Information gain is the new information a page adds compared with other pages on the same topic. A page with high information gain says something the others do not: a figure, a method, a result, an example or a clear conclusion. Both Google’s guidance on helpful content and the way AI assistants choose sources favour such pages, because repeating what ten other sites already say gives a reader, or a model, no reason to pick yours.
Where the idea comes from
In information theory, information gain measures how much uncertainty is removed by learning something new. In search, the term became common after Google was granted a patent in 2022 describing how to score documents by how much new information they would give a user who has already seen other documents on the topic.
Patents show which ideas a company has explored, and many are never used in production. The more useful evidence is Google’s own public guidance on creating helpful content, which asks site owners whether their content offers original information, research or analysis, whether it goes beyond the obvious, and whether it adds substantial value compared with other pages in the results.
Why AI search rewards it
When an AI assistant writes an answer from several retrieved passages, facts that appear in many sources are merged into one sentence and can be credited to any of them. A fact that only one page contains has one possible source. That page gets the citation.
The 2023 GEO research paper by Aggarwal and colleagues found that adding statistics, quotations and cited sources raised a page’s visibility in AI-generated answers by up to about 40% in their tests. Each of these is a form of information that a generic page lacks.
What counts as information gain
| Type | Example |
|---|---|
| Own data | Average load time across 40 sites you audited, before and after a fix |
| Prices and ranges | What a service typically costs and what makes it cost more |
| Method | The exact steps you follow, including the order and why |
| Results | What happened when you tried something, including when it did not work |
| Comparison | A table comparing options on criteria nobody else has laid out |
| Clear conclusion | A direct recommendation for a stated situation, with reasons |
| Correction | Evidence that a widely repeated claim is wrong or outdated |
Longer content is not the same as more information. A 3,000-word page that restates the same five points as every competitor has less information gain than a 600-word page with one original finding.
How to find your information gain
- Search your target question and read the top results and the AI answers for it.
- List every point they make. This is what the topic already contains.
- List what you know that is missing: from projects, client questions, data, mistakes you have seen.
- Build the page around the second list, and cover the first list briefly for completeness.
- Put your new points near the top of their sections, where both readers and models find them first.
The guides on this site follow the same approach. Several of them correct common claims, for example that llms.txt improves AI rankings or that text-to-HTML ratio is a ranking factor, because a documented correction is information most pages on those topics do not offer.
Frequently asked questions
Is information gain a confirmed Google ranking factor?
Google has not confirmed a ranking signal by that name. Its public guidance asks for original, substantial content, and the 2022 patent shows the idea has been studied. It is best treated as a description of what the guidance rewards.
Can AI-written content have information gain?
Only if the information comes from somewhere new, such as your data or experience. A model writing from its training data tends to reproduce what is already common, which is the opposite of information gain.
Do I need original research to rank?
No. Original research is the strongest form, but first-hand experience, specific examples and clear conclusions also add information that generic pages lack.