your facts about each option: prices, notes from demo calls, a few lines from each proposal. The more real detail you paste in, the less the AI has to guess.

Step 1: Write the decision as one neutral sentence (2 minutes)

Most lopsided matrices start with a loaded question: “Should we go with Option B, it seems cheaper?” The AI picks up your lean and builds a table that agrees with you. That is a known habit of these tools. Anthropic’s researchers found that five leading AI assistants consistently tended to tell people what they wanted to hear, a behaviour they call sycophancy (Anthropic, Towards understanding sycophancy in language models).

Loaded: “Is HubSpot or Pipedrive better? I like Pipedrive.”
Neutral: “Which CRM should a 4-person consultancy use to track about 200 client contacts and follow-ups, starting next month?”

Size, use and deadline are in the neutral version. Your preference is not.

Step 2: Let the AI propose criteria and weights, without the options (5 minutes)

This is the step most prompt templates skip, and it is the one that keeps the matrix honest. Ask for criteria and weights first, and do not name your options yet. Paste this:

I need to build a weighted decision matrix.

The decision: [your neutral sentence from step 1]
What matters to us, in my own words: [e.g. we are not technical, money is tight this year, everything lives in Gmail and Google Calendar, we might switch again in 2 years]

Step 1 only. Propose 4 to 6 criteria for this decision. For each one give:
- a one-line definition I can score from 1 to 5
- a weight in percent (all weights add up to 100)
- one sentence on why it deserves that weight

Do not suggest any product or option yet. Do not tell me what you would choose. Keep criteria separate, so no two of them measure the same thing.

Now read the weights like a manager, not a reader. Would you defend each one to your business partner? If “price” comes back at 40% and you know the team will quit a tool they hate, move points to ease of use. Change the numbers until they feel right, then write them down. From here on they are locked.

One check that catches most bad setups: ask “Is there any criterion here that is really a must-have, not a trade-off?” A must-have (for example “works in our country” or “meets our client’s data rules”) does not belong in the scores. It is a filter: an option that fails it is out before you start.

Step 3: Score each option, with a reason per score (5 minutes)

Now bring in the options and your facts. Open a fresh chat, or say clearly that the weights are fixed:

Here are my locked criteria and weights. Do not change them:
[paste your final list with weights]

Here are the options and what I know about each:
Option A: [your notes, prices, demo impressions]
Option B: [...]
Option C: [...]

Score every option on every criterion from 1 (poor) to 5 (excellent).
Rules:
- give a one-line reason next to every score, based only on my notes
- if my notes do not say, write UNKNOWN instead of a score and tell me what to find out
- show the result as a table: criteria in rows, options in columns, weights in their own column
- do not calculate totals and do not pick a winner

Two parts of that prompt matter most. The reasons turn a number you have to trust into a claim you can check. And the UNKNOWN rule stops the AI from quietly turning a gap in your knowledge into a low score. A missing fact is a question to answer, usually with one email to a sales rep or a look at the vendor’s help pages, not a point to deduct.

Why no totals? Chat assistants are good with words and less reliable with arithmetic. A spreadsheet never gets a multiplication wrong, so let it do that part.

Step 4: Let a spreadsheet do the math (3 minutes)

Ask the AI: “Give me this table as CSV I can paste into Google Sheets.” Paste it in so that the weights sit in column B (rows 2 to 6) and each option has its own column of scores next to it. Then, under Option A’s scores, type:

=SUMPRODUCT($B$2:$B$6,C2:C6)/100

SUMPRODUCT multiplies each weight by its score and adds the results up. It works the same way in Excel and Google Sheets. Dividing by 100 brings the total back to the 1 to 5 scale, because the weights add up to 100. Drag the formula across to the other options. If your matrix has more rows, change the 6 to your last row. Formulas giving you trouble? Our guide on how to use AI to write Excel formulas walks you through it.

Step 5: Ask what would flip the winner (5 minutes)

A total at the bottom of a column is not yet a decision. Paste the finished table back into the chat and ask:

Here is my finished decision matrix with totals: [paste]

1. Which single weight, if I moved it by 10 points, would change the winner?
2. Which single score, if it is wrong, would change the winner?
3. What should I check this week before I decide?
Argue against the current winner as hard as you honestly can.

If the answer is “nothing realistic would change it,” you have a clear winner. If one small change flips it, you have a close call, and you now know exactly which question to settle before you sign anything.

Bonus check: ask for a fresh scoring pass with the options listed in a different order. Researchers testing AI models as judges found they can favour an answer based on its position in the list (Zheng et al., Judging LLM-as-a-Judge). If the scores move when only the order changed, trust the reasons over the numbers.

A worked example: choosing a CRM for a small team

Here is what the steps produce for the consultancy from step 1. The options are kept anonymous on purpose: the method is the point, not the brands, and real scores depend on your own notes and the prices on the day you check.

After step 2 and a few edits, the locked criteria were: setup effort for a non-technical team (25%), monthly cost for 4 users (20%), Gmail and calendar connection (25%), ease of daily use (20%) and getting the data out later (10%). In step 3 the AI marked Option B’s data export as UNKNOWN. One look at B’s help pages showed it exports contacts only, not the notes, so that became a 2.

Criterion	Weight	Option A	Option B	Option C
Setup effort	25	4	5	2
Monthly cost	20	3	5	2
Gmail and calendar	25	5	3	5
Ease of daily use	20	3	4	5
Data export	10	4	2	5
Weighted total	100	3.85	4.00	3.65

Option B wins, but only just. Step 5 showed why that matters: move 10 points from cost to the Gmail and calendar connection (35% instead of 25%, cost 10% instead of 20%) and Option A comes out on top, 4.05 against 3.80. So the real question was never “which CRM is best.” It was “how much do we live in our inbox?” For a team like this one, that is a ten-minute conversation. If almost every client conversation starts in email, A is the better pick, and the choice fits in one sentence that nobody has to win an argument over.

That is the quiet value of the method. It does not decide for you. It shows you which disagreement you are really having.

When a decision matrix helps, and when it doesn’t

It works best when you have 2 to 5 real options, several things that matter at once, and other people who need to understand the choice. Typical work examples:

Picking software: a CRM, a booking tool, an AI voice agent platform, or an automation tool (our n8n vs Zapier comparison gives you ready-made criteria).
Hiring or choosing a supplier: score candidates or quotes on the same criteria, so the loudest one does not win by default.
Choosing which project goes first: impact, effort, risk and urgency as the criteria. Once you have the winner, turn it into a plan with our guide on how to use AI to create a project plan.

It helps much less when there is only one real option, when a single must-have decides everything, or when the choice is mostly about values (whether to take on a business partner, for example). Then a pros and cons list or an honest conversation does the job better.

Decision matrix, SWOT or mind map: which one do you need?

These three get mixed up, and they answer different questions:

Mind map: “What is even on the table?” Use it first, when the problem is still a tangle. See how to make a mind map with AI.
SWOT analysis: “What are the strengths, weaknesses, opportunities and threats of one plan?” See how to do a SWOT analysis with AI.
Decision matrix: “Which of these options do we pick?” Use it last, when you have the options and need to choose.

A good order for a big choice: mind map to find the options, a quick SWOT on the top two if the stakes are high, and the matrix to make the call. If the chosen option is a process change, our guide on how to make a flowchart with AI helps you show the team how the new way works.

When the matrix comes out wrong

Every option scores about the same. Your criteria are too vague (“quality”, “value”). Ask: “Rewrite each criterion so that a 1 and a 5 look clearly different for a team like ours.”
The winner feels wrong. Take that feeling seriously. Usually a criterion is missing. Ask: “I expected Option C to win. Which criterion would I need to add for that to happen, and is it a fair one?” If it is fair, add it and re-score. If it isn’t, you just learned something about your gut.
The AI invents facts about a product. It can describe features or prices that are out of date or wrong. Score only on what you checked yourself, and use the UNKNOWN rule from step 3.
The scores look suspiciously like your first preference. Go back to step 1 and check your sentence for a hint of what you wanted.

Questions people ask

Can ChatGPT make a decision matrix?

Yes. It can suggest criteria, propose weights and score options with reasons, and it can give you the table as CSV for a spreadsheet. Keep the weights in your own hands and let the spreadsheet calculate the totals, as described in steps 2 and 4.

Is Claude or ChatGPT better for this?

Both handle it well, and the method matters far more than the tool. If you want to compare them for your own work, our ChatGPT vs Claude guide goes through the differences.

What scale should I use, 1 to 5 or 1 to 10?

1 to 5 is plenty for most work decisions. A wider scale invites fake precision: nobody can really tell a 7 from an 8 on “ease of use.” What matters more is that a 5 means the same level of good on every criterion.

Should I paste supplier quotes or CVs into an AI tool?

Only if your company allows it. When in doubt, leave out names, personal details and confidential figures, and describe each option in your own words. The matrix needs your judgment of the facts, not the original documents.

Sources

Your next 20 minutes

Pick the decision you have been putting off this week. Write it as one neutral sentence, run the step 2 prompt, and fix the weights before you look at a single option. You will probably see the answer forming before the spreadsheet is done.

Want more small, practical wins like this one? Join our free Launchpad.