Heading hierarchy that AI quotes verbatim

## Why H2s and H3s matter to AI engines

When an AI engine answers "best CRM for startups", it doesn't read your whole page top to bottom. It scans section headings, pattern-matches them against the query, and lifts the matching sections back as the answer. Your headings are the index AI uses to find quotable content.

This means heading hierarchy isn't a style choice — it's the primary structural signal that determines what gets quoted. Two pages with identical prose can have 3-5x different citation rates based solely on heading structure. The signal is that strong, and most teams ignore it entirely because their CMS makes "heading" a styling decision rather than a semantic one.

## The rules that work

**One H1 per page** — the topic. Pages with multiple H1s confuse the AI's topic extraction. Most CMSes get this right by default; double-check after any template migration. The H1 should mirror the buyer query the page is targeting, not the brand voice you'd use in a hero section.

**Section H2s phrased as questions.** "How does X work?", "When should you use Y?", "What's the difference between A and B?". Question-phrased H2s match buyer queries letter-perfect and get lifted directly into AI answers. The pattern is so reliable that we use it as a diagnostic: pages with zero question-phrased H2s rarely break into the top citations for their topic.

**H3s frame the supporting points.** Each H2 typically contains 2-5 H3s that scope its subtopics. H3s should be short (under 50 chars) and ideally pattern-match secondary buyer queries. Treat each H3 as a sub-quotation the AI can lift in isolation — the whole section may be too long to surface in a single answer, but a well-framed H3 chunk gives the engine a clean handle.

**Real markup, not styled paragraphs.** Bold-uppercase paragraphs that LOOK like headings but aren't real `<h2>` / `<h3>` tags get ignored by AI engines. Use real markup or you lose the signal. This is a common failure mode in design-first CMSes that prioritise visual hierarchy over semantic hierarchy — the page looks structured to humans and looks like a wall of prose to crawlers.

**No skipping levels.** H1 → H3 (skipping H2) breaks the document structure AI engines rely on. Linter your CMS templates. Skipped levels also confuse screen readers, so the accessibility argument and the GEO argument align.

**Keep H2s declarative, not coy.** "The Surprising Truth About X" might be a good blog headline; it's a terrible H2 for AI. Engines can't pattern-match "surprising truth" to any specific buyer query. Lead with the question, not the tease.

## Audit your headings now

Run a platform scan from `/dashboard/audit`. The per-page report flags:

- Pages with no H2s at all - H2s longer than 80 chars (won't compress into AI answers) - H2s that are pure marketing ("Our Approach", "The Difference") — flagged as low-extractability - Pages with skipped heading levels - Styled paragraphs masquerading as headings - Pages where the H1 doesn't match the URL slug or the schema's `headline` field (entity-disambiguation risk)

Each issue ships with a specific rewrite suggestion you can apply or hand to our editorial team. The recommendations panel batches them by impact so you can ship the high-leverage fixes first.

## The ROI

For most B2B SaaS sites, rewriting H2s on the top 10 pages — keeping the prose unchanged — produces a 15-30% lift in AI citation rate within 60 days. It's one of the highest-leverage projects we can run, and it doesn't require any new content. The cost is one writer-day of work; the payoff is permanent.

## A worked example

Before: `## Our Approach` After: `## How does {product} optimise your workflow?`

Before: `## What We Do` After: `## What is {product} and who is it for?`

Before: `## Use Cases` After: `## When should you use {product} instead of {alternative}?`

Before: `## Features` After: `## Which features does {product} include in each plan?`

Before: `## Pricing Overview` After: `## How much does {product} cost?`

The new versions are queryable. The old ones aren't. AI engines see the new H2s, match them to buyer queries, and lift the sections beneath. Same content, dramatically different visibility.

## The interplay with schema

Heading structure and schema markup work together — they're not substitutes. A page with great H2s but no FAQPage schema still gets cited; a page with FAQPage schema but generic H2s still gets cited; a page with both gets cited 3-5x more often than either alone. The compounding is real, and it's why we recommend tackling them as one project rather than two separate sprints. The schema-vocabulary lesson covers the schema side; pair the two for maximum lift.

## Maintenance: the part nobody plans for

Heading structures drift. A product gets renamed, a feature gets dropped, a comparison gets stale, but the H2 stays the same and slowly loses match strength with current buyer queries. Quarterly review of your top 10 pages' headings — checking whether they still mirror what buyers are actually asking — keeps the structural signal fresh. Rankply's monthly audit flags drift automatically, but the discipline of the review is the thing that turns it from a recommendation into a habit.

## How long an H2 should be

There's a sweet spot: 30-70 characters. Under 30 and the heading is usually too vague to match any specific buyer query ("Overview", "Details", "More info"). Over 70 and the heading gets truncated in AI answer previews, with the cut-off point landing mid-thought and producing low-quality lifts.

Test heading: "How much does {product} cost and what's included in each plan?" — 61 characters, clean question framing, two distinct sub-queries baked in. AI engines pull this kind of H2 into both pricing and feature comparison answers simultaneously.

## H2s for different page types

The pattern varies by intent:

**Product pages.** H2s should answer the buyer's pre-purchase questions explicitly: "What does X do?", "How does X work?", "How is X different from Y?", "How much does X cost?", "Who is X for?".

**Comparison pages.** H2s should mirror comparison-shaped queries: "How does X compare to Y?", "When should you choose X over Y?", "What are the key differences between X and Y?".

**Pricing pages.** H2s should answer pricing-shaped queries: "How much does X cost?", "What's included in each plan?", "Is there a free trial?", "Does X offer enterprise pricing?".

**Blog / educational content.** H2s should mirror the "how to" or "what is" queries the article addresses. Avoid clever titles; the engines reward direct framing.

**About / company pages.** H2s should answer entity-disambiguation queries: "Who is X?", "Where is X based?", "When was X founded?", "Who runs X?".

Every page type has a different set of buyer queries it's meant to answer; the H2s are how you signal that to the engines.

## How AI engines handle very long sections

When a section under an H2 is more than ~500 words, the engine usually doesn't lift the whole thing — it picks a sub-passage. The selection bias favours the first paragraph, the first bullet list, and the most numbers-dense passage. Knowing this changes how you write: front-load the answer, use a tight bullet list early, include a concrete stat in the first 200 words.

Long sections without these scannable anchors get summarised rather than quoted, which produces lower-quality citations (your phrasing gets paraphrased; your specific numbers get rounded).

## When to ignore the rules

There's one important exception: editorial content where the storytelling matters more than the quotability. Founders' essays, customer narratives, brand-defining manifestos — these can use less direct H2s because they're not optimising for buyer-query extraction. The cost is lower AI citation rate; the gain is voice and memorability. For these pieces, accept the trade-off consciously rather than forcing question-shaped H2s into prose that doesn't want them.

The 80/20 rule applies: 80% of your pages should follow the answerable-headings playbook (product, pricing, comparison, support, FAQ). The other 20% can break the rules deliberately if the editorial value warrants it.