An outline is the highest-leverage thing you can hand an AI writing tool. Not a prompt tip, not a tone instruction — a structure. When you ask for "a blog post about [topic]", the model picks the most common shape in its training data, which is a flat list of loosely related sections. When you paste a skeleton with named sections and a job for each one, the output has somewhere to go. Over thirty days of testing we found the outline mattered more than the tool.
How to use these templates
Each outline below is a skeleton, not a script. Replace every [bracketed] item with your specifics before you generate — a bracket left in the prompt produces a draft that hedges around it. Keep the section order: the order is where the argument lives. And treat the word counts as proportions, not targets; the point is that the middle of the piece is longer than the introduction.
1. The how-to outline
Best for: anything a reader wants to do, not just understand.
- Opening: the outcome in one sentence, then who it is for — [task] in [time], for [reader] who [situation].
- What you need first: prerequisites, access, files, decisions. Three to six bullets.
- The steps: numbered, one action each, with the failure mode noted under any step people get wrong.
- Checking it worked: how the reader verifies the result.
- What usually goes wrong: two or three specific failures and the fix.
- Next step: one sentence, one action.
Where AI helps: the steps, and especially the failure modes — asking for "what goes wrong at this step" produces genuinely useful material. Where it fails: prerequisites specific to your product. Write those yourself.
2. The comparison outline
Best for: two named options where the reader already knows both exist.
- The decision in one line: who should pick which, stated before the evidence.
- Where they are similar: short. Establishes that the difference is real, not a category error.
- The dimensions that decide it: three to five, each its own section — price, limits, fit for [use case], support, lock-in.
- A table: the same dimensions side by side, for readers who skip prose.
- Switching costs: what it takes to move later. The section writers omit and readers want.
- When neither fits: names the third path honestly.
Where AI helps: producing the dimension list so you notice one you forgot. Where it fails: the cells of the table. Every number goes in from your own test or the vendor page.
3. The list outline
Best for: breadth, and for pages meant to be bookmarked.
- What this list is and is not: scope in one line — [N] [items] for [reader], tested on [basis].
- How the items were chosen: two or three sentences. This is what separates a list from a scrape.
- The items: each with the same three parts — what it is, who it suits, the catch.
- How to choose among them: a short decision rule at the end, because a list of twenty leaves the reader with a new problem.
Where AI helps: consistency — every item getting the same three parts. Where it fails: "the catch". Models default to balanced praise; you have to ask for the drawback explicitly or it will not appear.
4. The case study outline
Best for: proof, and for the bottom of a funnel.
- The result first: one number, one sentence.
- Who they were: size, sector, the constraint that mattered.
- What they tried before: and why it stopped working.
- What changed: the actions, in order, with dates.
- What it cost: money and time. The section most case studies hide.
- What did not work: one honest paragraph, which is what makes the rest credible.
- What transfers: which parts apply to a reader in a different situation.
Where AI helps: turning rough notes into order. Where it fails: the numbers. A generated case study with invented figures is worse than no case study.
5. The opinion outline
Best for: differentiation, and for earning links.
- The claim: one sentence, arguable, specific.
- Why most people think otherwise: stated fairly. Steel-man it or the piece reads as a rant.
- What changed: the evidence or experience that moved you.
- The strongest counter-argument: and what you concede.
- Where the claim stops applying: scope, stated by you.
- What you would change your mind about: one falsifiable condition.
Where AI helps: generating the counter-arguments you did not consider. Where it fails: the claim itself. Models produce reasonable, unarguable sentences, which is the opposite of what this structure needs.
6. The roundup outline
Best for: recurring formats and for building internal links.
- What is being collected and over what period: the [N] [things] from [month].
- Selection rule: one sentence, applied consistently.
- Entries: short — what happened, one line of why it matters.
- The pattern: three or four entries in, what connects them.
- What to watch next: forward-looking close.
Where AI helps: consistent entry length. Where it fails: the pattern section, which requires actually having read them.
7. The pillar outline
Best for: a topic you want to own, with child pages underneath.
- Definition and scope: what is in and explicitly what is out.
- The framework: your own structure, three to seven parts.
- One section per part: each self-contained, each linking to its child page.
- How the parts interact: the section that makes it a framework rather than a list.
- Common mistakes: three to five.
- Where to start: a first action.
Where AI helps: keeping section depth even. Where it fails: the interaction section, again — that is the original thinking and it will not be generated.
8. The update outline
Best for: anything with a year in the title, and for keeping old pages alive.
- What changed since the last version: a dated list, newest first.
- What did not change: surprising, and useful — it stops the reader re-reading everything.
- The current state: rewritten sections, not patched ones.
- What is still uncertain: honest open questions.
- What to check next: a date.
Where AI helps: rewriting stale sections in a consistent voice. Where it fails: knowing what changed. That requires a diff, not a model.
What this costs, from our 30-day benchmark
Our 30-day test measured roughly 22 minutes of editing per 1,000 words on the low-cost dedicated writer against about 8 minutes on the general assistant — near $7 per piece of your time at $30 per hour. Outlining moved that number more than any prompt change we tried: the same tool, given a skeleton, needed far less reordering afterwards.
A workflow that holds up
Pick the structure before you open the tool, and write the bracketed parts yourself — those are the sentences carrying your actual knowledge. Generate section by section rather than in one pass; a single long generation loses the thread around the fourth heading. Ask for the failure modes and the drawbacks as their own step, because they will not appear otherwise. Then read the finished draft against the outline and check that each section still does the job it was given. If a section has drifted into general background, cut it rather than editing it — the fastest way to improve an AI draft is usually deletion.
Outline first, generate second
Paste a skeleton and the draft arrives with a spine. Start on the low-cost plan and test it on real work.
Try Rytr →Frequently Asked Questions
Do outlines actually change AI output quality?
More than prompt wording does. Without a structure the model defaults to the average shape of everything it has read; with named sections that each have a job, the draft arrives ordered.
Should I generate the whole post in one pass?
No. Generate section by section against the outline. A single long generation loses coherence around the fourth heading, which is where most drafts start repeating themselves.
Which outline suits a product comparison?
The comparison structure — decision first, then the dimensions that decide it, then switching costs. The switching-cost section is the one writers omit and readers want.
Can I use these with any tool?
Yes. The outline is the instruction; the tool fills it. We ran the same skeletons on a low-cost dedicated writer and on a general assistant and the difference was smaller than the difference made by outlining at all.