LLM Readability Content Rules

Every pronoun in a passage is a bet that whoever reads it will still have the previous sentence. Ivan Sp writes and edits content against a fixed set of structural rules that make text parseable by language models. The rules govern entity naming, sentence construction, heading hierarchy, and internal linking, and they apply to new writing and to rewrites of existing pages alike.

The Six Rules

Six rules govern every page written for AI readability:

State entity, attribute, and value explicitly. Write the relationship as a complete claim, in the shape of “[orthodontist] [specializes in] [Invisalign treatment],” rather than implying it.
Use subject-verb-object construction. Named subject, active verb, specific object.
Name the entity instead of using pronouns. Write “West Edmonton orthodontists” rather than “they,” because a passage read in isolation has no antecedent available.
Keep heading hierarchy sequential. Move from H1 to H2 to H3 without skipping levels.
Answer first under every heading. The direct answer occupies the first one or two sentences.
Avoid node-to-node linking. A deep page should not link straight to another deep page when the parent has never introduced that entity.

Why Pronouns Break Extraction

A pronoun resolves against something earlier in the document. Extraction systems read a chunk without the document around it, so the antecedent is simply gone. “They offer same-day appointments” becomes a claim with no subject attached.

The correction is not repeating a full business name in every sentence, which reads mechanically. Naming the entity once per paragraph, at first mention, resolves the chunk without damaging the prose.

Headings That Survive Isolation

Every heading gets tested by pulling it out of context and asking whether it still describes the section below. Vague headings fail immediately, and so do metaphorical ones.

Metaphors cause a specific problem. A heading such as “Our Compliments to the Chef” embeds poorly, because the vector it produces sits nowhere near any query a person would type. Rewriting to name the entity and the topic fixes both the embedding and the reader’s experience.

One production pass processed 3,600 H2 headings in sequential batches. Each rewrite recorded the old heading, the new heading, and the stated reason for the change, so the work stayed auditable rather than becoming a silent bulk replacement.

The Node-to-Node Linking Rule

Site structure carries semantic meaning, and linking flattens or reinforces it. Linking a deep specific page directly to another deep specific page, with no parent having introduced that entity, tells a crawler the two sit at the same level in an undifferentiated pile.

Links therefore travel through the hierarchy. A child page links to its parent, and the parent introduces sibling entities before any child references them. That discipline is why this page links upward to AI search optimization as the parent service rather than sideways to unrelated deep pages.

Reading Level and Sentence Length

Body copy targets a grade 8 reading level unless a topic genuinely requires technical depth. Sentences stay at 25 words or fewer, which is an extractability constraint rather than a comprehension one. Short declarative sentences get lifted verbatim into answers, and long compound sentences get paraphrased or skipped.

The Rules in Full

Rule one, name the entity: The subject is named explicitly rather than carried by a pronoun.
Rule two, complete the claim: The core claim finishes inside one sentence.
Rule three, answer first: The direct answer opens the section, with context beneath it.
Rule four, headings stand alone: Each heading describes its section without the surrounding page.
Rule five, consistent terminology: One term per concept, used the same way throughout.
Rule six, no node-to-node links: A parent page introduces a topic before deep pages link across.
Sentence length: Twelve to twenty words, which keeps a claim inside a single readable unit.
What the rules do not do: Improve a wrong or thin answer. Structure carries content rather than replacing it.

Test One Paragraph Against the Rules

Ivan Sp suggests a two-minute test on any paragraph you are pleased with.

Copy one paragraph from your strongest page into a blank document.
Delete the first sentence.
Read what remains and check whether the subject is still identifiable.
Count how many pronouns now have no antecedent anywhere in the text.

Most paragraphs collapse entirely when the first sentence goes. Naming the subject once more in the middle is usually the whole fix.

Where the Standard Readability Advice Breaks Down

How many sentences on your page would still make sense with the previous one deleted?

The usual advice: Write at a lower reading level, since simpler text is easier for everyone to process.
What holds up in practice: Reading level and machine readability measure different things. A simple sentence built on pronouns reads easily in sequence and cannot be extracted. A longer sentence naming its subject extracts cleanly.

Consider a page rewritten to a lower reading grade. Sentences get shorter and every repeated noun is replaced with a pronoun to reduce repetition.

The score improves and the page becomes less extractable. Each short sentence now depends on the one before it for its subject.

I keep sentences short and name the subject anyway. Repeating a noun reads slightly heavier and survives being quoted alone.

Optimising for a readability score can make a page harder for the systems that summarise it. The two goals diverge exactly where pronouns are concerned.

A page written entirely in pronouns reads well and cannot be quoted anywhere. Nothing in a readability score will warn you about that.

Four Things Commonly Assumed About Writing for Machines

Which of these rules does your content already follow?

I meet these four assumptions whenever readability is treated as the goal.

Commonly assumed	What actually holds
Simpler text is more machine readable.	Reading level measures sentence and word complexity. Extraction depends on whether a sentence carries its own subject.
Repeating a noun is poor writing.	Repetition reads slightly heavier and keeps a sentence self-contained. Pronoun substitution is what breaks extraction.
Shorter sentences are always better.	Twelve to twenty words fits a complete claim. Cutting below that usually splits one claim across two dependent sentences.
Synonyms make writing less repetitive.	Synonym drift makes a machine treat one concept as several. One term per concept is what keeps the meaning stable.
These rules make writing sound robotic.	Naming subjects and completing claims is ordinary clarity. The rules constrain structure rather than voice.

Every rule here costs a few words and buys a passage that can be quoted. The page still reads like a person wrote it, because a person did.

A Worked Example: The Same Paragraph Under the Rules

Consider one paragraph rewritten against the six rules. The comparison below is illustrative, written to show what changes. No client engagement is described here.

Property	Before	After
Sentences	5	5
Average sentence length	11 words	16 words
Pronouns carrying the subject	4	0
Sentences complete on their own	1 of 5	5 of 5
Distinct terms used for one concept	3	1
Reading grade	Slightly lower	Slightly higher
Extractable passages produced	0	3

I accept the slightly higher reading grade. The final row is what the trade buys, and no human reader notices the difference.

Pronoun-heavy writing is a set of instructions where every step says do the same again. Perfectly clear in order and meaningless from step four.

The terminology row matters more than it looks. Three words for one concept splits the evidence three ways in anything building a model of the page.

How the Rules Get Applied

I apply the rules to existing content rather than requiring a rewrite.

Extract the heading outline and confirm each heading describes its section alone.
Read the first two sentences under each heading and confirm the answer appears there.
Highlight every pronoun and identify where its antecedent actually sits.
Replace every pronoun whose antecedent sits in a previous paragraph.
Find claims split across two sentences and combine them into one complete statement.
List every term used for each core concept, and pick one to use throughout.
Replace synonym drift with the chosen term, including in headings and metadata.
Check sentence length sits between twelve and twenty words for the claim-carrying sentences.
Confirm no deep page links across to another deep page without the parent introducing it.
Re-read three passages in isolation and confirm each states something complete.

How This Engagement Is Scoped

I scope rule application by page count, since the work is sentence-level.

What arrives: A rule-by-rule pass per page, with every changed sentence shown against the original.
Typical duration: Three to five working days for a set of fifteen pages.
What the client supplies: The page list, and subject review where a combined claim changes emphasis.
Included: Outline check, pronoun replacement, claim consolidation, terminology unification, and link review.
Not included: Writing new content where a page never made the claim.
What changes the estimate: Very long pages, or content with heavy technical terminology to unify.
How it is quoted: After a short scoping call, once the page list is agreed.

Tools and Primary Sources

The references below cover the standards behind these rules. Each opens in a new tab.

Google AI features documentation describes how AI surfaces draw on and attribute passages.
Google featured snippet documentation describes passage selection, which rewards the same self-containment.
Creating helpful, people-first content is the quality standard structure exists to carry rather than replace.
WCAG headings and labels is the accessibility requirement behind the standalone heading rule.

Common Questions About LLM Readability Content Rules

Do these rules make writing sound robotic?

They can, when applied mechanically. Naming an entity in every sentence, or opening every section identically, reads as machine-generated. The rules set a floor for clarity, and sentence variety still has to be written in deliberately.

Do the same rules apply to Google’s classic ranking systems?

Largely yes. Explicit entity relationships, clear sentence structure, and sequential headings help traditional natural language processing as much as they help language models. Very little here trades one system against the other.

Can existing pages be fixed without a full rewrite?

Usually. Most failures are pronoun replacements, heading rewrites, and sentence reordering rather than new content. A full rewrite gets recommended only when the underlying facts are missing.

Do these rules apply to every page?

The first five apply anywhere text is meant to be found and quoted. The linking rule is structural and applies at site level rather than per page.

Discuss This Work

The first call is a free 30-minute consultation with no pitch attached. You receive an honest assessment of where the site stands and what can realistically be improved.

Book a free 30-minute consultation