What it is
A comparison table puts the options down one side and your criteria across the top. The model fills each cell. Because every option has to answer every column, nothing gets a free pass: if one supplier’s quote says nothing about the notice period, the empty cell shows it.
Four choices make the difference between a useful table and a pretty one. Name the criteria yourself, because the model’s default criteria are generic and may not include the one you care about. Say where the facts come from: the quotes you pasted, or the model’s knowledge with anything uncertain marked. Keep cells short. And keep the verdict out of the table, or ask for one on a priority you state, so the recommendation follows your weighting instead of the model’s.
When it helps
Use it for any choice between a handful of options on several criteria: tools, suppliers, plans, office locations, ways to staff a team. Use it to compare proposals or quotes you have in hand, where the facts are on the page and the hard part is lining them up. And use it when several people must agree, because a table is easy to talk through in a meeting and easy to correct cell by cell.
It is also a quick way to see what you do not know. A row full of “not stated” is the list of questions to send each supplier before you decide.
When it does not
Cells compress. “Includes support” in ten words cannot carry “only on the annual plan, and only by email”. Allow a notes column, or ask for footnotes on any cell with a condition.
Facts from memory go stale. Prices, plan limits and features change, and a model filling cells from what it learned in training can be out of date without saying so. Paste the sources, or tell it to mark anything it is not sure is current. The guide to reducing hallucinations covers why.
Scores look more exact than they are. A weighted score of 4.2 against 3.8 rests on the model’s judgment of each cell. If you score, set the weights yourself and ask for the reason behind each score.
Wide tables do not read on a phone. More than five or six columns is hard to scan anywhere. And a decision that turns on one criterion does not need a table at all; ask that question directly.
Before and after
Three decisions from different kinds of work. The scores are the Prompt Grader’s free estimate with Written for set to Any chat model. A score counts what the prompt states; it does not grade the answer.
Which project management tool should we use?
We are a 15-person marketing agency choosing between Asana, Trello and Monday.com. Today we track client work in shared spreadsheets. Compare the three tools in a table, one row per tool, with these columns in this order: price for 15 users per month, client guest access, time tracking built in, Google Drive integration, and setup effort for a non-technical team. If you are not sure a fact is current, write [CHECK] in that cell, because prices and plans change often. Do not pick a winner. Below the table, add two sentences on which tool fits a team whose main need is client guest access.
The question gets a recommendation of whichever tool is most often recommended, with reasons you cannot weigh. The table puts all three against the five things this agency cares about, and the [CHECK] cells tell you which prices to confirm on the vendors’ own pages.
Compare these three supplier quotes. """ [three quotes] """
Three quotes for office cleaning, pasted below: """ [three quotes] """ Our facilities manager will choose one this week and needs the trade-offs side by side. Build a table with one column per supplier and these rows: monthly price, what is included, contract length, notice period, insurance cover, and anything excluded. Use only the quotes. If a quote does not state something, write "not stated" in that cell. Keep each cell under 15 words. A good table lets someone see the cheapest option and its catch in ten seconds.
“Compare these” returns three paragraphs, one per supplier, each covering different points. The table answers the same six questions for all three, and the “not stated” cells, often the notice period or insurance, become the questions to ask before signing.
Should we hire a bookkeeper or outsource it?
We are a 25-person landscaping company deciding how to handle bookkeeping, so the owner can stop doing the books at night. We process about 400 transactions a month. Compare three options in a table: a part-time in-house bookkeeper, an outsourced bookkeeping firm, and accounting software with a monthly review by our accountant. Rows: likely yearly cost, time to set up, owner hours freed per month, and the main risk. Label any cost you are estimating as an estimate. If our numbers are not enough to estimate a cost, say what you would need to know. Leave out options that need a full-time hire. After the table, rank the options on owner hours freed alone, in two sentences.
The yes-or-no question gets a balanced essay that ends “it depends”. The table sets three options side by side, including the software route the yes-or-no question left out, labels every estimated cost, and ranks the options on the one thing the owner cares about.
The line to add
Compare {{options}} in a table, one row per option, with these columns in this order: {{criterion}}, {{criterion}}, {{criterion}}.
Take facts only from the material below. If something is not stated, write "not stated" in that cell. Keep each cell under {{number}} words.
After the table, say in two sentences which option fits best if {{your top priority}} matters most.
With only two or three options, put the options as columns and the criteria as rows; it reads better on a narrow screen. If there is no material to paste, replace the second paragraph with: “If you are not sure a fact is current, write [CHECK] in that cell.”
In ChatGPT, Claude, Gemini and Copilot
All four chat assistants can answer with a table when you ask for one. Microsoft’s own example Copilot prompt ends “Give me a day-by-day itinerary in a table” (Get started writing prompts), and Google’s guidance lists a table among the response formats you can specify (Prompt design strategies). Anthropic’s guidance asks you to be specific about the output format and constraints you want (Prompting best practices), which here means naming the columns.
For facts the model cannot have, such as this year’s prices, a tool that can search helps. Google’s guidance says grounding with Google Search should be enabled whenever Gemini may need obscure or recent facts (grounding and code execution). With any assistant that searches, ask for the source link behind each cell and open the ones the decision turns on.
To move the table into a spreadsheet, ask for CSV instead, or copy the table and clean it with Copy Cleaner.
Key takeaways
- Name the criteria yourself, in the order you care about them.
- Say where facts come from, and have gaps written as “not stated”.
- Mark anything from memory that may be out of date, and check it.
- Keep cells short and the table narrow.
- Keep the verdict separate, or tie it to a priority you state.
Questions
How do I get ChatGPT to compare options in a table?
Name the options and the criteria, and ask for a table with one row per option and one column per criterion, in your order. Say where the facts should come from, such as quotes you paste, and ask it to write “not stated” for anything missing. Keep cells short, and ask for the recommendation separately.
Can AI make a decision matrix with weighted scores?
It can, but the scores are only as good as the judgments behind them. Set the weights yourself, ask for a one-line reason for each score, and check the cells that decide the result. A table of facts with your own priority stated is often more useful than a weighted total.
Are AI comparisons of software and prices accurate?
Not reliably from memory. Prices, plans and features change, and a model may describe last year’s version without saying so. Paste the vendors’ current pages or quotes, use an assistant that searches the web and ask for a source link per cell, and confirm anything the decision depends on.
Prompts to use with this
- Brainstorm names and taglines against your criteriaYou need options for a product, event or campaign name, checked against your rules and audience.
- Compare vendor proposals against your requirementsYou have several quotes or proposals for the same need and want a fair side-by-side comparison against your criteria, with every claim traced to a proposal.
- Compare two lists in a spreadsheet and find what is missing or changedYou have two lists, such as last month's and this month's customers or a system export and your own tracker, and need what is in one but not the other and what changed.
- Screen applications against a role's must-have criteriaYou have a pile of applications and want each checked against the same job-related criteria, with a person making every decision.