An AI content outline is the structure you approve before a model writes a single paragraph: the title, the questions each section answers, what each section must prove and roughly how long it runs. Fixing a heading takes seconds; fixing a drafted section built on the wrong heading means rewriting the section.

Google’s own guidance on generative AI content names structure as one of the places AI helps: it calls generative AI useful “to add structure to original content”. The original part is yours: the angle, the sources and the examples from your brief. This page covers the step between brief and draft, and the introduction that comes out of it.

Summary

Ask the model for an outline before you ask for any paragraphs, and give it the brief as input. A usable AI content outline has headings phrased as the reader’s questions, one job per section, the evidence each section needs and a word budget. Check it with the skim test (do the headings alone tell the story?) and the overlap test (does any section answer the same question twice?). Then write the introduction from the outline: the answer, the condition it depends on and what the page covers, in that order.

What should an AI content outline contain?

A list of headings is a table of contents. An outline a model can draft from well carries six things:

ElementWhat it looks likeWhy the model needs it
Working titleOne line, keyword near the frontSets the promise every section must keep
Answer sentenceThe page’s main answer in under 30 wordsBecomes the start of the introduction
H2s as questionsThe reader’s questions in the order they ask themStops the model inventing filler sections
Job per sectionOne line: what the reader can do after this sectionGives each section an end point
Evidence per sectionThe source, example, table or number it needsTies each fact to the brief instead of the model’s memory
Word budgetA rough number per section that adds up to the length rangePrevents one bloated section and five thin ones

The order of the H2s follows the inverted pyramid, which Nielsen Norman Group describes as starting with the most important information so readers get the main point even if they stop after one paragraph. In an outline that means the section answering the search goes first, and background, history and edge cases go last or go.

The input for all of this is the brief. If you do not have one yet, start with the SEO content brief template; its must-cover sections field is the raw material for the outline.

Which prompts produce a usable outline?

Anthropic’s prompting guide for Claude recommends putting long inputs at the top of the prompt, above the instructions, and wrapping each kind of input in its own XML tag. It also suggests 3 to 5 examples when you want a set format. The five prompts below follow that layout; paste them into Claude, ChatGPT or Gemini. Run them in order; each one fixes what the previous one tends to miss.

1. Outline from the brief

<brief> [paste the filled brief] </brief>

Write an outline for this page. Give: a working title; the answer to the main search in one sentence of under 30 words; 6 to 9 H2s phrased as the questions the reader asks, in the order they would ask them; under each H2, one line starting with “Job:” saying what the reader can do after the section, one line starting with “Evidence:” naming the source, example or table it needs from the brief, and a word budget. Use only sources listed in the brief. Write no paragraphs.

2. Gap check against what ranks

<outline> [paste outline] </outline> <competitor_headings> [paste the H2s of the top 3 results] </competitor_headings>

List questions the competitor headings answer that the outline does not. For each, say whether it belongs on this page or on a separate page, and why. Then list one thing the outline covers that none of the competitors do.

3. Headings that survive a skim

Rewrite each H2 so the first two words carry the topic of the section. Keep them as questions. Flag any two headings that a reader would expect to answer the same thing.

4. The sceptical reader

Read the outline as the reader described in the brief, who is in a hurry and doubts the advice. List the 5 questions they would still have after reading every heading. Mark which ones need a new section, an FAQ, or one sentence inside an existing section.

5. Cut to the length range

The page should be [length range] words. Merge or cut sections until the word budgets add up to that range. Say what you cut and why. Never cut a section whose evidence line names first-hand material from the brief.

What does a weak outline look like next to a fixed one?

Example keyword: how to price a logo design project, for freelance designers. The first column is a generic outline built from the keyword alone; the second is the same topic rebuilt with prompts 1 to 5 and a brief. Both were written for this example.

Before: keyword onlyAfter: brief plus prompts 1 to 5
Title: Logo Design Pricing: A Complete GuideTitle: How to Price a Logo Design Project: 3 Methods With the Maths
IntroductionAnswer sentence: price from your hours, your rate and a revision buffer, then check the total against what the logo is worth to the client.
What is logo design pricing?How do you work out a price for one logo project? Job: calculate a quote. Evidence: worked example with hours x rate + buffer. 350 words.
Factors that affect logo pricingHourly, fixed or value-based: which method fits which client? Job: pick a method. Evidence: comparison table. 300 words.
Hourly vs fixed pricingWhat goes into the quote besides the design? Job: list rounds, files and usage rights. Evidence: copyable quote checklist. 250 words.
Tips for pricing your logoHow do you stop scope creep after the quote? Job: write the revision clause. Evidence: sample clause. 250 words.
ConclusionWhat do you say when the client says it is too expensive? Job: answer the objection. Evidence: three reply scripts. 250 words.
(none)FAQ candidates: deposit size, pricing a logo refresh, pricing for non-profits.

Six things changed between the columns:

  1. The definition section went. Someone searching how to price a project knows what pricing is.
  2. Every heading became a question the reader asks, in the order they ask it.
  3. Each section got a job, so the model knows when the section is finished.
  4. Each section names its evidence, so facts come from the brief and the worked example instead of the model’s memory.
  5. “Factors” and “Tips” disappeared. Both are buckets that collect generic advice.
  6. The conclusion became the last real question, the price objection, which is where the reader’s next action happens.

How do you check an outline before the draft starts?

Seven checks, about five minutes. Fix the outline until all seven pass; every failure here becomes a rewrite later.

CheckPass ifFix
1. First section answers the searchThe first H2 delivers what the keyword asks forMove the answering section up; push background down
2. Skim testReading only the headings tells the reader the whole answerRewrite vague headings with prompt 3
3. Overlap testNo two sections answer the same questionMerge them, or give one a different job
4. Evidence namedEvery section lists its source, example or tableAdd it from the brief or cut the section
5. First-hand material placedThe brief’s own data or examples appear in at least one sectionAssign each item to a section
6. FAQ adds new questionsNo FAQ repeats an H2Replace it with a question from prompt 4
7. Budgets add upSection budgets sum to the brief’s length rangeRun prompt 5

The skim test has research behind it. Nielsen Norman Group’s eyetracking work on scanning describes a layer-cake pattern in which readers scan headings and subheadings and skip the text below, and advises starting headings with the words that carry the most information. An outline that passes the skim test serves those readers before a word of body text exists.

How do you write the introduction from the outline?

Write the introduction last, from the approved outline. It has three jobs, in this order:

  1. Answer. The answer sentence from the outline, with the keyword in it when it reads naturally.
  2. Condition. What the answer depends on: budget, team size, location, skill level.
  3. Map. What the page covers, in one sentence built from the H2s.

Two to three short paragraphs carry all three. Nielsen Norman Group’s advice from the same scanning research is to put the most important points in the first two paragraphs, and its eyetracking shows the first lines of text get the most attention.

Do not write the introduction to win a featured snippet. Google’s featured snippet documentation says a site cannot mark a page as a featured snippet; Google’s systems decide. A plain answer near the top helps the reader either way.

Intro prompt to copy:

<outline> [paste the approved outline] </outline>

Write the introduction in 2 or 3 short paragraphs, under 120 words. Sentence 1 gives the answer sentence from the outline. Sentence 2 says what the answer depends on. Then one sentence saying what the page covers, built from the H2s. Never open with a definition, a question, a scene or a statistic that the outline does not source. Do not say “in this article”.

Which introduction formulas work for search articles?

Five formulas cover almost every search article. Each example below uses the logo pricing outline from above.

FormulaStructureExample openingUse it for
1. Answer firstAnswer + condition + mapPrice a logo project from your hours, your rate and a revision buffer, then test the total against what the logo is worth to the client. The method changes with the client’s size, so this guide covers hourly, fixed and value-based quotes.How-to and definition searches
2. Number firstA number from the page + what it means + mapTwenty hours at $60 an hour is a $1,200 logo before a single revision round. That arithmetic is the floor of your quote, and everything else on this page sits on top of it.Cost, salary and calculation searches
3. Mistake firstThe common mistake + its cost + the fixQuoting a logo by the hour without a revision limit turns a two-week job into a two-month one. Fix the rounds in the quote first, then set the price.Searches from people who already tried and failed
4. Situation firstA specific reader situation + the answerA bakery asks for a logo and wants a number by Friday. Start from the hours the brief will take, then add a buffer for two revision rounds.Searches with a clear trigger moment
5. Term firstPlain definition + why it matters now + mapValue-based pricing means charging for what the logo earns the client instead of the hours you spend. It suits clients with a clear revenue line; the rest of this guide shows when hourly is safer.Only when the keyword is a term the reader does not know

The number in formula 2 is arithmetic from the article’s own worked example, which is the safest kind of opening number. A market statistic in the first line needs a source the reader can click, or it should not be there.

What do AI introductions get wrong, and how do you fix them?

Below, a generic introduction written from the keyword alone, then the version written from the outline with formula 1. Both were written for this example.

Before: Logo design is one of the most important parts of any brand. Many designers struggle with knowing what to charge. This article will cover everything you need to know about pricing logo projects so you can set your rates with confidence.

After: Price a logo project from your hours, your rate and a buffer for revisions, then check the total against what the logo is worth to the client. Small local clients suit a fixed quote; funded companies can carry value-based pricing. Below: the formula with a worked example, a quote checklist and three replies for “that’s too expensive”.

Pattern in the AI introWhy it failsFix
Opens with how important the topic isThe reader already decided it matters by searchingOpen with the answer sentence
Names the reader’s struggleTells them what they know and delays the answerMove the struggle into the condition sentence or cut it
Promises “everything you need to know”Empty, and every competing page says itName three things the page actually contains
Opens with an unsourced statisticModels predict likely numbers; they do not look them upUse a number from the page’s own example, or a linked source
Restates the title as a sentenceUses the most-read line for nothingReplace with the condition: what the answer depends on

The statistic row is the dangerous one. Google’s guidance on AI content says generative models predict a likely sequence of words, so a confident percentage in an AI introduction is a guess until someone finds its source. The full fact-checking pass belongs in the AI SEO editing checklist.

The outline turns a brief into a plan; these pages cover the steps on either side.

Related

SEO content brief template: the 21-field brief this page’s prompts take as input

Related

how to brief AI for SEO content: the reasoning behind each brief field

Related

AI SEO content workflow: every step from keyword to Search Console, with time per step

Related

AI SEO content writer tools compared: tools that draft from your outline, with prices checked 5 Nov 2026

FAQ

Should AI write the outline, or should I?

Let the model propose it from your brief, then approve every heading and job yourself. The outline is where the angle and the evidence get locked in, so a person signs it off before any paragraphs exist.

How many H2s should an AI content outline have?

One per reader question in the brief. For a 1,500 to 2,500-word guide that lands around 6 to 9. Count the questions, then check the word budgets; a section with a budget under 150 words is a candidate for merging.

How long should the introduction be?

Two or three short paragraphs, under 120 words, with the answer in the first sentence. Nielsen Norman Group’s scanning research advises putting the most important points in the first two paragraphs.

Can a certain intro format win the featured snippet?

No format guarantees it. Google’s documentation says sites cannot mark a page as a featured snippet; its systems decide. A direct answer in the first sentence serves readers whether or not Google picks it.

Should I give the model an example outline?

Yes, when you want the same format every time. Anthropic’s prompting guide suggests 3 to 5 varied examples wrapped in their own tags. Use outlines from different topics so the model copies the format and leaves the content alone.

What should you do next?

Take the next article on your list, run prompt 1 with its brief, and put the result through the seven-point check before asking for a single paragraph. Then write the introduction with formula 1 and compare it with the intro your last article shipped with.