The original January 2026 observations are preserved below. The responsible conclusion is narrower than the original headline: H1s are useful page structure, but this dataset does not prove they independently cause higher rankings.
The January 2026 benchmark
| Reported value | Observation | Published coefficient |
|---|---|---|
| 93.5% | Reported with a single H1 | −0.1172 |
| 12.5% | Reported with multiple H1s | +0.3078 |
| 13.5% | Exact keyword in H1 | −0.2670 |
| 88.5% | Partial keyword match | −0.0282 |
The original page reports 93.5% with a “single H1” and 12.5% with multiple H1s. Those categories overlap or exceed 100% if interpreted as mutually exclusive, so raw field definitions must be verified before converting the percentages into counts.
How to interpret the benchmark
Correlation is not causation
The coefficients do not isolate the H1 from content quality, links, brand strength, page type, or search intent.
Rank encoding controls direction
If rank 1 is best, the sign of a coefficient must be interpreted against that encoding. The legacy documentation is not clear enough to turn these archived values into directional recommendations.
Small samples require restraint
Five keyword sets can reveal hypotheses worth testing. They cannot establish a universal rule for every query or industry.
The August 2026 follow-up
The August 2026 study covers 916 traditional observations and 1,645 AI observations across the same 31 queries. It uses a harmonized rendered-DOM standard and keeps the historical Google benchmark separate.
The current H1 comparison across traditional search and AI citations found H1 tags on 97.8% of AI-cited pages versus 98.1% of traditional results, an almost identical presence rate that does not establish an independent visibility effect.