Ask for structured output
Asking for a shape as well as a subject is one of the cheapest improvements available. A prompt that says only what to write about leaves the organisation to chance; a prompt that says how to organise it usually gets it.
Where structure is decided
It depends on the tool, and this determines where to ask.
| Tool | Structure comes from |
|---|---|
| Long-form post builder | The outline stage |
| Site-wide copy generator | The page plan, which you can edit |
| Paragraph assistant | Your prompt alone |
For the first two, ask for structure in the prompt that produces the plan, then correct the plan directly. The site-wide tool lets you edit, reorder, add and remove entries before any copy is written — that editing is free and more reliable than re-prompting.
For the paragraph assistant there is no plan stage, so the prompt is the only lever.
Asking for a shape
Name the organising principle:
Structure it as a problem, what we tried, what worked, and what we would do differently.
Organise it as three options with the situation each one suits.
Structure it as an argument that builds to a single recommendation, not as a list.
Alternate: each claim followed immediately by the evidence for it.
That last pattern is worth knowing. Generated content tends to make several claims and then support them vaguely; forcing claim-then-evidence produces noticeably more substantial copy.
Asking for a format
Formats work when they suit the content and produce nonsense when imposed on it:
| Format | Prompt | Suits |
|---|---|---|
| Comparison | Compare the two approaches across cost, time and risk | Decisions between options |
| Steps | A numbered sequence a reader can follow | Procedures |
| Q and A | Questions a buyer actually asks, with direct answers | Objection handling |
| Definition first | Define it in one sentence, then explain when it applies | Explainers |
| Table | Present the differences as a table | Genuinely tabular data |
Only ask for a table when the content is tabular. A table of prose is worse than prose.
Controlling section behaviour
In the outline-driven tools, structural instructions in the prompt are the most effective single thing you can write:
- Prefer fewer, deeper sections — the fix for a fragmented article.
- Must include a section on … — guarantees coverage.
- No history or background section — removes the standard generic opener.
- No concluding summary — removes the paragraph that repeats the article.
- Each section should answer one question — produces headings that are useful in a table of contents.
Length by structure
Length is a form control, not a prompt instruction — the word count field, or the per-page word counts in the site-wide plan. But structure interacts with it: the same total word count spread across four sections or twelve produces very different depth.
If you want depth, reduce the section count rather than raising the word count.
What you cannot ask for
Specific WordPress blocks or layout. The tools generate content, not layout. Structure means the organisation of ideas.
Guaranteed exact lengths. Word counts are targets.
Machine-readable output. There is no JSON or schema mode in the public tools. Do not ask for markup formats; you will get text that looks like them.
Consistency across separately generated sections. Each section is generated independently, so cross-references between them are not reliable. See Generate and review long-form sections.
Verifying what you got
Structure is the easiest thing to check and the most consequential to get wrong:
- Do the headings describe their sections?
- Does the order make sense to a reader arriving cold?
- Does each section do a distinct job?
- Is the heading hierarchy correct and sequential?
- Does the shape you asked for actually appear?
If the structure is wrong in an outline-driven tool, fix it in the plan before generating the copy. Fixing structure after generation means rewriting the copy too.