Overview
AI tools are changing how insurance consumers find answers. I wanted to know if the way we write affects how often those tools cite us. So I tested it.
Hypothesis
LLMs cite claims, not paragraphs. A clean, self-contained sentence is extractable. A long, multi-clause sentence loaded with legal language is not. Insurance copy defaults to the latter. I hypothesized that simplifying sentence structure — not substance — would increase AIO appearances and LLM citations without touching traditional rankings.
Approach
I edited four motorcycle insurance pages and one snowmobile article. Three strategies:
Break multi-clause sentences into shorter, extractable units
Replace legal phrasing with plainer equivalents without changing meaning
Cut filler and swap longer words for shorter ones
Every edit stayed within Progressive's guidelines and keyword targets.
5
Pages tested
4 of 5
Crossed readability threshold
875%
Citation growth for MA
Results
Pages that crossed a readability category — Difficult into Fairly Difficult — saw the biggest jumps:
Indiana: AIO +200%, LLM Citations +136%
Kentucky: AIO 0 → 5, LLM Citations +850%
Massachusetts: AIO 0 → 13, LLM Citations +875%
Snowmobile: AIO +67%, LLM Citations +149%
California improved its score but stayed within "Difficult." AIO dropped 15%. LLM citations still grew 72%. The exception that proved the rule.
Finding
The score didn't matter as much as the category. Crossing a readability threshold produced step-function jumps in citations. Staying within one produced modest or mixed results.
Traditional rankings held steady throughout — confirming the hypothesis. We changed something language models care about without touching what traditional search weighs.
For regulated industries with language constraints, the play isn't to maximize readability. It's to find the pages closest to a category threshold and edit them across it first.