An analysis of 6.8 million subheadings reveals a measurable correlation between specific heading structures and ChatGPT citation rates.

While we often focus on overall word count or keyword density, subheading formatting appears to act as a significant sorting signal for AI retrieval systems.

We examined H2 to H4 headings across 815,000 query and page pairs to identify patterns associated with higher AI visibility.

Here is what the data indicates:

1. Question Formats Show Higher Alignment

ChatGPT fanout queries are frequently phrased as questions. Consequently, question style headings appear to align more naturally within the embedding space, matching fanout queries at 1.5x the rate of declarative headings.

1. The 20 to 39 Character Range Yields Peak Rates

Heading length shows a clear impact on performance. The 20 to 39 character range correlates with the highest citation rate at 32.7 percent. Conversely, headings over 60 characters correspond with a 6 percentage point drop in citation rates, while extremely short headings under 20 characters (like "Overview") exhibit poor matching capabilities.

The Practical Application:

Testing a structure where a quarter to half of your subheadings are phrased as user questions is a data-backed starting point.

For example, "What does X cost?" shows better retrieval alignment than both the overly broad "Pricing" and the diluted "A complete breakdown of pricing options for X".

Structuring your H2s to mirror natural user prompts rather than traditional blog titles may significantly improve your probability of being cited.

(All details in this week's Growth Memo)