Information Gain Scoring

In 2018 Google was granted US patent 10,108,919, 'Search result ranking based on contextual factors.' The relevant claim describes scoring a candidate document by the additional information it provides relative to documents the user has already seen. The patent is not a confirmation that this exact mechanism is in production, but it is a clear public statement of intent.

In 2026, with retrieval-augmented LLMs everywhere, the operational pressure is the same regardless of whether Google ships that exact ranking factor. A model that has already retrieved Page A has no incentive to also retrieve Page B unless Page B says something Page A did not. Information Gain is the metric for that incremental value.

Our /methodology page contains the formulas; here is the operator's intuition for each.

Lexical novelty

Different vocabulary signals a different angle. We compare your top 12 unigrams and 8 bigrams against the competitor's, after stop-word removal. This is not semantic similarity — it is lexical surface diff. It catches the case where two pages say the same thing in different words (low novelty) and flags the case where you genuinely use different terminology (high novelty).

Verifiable facts

We count four classes: 4-digit years, currency amounts, percentages, time durations. More verifiable facts = more useful to extractive systems. Unverifiable claims (vague adjectives, marketing language) score zero.

Original-data signals

First-person research markers ('we tested,' 'our 2026 sample,' 'internal data,' 'we measured') are the strongest single predictor of LLM citation in our test set. The marker has to be backed by actual data on the page; pages claiming originality with nothing underneath get downgraded by helpful-content systems.

Topical depth

More sentences on the topic means more discrete claims. Past a point this becomes filler; we cap the contribution at 15 points.

Answerable structure

Question marks and explicit headings are a proxy for FAQ density. Pages built question-first are extracted preferentially.

1. 1Pick the top 5 highest-revenue or highest-traffic URLs.
2. 2For each, identify the current #1 result for the primary query.
3. 3Run both through /tools/information-gain-auditor.
4. 4If your score is < 65 or the gap to competitor depth > 30%, add the page to the rewrite queue.
5. 5For pages in the rewrite queue, do an IG diff: list 5 things the competitor does not say, draft those as new H3s.
6. 6Publish, change the dateModified, ping IndexNow.