What it is
The format is the structure of the answer. Common ones at work are a table, a bulleted or numbered list, sections with set headings, an email with a subject line, step-by-step instructions and plain paragraphs. On the ten checks this is Form, the fifth check.
The more exactly you name it, the less you reshape afterward. “A table” leaves the columns, their order and their names to the model. “A table with one row per supplier and columns for price, lead time and minimum order” does not, and every row comes back comparable with the next.
When it helps
It helps whenever the answer is going somewhere that already has a shape. A spreadsheet wants columns. An email wants a subject line and a short opening. A ticket wants a summary line, steps to reproduce and the expected result. Asking for that shape saves the copying and rearranging that otherwise follows every answer.
It helps when you are comparing things, because a table makes gaps visible: an empty cell is easier to spot than a missing sentence. Comparing options in a table is a pattern of its own for that reason. And it helps with anything you produce every week. If the minutes or the status update always have the same sections in the same order, readers learn where to look.
When it does not
Reasoning reads better as prose. An argument broken into bullets can hide the step where the logic does not follow, and a recommendation needs its “because”. For a decision memo or an explanation, ask for paragraphs, and save lists for things that really are separate items.
A format is not content. A neat table of guesses is still guesses, and a column the material has no data for will be filled anyway unless you allow for it. Give the model a way to say so, such as “yes, no or unclear”, or follow the pattern of saying what to do when unsure.
Chat apps display formatting with Markdown, the asterisks and pound signs behind bold text and headings. Pasted into some emails, forms and documents, those symbols come along as text. Ask for plain text when that is where the answer is going, or paste it through Copy Cleaner, which keeps real headings, lists and tables and drops the symbols.
When software will read the answer, a described format is not strict enough. A person can read a table with one odd row; a script cannot. Ask for JSON with fixed keys, the subject of the extract to JSON pattern.
Before and after
Two requests where the after version names the format. Each score comes from the Prompt Grader’s free ten-check estimate, written for any chat model.
Compare the three payroll providers in the quotes below for us. """ [the three quotes, pasted in full] """
Compare the three payroll providers in the quotes below for us. Format: a table with one row per provider and these columns: monthly cost at our headcount of 45, setup fee, minimum contract length, and whether it connects to our accounting software (yes, no or unclear). Under the table, one sentence on which is cheapest over two years. """ [the three quotes, pasted in full] """
Without a format, a comparison can come back as a paragraph per provider, each covering different points, so you build the table yourself. With the columns named, the three quotes line up on the same four questions and the “unclear” option shows where a quote was silent. The sentence under the table also sets a size, which the estimate counts.
Write up what we agreed on the call with the client, from my notes below. """ [your call notes] """
Write up what we agreed on the call with the client, from my notes below, as an email I can send them. Format: a subject line, a two-sentence opening, then two bulleted lists headed "Agreed" and "Next steps". Put the owner and the date at the end of each next step. """ [your call notes] """
“Write up” could produce notes for your files, a summary for your manager or an email to the client. The second version gets the email, with the next steps in a place where the client can see what they owe and by when.
The line to add
Format: {{shape}}, with {{the_parts_it_must_have}} in this order.
Filled in, that might be “Format: a table with one row per supplier and columns for price, lead time and minimum order” or “Format: an email with a subject line, a two-sentence opening, then the decisions as bullets.” When the result has a natural shape you already use, describe that one. When you have a past example in that shape, show it, which is faster than describing it.
In ChatGPT, Claude, Gemini and Copilot
All four take a plain description of the format, and their defaults differ, which is the reason to give one. OpenAI’s guide to its latest models says its newest model tends to use lists, tables and Markdown to make responses scannable, and that you should say so if you want prose with less formatting. Google’s guide for Gemini lists formats you can ask for, among them a table, a bulleted list, an elevator pitch, keywords, a sentence or a paragraph.
Anthropic’s guide for Claude adds two habits worth using anywhere. Tell the model what to do instead of what not to do: ask for flowing prose paragraphs rather than saying “no Markdown”. And match the style of your prompt to the style you want back, since removing Markdown from the prompt can reduce Markdown in the answer.
Microsoft’s guide to Copilot prompts calls this part expectations, how Copilot should respond, and one of its sample prompts ends by asking for a day-by-day itinerary in a table.
Key takeaways
- Name the shape the destination needs: columns for a spreadsheet, a subject line for an email.
- Name the parts, not just the type. “A table with these columns” beats “a table”.
- Keep reasoning in paragraphs and lists for separate items.
- Give the format a way to show gaps, such as “unclear” in a column.
- For software, ask for JSON with fixed keys instead of a described layout.
Questions
How do I get ChatGPT to answer in a table?
Ask for a table and name the columns: “a table with one row per supplier and columns for price, lead time and minimum order.” Naming the columns keeps every row comparable. To move the table into Word, Google Docs or a spreadsheet without stray symbols, paste the answer through Copy Cleaner.
How do I stop AI from using bullet points?
Ask for what you want instead: “write three paragraphs of plain prose.” Anthropic’s guidance recommends describing the format you want over listing formats to avoid, and notes that the style of the prompt can carry into the answer, so write the prompt itself in plain sentences.
Why does my AI answer have asterisks and pound signs when I paste it?
Those are Markdown, the formatting chat apps use for bold text, lists and headings. The app shows them as formatting, but some places you paste into show the raw symbols. Ask for plain text when the answer is going into an email or a form, or clean it with Copy Cleaner, which turns them into real formatting.
Prompts to use with this
- Write a professional email from rough bullet pointsYou need to send a clear, polite email and only have a few points jotted down.