Specify the Output Format

You asked the right question, provided the right context, and added precise constraints, yet ChatGPT responded with a wall of text when you wanted a table. Or it gave you a list when you expected a ready-to-send email. Output format is the most neglected pillar of prompting, yet it determines whether the response is a deliverable or just a draft.

Why format is decisive

ChatGPT does not guess what you will do with its response. Without instructions, it chooses the most neutral form: paragraphs. However, your actual need is often a table to paste into Excel, a list to insert into an email, JSON to pass to your application, or an outline to expand into a presentation. Specifying the format eliminates the reformatting step: the time spent cutting, reorganizing, and retitling a response that was factually correct but structurally useless. The ChatGPT documentation goes further: tell the model how you intend to use the result, and it will better choose the length, level of detail, and organization. “A one-page summary for a director to scan before a meeting, with decisions and next steps first” says far more than “write a summary.”

The most useful formats

Tables are the format of choice for comparisons and any structured data. Explicitly name the expected columns: if you simply say “table,” the model will invent its own, often choosing columns that suit it rather than those you need.

Compare the 4 main project management tools (Asana, Monday, Notion, Trello)
for a marketing team of 8 people.

Format: table with the following columns:
| Tool | Price/month/user | Strengths | Weaknesses | Ideal for |

Structured lists work well for action plans, checklists, and anything that unfolds over time. Again, describe the content of each entry: a list where every item includes a duration, a deliverable, and a tool transforms directly into a schedule, which a simple enumeration of titles does not allow.

Give me an action plan to launch a B2B newsletter in 30 days.

Format: numbered list with each step containing:
- Step number and title
- Estimated duration
- Concrete deliverable
- Recommended tool

For anything related to communication, ask for the finished deliverable—subject line, body, and signature—rather than content you would have to dress up yourself.

Write a thank-you email after a job interview for a Digital Project Manager
position at a software publisher.

Format: complete email with:
- Email subject
- Message body (max 150 words)
- Professional signature

For technical formats, ChatGPT is highly reliable as long as you specify the language, structure, and type of each field. ChatGPT can also produce real files: with ChatGPT Work, it creates or modifies documents, spreadsheets, and presentations, depending on your plan and workspace settings. In that case, name the file format and location, what must remain unchanged, and, for a spreadsheet, the tabs, columns, formulas, and charts that matter. For content to paste into another tool or pass to an application, request CSV, Markdown, or JSON instead.

Generate a JSON configuration file for 5 product categories
in an online clothing store.

Format: valid JSON with each category containing:
- id (string)
- name (string)
- description (string, max 50 words)
- slug (string, URL-friendly)
- productCount (number)

For long documents, you must impose a hierarchy, heading level by heading level; otherwise, you will get flat text that is impossible to integrate into your knowledge base.

Write an onboarding guide for new sales representatives.

Markdown format:
- Main title in H2
- Subsections in H3
- Key points in bold
- Numbered steps for procedures
- "Tip" callout boxes for practical advice

Combining multiple formats in a single prompt

You are not forced to choose a single form. A hybrid response, where the diagnosis is written in prose, problems are tabulated, and the plan is numbered, produces a complete document you can project in a meeting without touching it.

Analyze our tech blog's content strategy (50 published articles,
organic traffic down 20% over 3 months).

Response format:
1. A diagnostic paragraph (max 100 words)
2. A table of the 5 identified problems with their impact and priority
3. A 5-step numbered action plan with deadlines
4. A concluding paragraph with the expected outcome

Enforcing a strict format with an example

When the expected structure is very precise, description is not enough: show the template. The OpenAI Help Center sums it up in two words: show as well as tell. An output template is followed more faithfully than a description, including separators and formatting, making the response easier to reuse in a catalog or content management system. Still, verify every output before importing it.

Generate 5 product descriptions for our online catalog.

Exact format for each product:
---
**[Product Name]**
*Category: [category]*
Price: [price] EUR

[Description in max 2 sentences. The first describes the product,
the second highlights the main benefit.]

Tags: [tag1], [tag2], [tag3]
---

This rigor has a trade-off: for creative content, a template that is too tight stifles production and yields five variants of the same text. In those cases, frame the substance and let the form breathe.

Professional formats to know

Some formats are standards your team recognizes at a glance. Naming them in the prompt is enough to get a document compliant with conventions without having to describe the structure.

A complete case: preparing a client review

The following prompt shows the level of precision to aim for when multiple deliverables coexist. Each part of the response has its own form, adapted to its use during the meeting, and the last line prepares for the moment we dread: the tough question asked at the end. Take a prompt you already use, add a specification of this level, and compare the results.

I am preparing for a client meeting tomorrow morning. Help me structure my notes.

Context: Quarterly review with a SaaS client, 6 months of collaboration,
the client has expressed frustration about support response times.

Expected format:
1. **Agenda**: 5 bullet points, with estimated duration for each
2. **Satisfaction points**: table | Point | Proof | Estimated Score |
3. **Friction points**: table | Problem | Root Cause | Proposed Solution |
4. **Commitments to make**: numbered list with owner and date
5. **Anticipated trap question**: the difficult question the client might ask + prepared answer

Requesting a final verification

For important work, the ChatGPT documentation advises ending the prompt with a verification step: confirm that every action has an owner and a deadline, that the figures in the table match the source, or flag any information that could not be verified. This instruction does not replace your own review: always read the result before using or sharing it.

Before finishing, verify that every commitment has an owner and a date,
and flag any information you could not confirm in my notes.

Your turn

Think about usage before form. Take three recent ChatGPT responses you had to reformat. For each, write one sentence on what you did with it next: pasted into a spreadsheet, sent via email, projected in a meeting. Rewrite the original prompt by adding this usage, the precise format, and a final verification, then compare the results.

Official sources: Prompting (ChatGPT documentation) · Best practices for prompt engineering with the OpenAI API · Creating and editing documents, spreadsheets, and presentations with ChatGPT Work (OpenAI Help Center)

Key takeaways

• Tell ChatGPT what you will do with the result: it will infer the length, level of detail, and organization
• Adapt the format to the final use; with ChatGPT Work, ChatGPT can create real documents, spreadsheets, and presentations
• Show a format example when the structure must be very precise
• Combine multiple formats and name standard professional formats (SWOT, OKR, brief, user story)
• For important work, request a final verification, then review it yourself