daBongo LMS AI Training Courses

Crafting Effective Prompts – Structure, Context, and Constraints

Lesson 3: Requesting Structured Output and Specific Formats

Log in and enroll to track lesson completion.

Lesson Objectives

By the end of this lesson, students should be able to:

  • Request specific output formats by name or by example
  • Match output format to the downstream use case
  • Ask Claude to produce structured data formats for system or workflow integration

Lesson Content

Format is not cosmetic – it determines usability.

A well-reasoned Claude response in the wrong format requires significant rework before you can use it. A response in the right format can go directly into a document, a system, or a decision. Specifying format is not a cosmetic preference – it is part of the task definition.

Common format types and when to use them:

  • Numbered list: Step-by-step instructions, ranked priorities, sequential processes
  • Bullet list: Unranked items, features, considerations, brainstorm outputs
  • Table: Comparisons, feature matrices, schedules, structured data with categories
  • Prose paragraphs: Narrative explanations, essays, emails, reports
  • Checklist: Audit steps, review criteria, pre-launch verification
  • Two-column comparison: Side-by-side evaluation of two options
  • JSON / XML / CSV: When output needs to feed into a system or tool
  • Outline (H1/H2/H3): Document structure, course curriculum, presentation structure
  • Q&A format: FAQ documents, interview prep, knowledge base articles
  • Slide deck outline: Presentations with speaker notes

Specifying format by example.

If you cannot describe the format you want in words, show it. Paste a small example of the structure and ask Claude to follow it. This is particularly useful for custom data formats, branded templates, or unusual structural requirements.

Example: "Use this structure for each item: [Name] | [Description: one sentence] | [Risk level: Low/Medium/High] | [Owner: TBD]"

Requesting format for downstream use.

If the output is going into a system, tell Claude what the system expects. "Format this as a Markdown table," "Return this as valid JSON with these keys: name, priority, deadline, owner," "Format as a numbered list compatible with Microsoft Word styles." Claude handles a wide range of format requests – but only if you ask.

Format and length are related.

A table with 20 rows is often more scannable than 20 prose paragraphs. A numbered five-step process is faster to follow than a narrative description. When you choose format, consider how the reader will use the output – reading linearly, scanning for a specific item, copy-pasting into a system, presenting to an audience.

Practical Example

A team lead needs a project risk register.

Weak prompt

Tell me the risks in my project.

This produces a narrative paragraph of generic risk categories with no structure.

Improved prompt

Create a risk register for a website redesign project. The team has 4 people, a 90-day timeline, and a third-party vendor handling development. Known constraints: tight deadline, single developer dependency, no staging environment yet.
Format: Markdown table with columns – Risk Description | Likelihood (High/Medium/Low) | Impact (High/Medium/Low) | Mitigation Action | Owner (leave as TBD).
Include at least 8 risks. Prioritize them from highest to lowest combined risk.

The output is a structured, immediately usable risk register that can be pasted directly into a project document.

Lesser-Known Tip

Claude can maintain a consistent format across a long output if you give it the format once at the top and ask it to follow it for all subsequent items. For large batch tasks – summarizing 20 documents, generating 15 interview questions with rationale, creating a full curriculum outline – establish the format template in your opening instruction and Claude will apply it consistently throughout.

Safety Notes

When requesting JSON, XML, or other structured data formats that will feed into a system, validate the output before processing it. Claude can produce well-structured-looking code that has subtle errors – mismatched brackets, incorrect field names, invalid values. Always run structured output through a validator or review it manually before feeding it into an automated workflow.

Practice Task

Choose a comparison decision you are currently facing (two tools, two vendors, two job options, two approaches to a problem). Ask Claude to produce a structured two-column or table comparison. Specify the exact columns, the criteria rows, and whether you want Claude to recommend an option or just present the data. Evaluate whether the output is immediately useful or requires editing.

Completion Check

You should be able to name six output formats, explain when each is appropriate, and write a prompt that requests a specific format by name or by example.

Lesson Quiz

Log in and enroll to take this lesson quiz.

Scroll to Top