How to structure headings to be cited by AI?

Headings (H1, H2, H3) function as navigation anchors for RAG systems — they delimit where each block of information begins and ends, making it easier to extract specific answers. A question about "delivery time" in an article with the heading ## Delivery time and logistics is much easier to locate and extract than the same information buried in an unheaded paragraph. Heading structure is not just a matter of visual formatting — it's citability infrastructure.

How RAG systems use headings

When a RAG system processes a document, it segments the text into chunks. Headings are the natural dividers between chunks: the content between one H2 and the next H2 tends to be treated as a semantic unit. This has direct implications:

• An H2 with text relevant to the query receives a high score in the retrieval stage
• The content immediately after the heading is the first candidate for extraction
• Vague headings (like "Introduction" or "Final Thoughts") produce low-value semantic chunks

Practical rules for efficient headings

H1: single title, direct question or statement

The H1 should reflect exactly what the page answers. The most efficient pattern for AIO is the question format — which mirrors how users phrase queries in AIs.

Less efficient	More efficient
"Business credit"	"How to get a business loan without collateral?"
"Our solutions"	"What HR outsourcing services exist for SMEs?"
"Learn about rental"	"How does short-term rental work for property owners?"

H2: subtopics that answer aspects of the main question

Each H2 should function as a partial, self-contained answer. Someone who reads only the H2 and the paragraph immediately below it should leave with complete information.

Example for an article about auto financing:

## Which lender has the lowest interest rate for car financing?
## What is the minimum down payment required by lenders?
## How to calculate the maximum monthly payment relative to income?
## Can I get a car loan with bad credit?
## Auto loan vs. lease: what's the practical difference?

Each H2 is a question a car buyer would ask ChatGPT. The article, structured this way, captures multiple queries with a single document.

H3: specific details within each subtopic

H3 is for granularity — when a subtopic has variants, special cases, or examples that deserve their own highlight.

Example for an article about service contracts:

## Required clauses in a service agreement
### Confidentiality clause
### Non-compete clause
### Unilateral termination clause
### Late delivery penalty

What to avoid in headings

Generic headings with no semantic value:

• "Introduction" → replace with a statement or question about the topic
• "Conclusion" → replace with "What to do from here" or similar
• "See more" → that's a CTA, not a heading

Headings that promise more than they deliver:

If an H2 says "Everything about inventory management for restaurants" and the content is a 100-word paragraph, the RAG system learns that this source has low density — affecting the overall page score.

Excessive headings without content:

Three lines of text between each H2 are not enough for the chunk to be valued. The ideal is at least 150–200 words per H2 section.

Headings as a strategy for capturing long-tail queries

One of the most efficient AIO strategies is using H2 and H3 in the format of long-tail questions that mirror how users formulate queries for AIs. An article about corporate health insurance might have:

## Is group health insurance required for companies with fewer than 10 employees?
## What is the difference between co-pay and deductible in health insurance?
## How does the waiting period work in group health insurance?
## Can employees include dependents on the company plan?

Each heading captures a different query that potentially leads to the same article — multiplying citation opportunities.

FRT Digital structures heading architecture as part of content optimization in the AIO service. To identify which articles on your site have suboptimal heading structure, start with the AIO Score Audit.