Skip to main content Scroll Top

AEO Content Writing: Six Rules at Sentence Level

Updated September 2026 · Written and maintained by the Progression Agency strategy team

Four things genuinely differ from SEO copywriting, three were always true, and two were never a good idea — plus the two-minute test that catches most extraction failures.

On this page · 11 sections
  1. Writing for extraction is mostly just writing well
  2. What actually differs from SEO copywriting
  3. The same section, written two ways
  4. The only test that matters
  5. What to restructure first
  6. What to be sceptical of
  7. Everything else we have written on search, AI and getting found
  8. The same section, before and after
  9. The six rules, with the failure each one prevents
  10. What differs from SEO writing, and what does not
  11. Where to spend the rewriting effort

The short answerSix rules change at sentence level: the heading asks a question, the first sentence answers it completely and stands alone, no pronouns reach backwards out of the passage, numbers carry their source inline, one idea per section, and context comes after the answer. Every one of them improves the page for human readers too, which is why a premium rate for ‘AI-optimized’ writing is hard to justify. Almost all of this is restructuring pages you already have rather than publishing new ones.

Page-type return figures are our own observation from client work, labelled as such. Retrieval internals are not published by any provider; the rules here follow from passage-level selection, which is documented.

The six sentence-level rulesThe six sentence-level rules
All six improve the page for human readers too, which is the honest reason to do them rather than a coincidence.

Writing for extraction is mostly just writing well

Six rules change at sentence level, and every one of them improves the page for human readers too. That is not a happy coincidence — it is the reason to be sceptical of anyone charging a premium for ‘AI-optimized’ content.

The rules: the heading asks a question, the first sentence answers it completely and stands alone, pronouns never reach backwards out of the passage, numbers carry their source inline, one idea per section, and context comes after the answer rather than before it.

Almost all of this is restructuring pages you already have. No new words are required, which makes it a rewriting job rather than a publishing programme.

Rule 1 — Heading asks a question. Or makes a specific promise.
Rule 2 — First sentence answers it. Completely, and alone.
Rule 3 — No backward pronouns. They break when lifted.
Rule 4 — Numbers carry sources. Inline, not elsewhere.
Rule 5 — One idea per section. Two answers, neither extracts.
Rule 6 — Context after. Not before the answer.
What differs from SEO copywriting, and what does notWhat differs from SEO copywriting, and what does not
Four genuine differences, three things that were always true, two that were never a good idea. This is a smaller change than most ‘AI content’ offerings imply.

The heading asks a question

Or makes a specific promise. A heading reading ‘Our approach’ promises nothing a retrieval system can match a question against.

The first sentence answers it

Completely, with no preamble. If the reader needs the paragraph above to understand it, retrieval cannot use it either.

No pronouns reaching backwards

‘It’, ‘this’ and ‘they’ are fine inside a passage and fatal across its boundary. A lifted passage beginning ‘This means that…’ means nothing.

Numbers carry their source inline

Not in a footnote, not in a sources section at the bottom. An unsourced statistic is among the first things a cautious system declines to repeat.

One idea per section

Two answers under one heading usually means neither extracts cleanly, because the retrieved chunk contains half of each.

Context after the answer

Human readers tolerate context-then-answer. Retrieval does not, and answer-then-context reads better for skimmers anyway.

What actually differs from SEO copywriting

Four things genuinely differ: question headings, answer-first construction, pronoun discipline across passage boundaries, and inline sourcing. Three things were always true of good writing and still are. Two things were never a good idea and remain so.

That is a smaller change than most AI content offerings imply. If a brief arrives with keyword density targets and a monthly article quota attached to the words ‘AI optimization’, it is an SEO brief from 2014 with a new cover sheet.

Differs — Question headings. SEO tolerates labels.
Differs — Answer-first. SEO often builds to it.
Differs — Pronoun discipline. Only matters when lifted.
Differs — Inline sourcing. Stronger requirement here.
Same — Clear plain writing. Always was.
Same — Real subject knowledge. Always was.
Article quotas — Not a lever. Volume does nothing.
Keyword density — Not a lever. Obsolete for both channels.
FAQ padding — Not a lever. Unasked questions retrieve nothing.
Word count targets — Not a lever. Length is not structure.
AI-generated volume — Not a lever. Mistakes the lever for the goal.
Premium 'AI' rates — Not a lever. It is ordinary good writing.

The same section, written two ways

Identical information, identical length, entirely different extractability. The difference is structural rather than a matter of quality or effort, which is why this work is cheap relative to what it returns.

Marketing prose tends to hedge precisely where an answer should be — around price, scope and outcomes — because hedging feels safer commercially. It is also exactly where retrieval needs a clean claim, which is why pricing and service pages are usually the worst offenders.

Two ways to write the same sectionTwo ways to write the same section
Identical information, identical length. The difference is entirely structural, which is why this is a rewriting job rather than a writing-more job.
Extracts well — Direct answers. First sentence, complete.
Extracts well — Tables. Self-contained rows.
Extracts well — Plain definitions. Quotable in one line.
Extracts well — Sourced statistics. Survive the reliability filter.
Extracts badly — Hedged prose. No clean claim to lift.
Extracts badly — Narrative build-up. Answer arrives too late.

The only test that matters

Copy a section out on its own, paste it into a blank document, and read it as a stranger would. If it answers an identifiable question, every pronoun resolves inside it, and every number carries a source, it can be quoted. If it does not, nothing else you do will matter.

Two minutes per section, no tools required. This single test catches more extraction problems than any software we have used, because the failure is almost always obvious once the passage is separated from its surroundings.

The only test that mattersThe only test that matters
Six steps, two minutes per section. This single test catches more extraction problems than any tool we have used.
Copy the section out — Test. Nothing above or below.
Read with no context — Test. As a stranger would.
Resolve every pronoun — Test. Inside the passage.
Check every number — Test. Source inline.
Identify the question — Test. It should be obvious.
Holds alone? — Test. Then it can be quoted.

What to restructure first

Pricing pages and service pages, before anything else. They carry the highest-value questions and they are usually the worst written for extraction, because commercial caution produces hedging exactly where a clean answer belongs.

FAQ sections come third and are the cheapest win, since they are already question-shaped and often need nothing more than moving the answer up a sentence. Blog archives come last — long tail, low return per page.

Which content is worth restructuring firstWhich content is worth restructuring first
Top-right first. Pricing and service pages carry the highest-value questions and are usually the worst written for extraction, because marketing prose hedges exactly where an answer should be.
Where restructuring effort returns mostWhere restructuring effort returns most
Our observation, labelled as such. The top three are where most sites have the largest gap between what they know and what a machine can lift.
Priority 1 — Pricing pages. High value, badly hedged.
Priority 2 — Service pages. The commercial core.
Priority 3 — FAQ sections. Already question-shaped.
Priority 4 — Comparison pages. Tables extract cleanly.
Priority 5 — Definitions. Quotable, lower value.
Last — Blog archive. Long tail, low return.

Pricing pages

The highest-value question anyone asks about you, and the one most commonly answered with ‘it depends on your needs’. Give a band, state what moves it, then hedge.

Service pages

What you do, for whom, and what is included. Usually written as positioning rather than as answers, which extracts badly.

FAQ sections

Already in the right shape. The common fault is an answer that opens with ‘Great question’ or restates the question before answering it.

Comparison pages

Tables extract exceptionally well because a row is already a self-contained fact with its dimensions labelled.

Definitions

Highly quotable and lower commercial value. Worth doing, not worth doing first.

What to be sceptical of

A premium rate for ‘AI-optimized’ writing. Monthly article quotas aimed at citation. Keyword density targets in an AI brief. FAQ padding with questions nobody asks. And generating content at volume with AI in order to win AI citation.

That last one deserves naming directly. Producing large volumes of machine-written content to be cited by machines mistakes the lever for the goal — retrieval selects a passage on relevance and apparent reliability, and volume affects neither.

What to be sceptical of in AI content servicesWhat to be sceptical of in AI content services
The last row deserves naming. Producing large volumes of machine-written content to win machine citation is the clearest case of mistaking the lever for the goal.
Writing for extraction, in numbersWriting for extraction, in numbers
The zero in the middle is the commercial point. Almost all of this is restructuring pages you already have.

Want your highest-value pages restructured?

An AI visibility audit identifies which pages carry your most valuable questions and which of them currently fail the standalone test — so the rewriting effort goes where it returns most.

/ai-visibility-audit

Everything else we have written on search, AI and getting found

AI, AEO and what is changing

Websites and design

Choosing and working with an agency

Social, content and brand

By industry and by situation

Want this done for your site?We build and maintain the search, content and paid programmes described on this page.

Get a free proposal

The same section, before and after

One section in its original and rewritten forms, with every change annotated.

Identical information, identical length
BeforeAfter
HeadingOur pricing philosophyHow much does an AEO audit cost?
Sentence 1We believe every business deserves transparency.A published AEO audit costs $1,000 to $4,000 for 10 to 20 hours of work.
Sentence 2That is why we take a consultative approach.The range depends on how many commercial templates you have and whether they render without JavaScript.
Sentence 3Every engagement begins with a conversation.Both are testable in an afternoon, which is why we run them before quoting.
Pronouns reaching back‘That’, ‘Every’ — unresolvable aloneNone
Numbers sourcedNone presentRange attributed on the page
Survives alone?NoYes

The six rules, with the failure each one prevents

Six sentence-level rules, each paired with the specific failure it exists to prevent.

What goes wrong without each rule
RuleFailure it preventsHow to check it
Heading asks a questionNothing to match a query againstRead the heading alone — is it a question?
First sentence answers completelyRetrieval lifts a fragmentRead sentence one alone
No backward pronounsLifted passage opens with ‘This means…’Search the section for It/This/They
Numbers carry sources inlineStatistic dropped at the reliability stepEvery figure has an attribution beside it
One idea per sectionChunk contains half of two answersCount the questions the section answers
Context after the answerAnswer arrives too late to extractCheck what sentence one is doing

What differs from SEO writing, and what does not

Most of this is ordinary good writing; these are the places where it genuinely diverges.

A smaller change than most AI content briefs imply
PracticeSEO writingAEO writingVerdict
Question headingsOptionalRequiredDiffers
Answer positionOften built towardFirst sentenceDiffers
Pronoun disciplineRarely consideredRequired across boundariesDiffers
Inline sourcingGood practiceMaterially affects selectionDiffers
Clear plain writingRequiredRequiredIdentical
Subject knowledgeRequiredRequiredIdentical
Logical structureRequiredRequiredIdentical
Keyword densityObsoleteIrrelevantNeither
Publishing cadenceDiminishingNot a leverNeither

Where to spend the rewriting effort

Not every page repays a rewrite, and these are the ones that do.

Ordered by return, from our own work
Page typeReturnWhyTypical fault
PricingHighestHighest-value question asked about you‘It depends on your needs’
ServiceHighestThe commercial corePositioning rather than answers
FAQHighAlready question-shapedRestates the question before answering
ComparisonHighTables extract cleanlyProse where a table belongs
DefinitionsMediumQuotable, lower commercial valueApproached gradually
Case studiesMediumEvidence, if made specificAnonymised into uselessness
Blog archiveLowLong tail, low per-page returnEverything above, at scale

Getting found in search

AI, AEO and what is changing

Paid media and lead generation

Websites and design

Choosing and working with an agency

Social, content and brand

By industry and by situation

Frequently asked questions

How do I write content that AI will cite?
Six rules at sentence level: the heading asks a question, the first sentence answers it completely and stands alone, no pronouns reach backwards out of the passage, numbers carry their source inline, one idea per section, and context comes after the answer rather than before it.
Is AEO writing different from SEO writing?
Four things genuinely differ — question headings, answer-first construction, pronoun discipline across passage boundaries, and inline sourcing. Clear writing, real subject knowledge and logical structure were always required. Keyword density and publishing cadence were never good ideas and still are not.
Do I need to write new content?
Usually not. Almost all of this is restructuring pages you already have, which is why it is a rewriting job rather than a publishing programme.
What is the single most important rule?
The first sentence of a section must answer the question its heading asks, completely, and stand alone without the paragraph above it.
How do I test whether a passage will extract?
Copy the section out on its own, paste it into a blank document, and read it as a stranger. If it answers an identifiable question, every pronoun resolves inside it and every number carries a source, it can be quoted.
Why do pronouns matter so much?
Because retrieval lifts a passage out of its surroundings. A lifted passage beginning ‘This means that…’ refers to nothing, and a system is unlikely to use it.
Why does sourcing affect citation?
Passages are filtered on apparent reliability before an answer is composed. An unsourced statistic is among the first things a cautious system declines to repeat.
Which pages should I restructure first?
Pricing and service pages. They carry the highest-value questions and are usually the worst written for extraction, because commercial caution produces hedging exactly where a clean answer belongs.
Why are pricing pages usually the worst?
Because ‘it depends on your needs’ feels safer than a number. Give a band, state what moves it, then add the caveats — that structure is both more useful and more extractable.
Do tables help?
Considerably. A table row is already a self-contained fact with its dimensions labelled, which is close to the shape retrieval wants.
Should I pay a premium for AI-optimized content?
Be sceptical. Most of what makes content extractable is ordinary good writing plus four structural rules. A premium rate for applying them is difficult to justify.
Do article quotas help AI visibility?
No. Volume is not the lever — retrieval selects a passage on relevance and apparent reliability, and neither is affected by how many articles you publish.
Does keyword density matter?
No. Similarity is computed on meaning rather than term frequency, so repeating a phrase does not increase your odds of retrieval and tends to make the writing worse.
Should I pad my FAQ with more questions?
No. Questions nobody asks retrieve against nothing. Better to answer ten real buyer questions properly than forty invented ones badly.
Can I use AI to write content that AI will cite?
You can use it as a drafting tool, but generating volume with AI specifically to win AI citation mistakes the lever for the goal. Structure and reliability drive selection; provenance of the draft does not.
How long should a section be?
Long enough to answer its heading and short enough to contain one idea. Length is not the variable — self-containment is.
Where should context and background go?
After the answer. Human readers cope with either order and skimmers prefer answer-first, so this costs nothing in readability.
What about case studies?
Worth restructuring if they contain specifics. Anonymised case studies with no named figures extract poorly because there is no concrete claim to lift.
How do I brief a writer for this?
Give them the six rules, the standalone test, and the page-type priority order. Do not give them a keyword density target or a word count.
Does this hurt readability?
No, and generally the opposite. Answer-first writing suits skimmers, resolved pronouns are clearer, and inline sourcing builds trust with human readers too.
What is the fastest improvement I can make today?
Take your pricing page, find the section whose heading asks what something costs, and move the actual number into the first sentence. That single edit is usually the largest single gain available.
How do I know it worked?
Extraction is checkable by reading, within days. Citation frequency takes months and needs a dated baseline with verbatim answers, which is a separate discipline from the writing itself.

Want this done for your site?We build and maintain the search, content and paid programmes described on this page.

Get a free proposal

Get a free marketing proposal

Tell us what you are trying to grow and we will come back with a plan, not a pitch deck. Same-day reply on weekdays.

Our website uses cookies from third party services to improve your browsing experience. Read more about this and how you can control cookies by clicking "Privacy Preferences".
Privacy Preferences
When you visit our website, it may store information through your browser from specific services, usually in form of cookies. Here you can change your privacy preferences. Please note that blocking some types of cookies may impact your experience on our website and the services we offer.
Privacy Policy
You have read and agreed to our privacy policy
Required
Contact Us