Control How Your Step Output Is Formatted

Everything in a step's response comes from the step prompt, so formatting is something you ask for. If your prompt says nothing about format, the AI formats by instinct and the result varies from user to user. Adding two or three lines about formatting to the prompt is usually all it takes.

The step editor supports a specific set of formatting. Ask for those, and the response renders cleanly and stays intact when the user edits it.

Add a formatting instruction to your prompt

Open the step from inside its milestone and scroll to the Prompt field. Add a short formatting paragraph at the end, after your instructions about what the step should accomplish. This works for most steps:

Format the response in markdown. Use ## and ### headings, short paragraphs, bold for key terms, and bullet or numbered lists. Use a markdown table when you are comparing options. Use no other markdown formatting.

Save the step, then use Preview to see the response the way a user will.

Formatting the editor supports

• Headings, levels 1 to 4. Ask for them as #, ##, ### and ####.
• Bold with **double asterisks** and *italic* with *single asterisks*.
• Strikethrough with ~~double tildes~~.
• Bullet lists with - at the start of a line, including indented sub-bullets.
• Numbered lists with 1. at the start of a line.
• Tables in standard markdown pipe format, with a header row and a | --- | divider row underneath.
• Paragraphs and line breaks. A blank line starts a new paragraph, a single line break stays a line break.

If a formatting type is not on this list, it is not currently supported in step output. Links, quote blocks, code blocks, horizontal rules, highlighting and checkboxes will not render, and the formatting is dropped as soon as the response is saved. Links are the one people hit most often: the clickable link disappears and only the label text remains, so if a user needs a web address, ask the AI to write the address out as plain text instead.

Show the AI the exact shape you want

The paragraph above is enough for most steps. Use this next pattern only when you want tight control over the output, usually a deliverable that has to come back in the same shape every time.

Naming the formatting gets you halfway. Showing the AI a small model of the finished response gets you the rest, because an example fixes the section order, the heading levels and the length, not just the formatting. Write the prompt in three parts: the task, the formatting rules, then a skeleton for the AI to fill in.

A skeleton for an ideal client profile step might look like this inside your prompt:

• ## Ideal Client Profile followed by [One short paragraph naming who this person is.]
• ### What they want followed by three bullets, each starting with a bold label: **Outcome:**, **Timeline:**, **Budget signal:**
• ### Where they are stuck followed by two bullets: **Belief:** and **Behaviour:**
• ## Fit check followed by a three-column table: Signal, Great fit, Poor fit
• ## What to do next followed by three numbered actions, one sentence each

Put square-bracket guidance inside the skeleton wherever you want to control length, such as [One short paragraph naming who this person is.]. The AI replaces the bracketed text and keeps the structure. Finish the formatting rules with Use no other markdown formatting so the AI does not reach for links or dividers out of habit.

Check the result

Select Preview on the step and read the response as a user would see it. If something looks like plain text where you expected formatting, check it against the supported list above. The most common cause is asking for a formatting type the editor does not support.