<b>Is "information gain" actually measurable, or just a patent buzzword?</b>

<b>Is "information gain" actually measurable, or just a patent buzzword?</b>

Google's 2020 information-gain patent describes ranking a document by how much <i>new</i> information it adds versus documents the user already saw. The question for us: can we proxy it before publishing?

I tested one crude proxy. For 60 target queries I took the top 10 results, extracted every distinct claim/statistic/entity, and built a "coverage set." Then I scored 60 of my own drafts by what fraction of their content was NOT already in that coverage set — a rough novelty ratio.

— Drafts in the top novelty tercile reached their peak position a median of 11 days sooner.
— They also earned 2.3x more unsolicited backlinks over 6 months.
— But novelty above ~40% showed no extra benefit and sometimes hurt — likely because "novel" past a point means "off-topic."

The takeaway: information gain looks real and partially proxy-able, but it's a curve, not a slope. Saying something the SERP hasn't said helps; saying something the SERP isn't about doesn't.

Method note: claims extracted manually for 20 queries, LLM-assisted for 40 (spot-checked).

Confidence: low-medium — proxy is noisy, sample modest.