Information gain describes how much new information a sentence or paragraph contributes — measured against what an AI system already knows or obtains from other sources. A paragraph that repeats what twenty other pages say has an information gain close to zero. A paragraph with its own figure, its own measurement or a concrete experience has a high gain.
How does information gain work?
When generating an answer, the model has several passages from different sources in front of it. What gets cited is whatever carries a statement the answer needs and that other passages do not provide. If a passage merely confirms what the model knows from its model knowledge, it flows in without a mention or is not used at all. Information gain therefore does not decide whether you are found, but whether you receive a citation.
Closely related is the term uniqueness: does the page contain information that is not already in the model knowledge and for which only you qualify as the source?
Why does information gain matter for AI visibility?
Because standard definitions and general advice texts have long been in the training data. Anyone who writes them again competes with the model’s memory — and loses, because the model needs no source to say something it already knows. Content succeeds in AI answers only with new figures, data, facts and perspectives. That applies to technical articles as much as to product pages.
What does this mean for your website?
Get to the point and reduce promotional language to a minimum — every sentence should contribute something. Publish your own data: studies, surveys, measurement series, experience from projects, concrete prices and the limits of your service. Make authorship and evidence visible (E-E-A-T) so that the uniqueness is also credible. And check existing pages with one question: which sentence here appears nowhere else?