# How do you ask ChatGPT for a specific format?

Date: 2023-08-14 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/ask-chatgpt-specific-format/

**Answer:** Name the format and spell out its parts: for a table, list the columns; for a checklist, say how to group the items; for an email, give the sections and mark where your own details go. The more exactly you describe the shape, or show a small sample of it, the less you have to rearrange afterward.

ChatGPT's default answer is a few paragraphs, sometimes with a list. That suits a conversation, but most real work has a destination: a spreadsheet, a meeting agenda, an email, a slide. When the answer arrives in the wrong shape, you spend your time cutting and rearranging instead of using it.

The fix is to treat format as part of the request. *A Beginner's Guide to ChatGPT* lists it among the core habits of a good prompt: tell the model what output you expect, including its structure and level of detail. OpenAI's prompting guide for developers goes a step further and notes that models follow a format more closely when you show it to them, not just name it.

## Three ways to describe a format

**Name it.** "As a table", "as a numbered checklist", "as a three-paragraph email". This alone helps, but it leaves the details to the model.

**Specify its parts.** Column names, the order of sections, the number of items, the length of each line. This is where most of the gain comes from.

**Show a sample.** Type the first row of the table, or a skeleton with empty slots, and ask the model to follow it. This is the most reliable option when the output must match a template you already use.

## Worked example 1: a table

```prompt
Compare three ways to back up family photos: an external hard drive, a cloud storage service, and printing an album of the best ones.

Format: a table with these columns, in this order: Option | Rough cost | Effort to set up | Main risk | Best for.
Keep each cell under 12 words. After the table, add one sentence on how to combine two of the options.
```

Naming the columns does two jobs. It decides what gets compared, and it forces short, parallel answers that are easy to scan. If you plan to paste the table into a spreadsheet, say so, and ask for plain numbers without extra words in the cost column.

## Worked example 2: a checklist

```prompt
I'm moving to a new apartment in six weeks. Make a checklist of what to do.

Format: group the tasks under these headings: 6 weeks before, 2 weeks before, moving week, first week after. Start every item with a verb. One line per item, no explanations. Put a checkbox [ ] at the start of each line.
```

The headings turn a long list into a timeline. "Start every item with a verb" is a small rule with a large effect: it turns vague entries like "utilities" into actions like "Book the internet installation."

## Worked example 3: an email skeleton

```prompt
Write a polite reminder email to a client whose invoice is 30 days overdue. This is the first reminder, and the relationship is good.

Format: use this skeleton and keep my placeholders in square brackets exactly as written.
Subject: [one line]
Greeting: Hello [client name],
Paragraph 1: friendly reminder with [invoice number], [amount] and [due date].
Paragraph 2: how to pay, and an offer to help if something is wrong with the invoice.
Closing: thanks, then [my name] and [company].
Keep the body under 100 words.
```

The square brackets keep ChatGPT from inventing a client name, an amount or a date. That matters: a model asked to fill in details will often produce plausible ones, and an email with a made-up invoice number is worse than one with a blank.

> **Try it:** Take a request you made to ChatGPT recently and run it again with a format line added: name the format, list its parts, and set a length. Compare the two answers and count how many edits each one would need before you could use it.

## When the format slips

Long conversations drift. After several exchanges, ChatGPT may forget a column or return to paragraphs. Two habits help. First, restate the format in the follow-up ("Same table format as before, now for video files"). Second, correct one thing at a time: "Shorten every cell in the Main risk column to five words or fewer" works better than "Make it tidier." OpenAI's guide suggests the same principle for all prompts: replace a fuzzy description with a precise one, and say what to do rather than only what to avoid.

## Which formats save the most editing time?

Formats that match the destination save the most time. A table with the exact columns of your spreadsheet can be copied across almost as is. A skeleton that mirrors your usual email or report means you only fill in blanks and adjust tone. A checklist whose items start with verbs can go straight into a to-do app.

Next come formats that limit length. Word or line limits per item stop the model from padding, which is often the biggest editing job of all. Asking for "no introduction and no summary" removes the opening and closing paragraphs that you would otherwise delete by hand.

Finally, placeholders save time in a less obvious way: they save checking. Every name, number or date that the model leaves blank is one you do not have to verify. For anything you will send to another person, a skeleton with brackets is usually faster to finish than a complete draft you have to fact-check line by line.

## Sources

- A Beginner's Guide to ChatGPT: https://www.amazon.com/dp/B0BZRRPXZW
- Best practices for prompt engineering with OpenAI API: https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-openai-api