Information Gain: Why AI Cites Some Pages and Not Others

AI answers cite the pages that add something new. What information gain means, where the idea comes from, and how to add it to your pages.

Indexed Research

Research team · 6 min read

Published

Key Takeaways

  • Information gain is how much new information a page adds beyond what other sources on the same topic already say.
  • AI answers combine several sources and cite the ones that support specific claims, so pages that only repeat others are easy to skip.
  • The term comes from a Google patent granted in 2022, but Google has never confirmed it as a ranking factor.
  • Google's helpful content guidance asks whether a page offers original information and substantial value compared with other results.
  • Original data, first-hand testing, fresh dated facts and worked examples are the most reliable ways to add information gain.

Ask ChatGPT, Perplexity or Google's AI Mode a question in your industry and look at the sources it cites. Often they aren't the biggest sites or the longest guides. They're the pages that said something the others didn't: a number, a test result, a worked example, a clear opinion with evidence behind it. That quality has a name in SEO, information gain, and it's one of the most useful ideas for deciding what to write if you want AI systems to cite you.

Quick answer: Information gain is how much new information a page adds beyond what a reader, or an AI system, has already seen in other sources on the same topic. It matters for AI citations because an answer engine builds one response from several sources. A page that repeats the other sources gives it nothing extra to cite, while a page with original data, testing or specific detail does.

Want to know whether AI tools cite your site today? Run a free check with our LLM visibility checker, or see how Indexed approaches AI SEO.

Where the Term Comes From

The phrase entered SEO through a Google patent titled "Contextual estimation of link information gain" (US 11,354,342 B2), filed in October 2018 and granted in June 2022. It describes scoring a document by how much additional information it contains compared with documents the user has already seen, then using that score to decide which documents to show next. The patent's detailed description also mentions information being delivered through an automated assistant.

Two cautions. A patent shows what Google has worked on, not what it runs in Search today, and Google has never confirmed an "information gain score" as a ranking factor. But you don't need the patent to be live for the idea to hold up, because Google's public guidance says the same thing in plainer words.

What Google's Own Guidance Says

Google's guide to creating helpful content asks site owners to judge their pages with questions such as:

  • "Does the content provide original information, reporting, research, or analysis?"
  • "Does the content provide substantial value when compared to other pages in search results?"
  • Whether content that draws on other sources avoids simply copying or rewriting them, and instead adds substantial value and originality

For AI features specifically, Google's May 2025 post on succeeding in AI search tells site owners to focus on "unique, non-commodity content." Commodity content is the opposite of information gain: the definition, the five tips and the generic FAQ that every competing page already has.

Why Information Gain Matters More for AI Citations

A classic search results page shows ten links, and a reader can click any of them. An AI answer is different. The engine retrieves several pages, writes one response, and cites a handful of sources to support specific claims. That changes the economics of being "pretty good."

  • Repeated facts need only one source. If five pages give the same definition, the engine can support that sentence by citing any one of them. The other four add nothing.
  • Specific claims need a specific source. A number, a dated change or a test result can only be supported by the page that published it, so that page gets the citation.
  • Research backs this up. The GEO study from Princeton and other researchers, accepted at KDD 2024, found that methods such as adding statistics, quotations and cited sources can boost a page's visibility in generative engine responses by up to 40%. Each of those methods adds checkable information.

This is our reading of how answer engines behave, not a documented rule from any vendor. It is easy to test for yourself: compare the cited and uncited pages for a prompt you care about and look for the fact only the cited page has.

Seven Kinds of Information Gain That Earn Citations

TypeWhat it looks likeWhy AI cites it
Original dataA survey, an analysis of your own customer data, a benchmarkThe number exists nowhere else
First-hand testing"We ran this for 90 days and here is what changed"Experience claims need a named source
Fresh, dated factsA pricing change, a policy update, a rollout dateEngines prefer the most current source for time-sensitive claims
Worked examplesA real calculation, a filled-in template, a before and afterShows how, not just what
Tabulated comparisonsA table nobody else built, such as tools by feature and priceEasy to extract and quote
Expert opinion with reasonsA clear position that disagrees with the consensus, with evidenceGives the answer a second view to present
Edge cases"This advice doesn't apply if…"Answers the follow-up question the engine fans out to

Commodity sentence vs information-gain sentence

Commodity (every page says it)Information gain (only you say it)
"Domain Rating can change over time.""Ahrefs' DR4 model replaces DR3 in its interface on December 1, 2026, and about half as many sites will have a score."
"Long-tail keywords convert better.""In our own Search Console, queries of eight words or more earned about 38,000 impressions and 11 clicks in three months."
"Update your content regularly.""We found 11 posts whose Search Console queries asked questions the posts never answered, and added a direct answer to each."

The second column is harder to write because you have to know something. That is the point.

How to Add Information Gain to a Page

  1. Read what already ranks and what AI already says. Open the top five results and ask the question in ChatGPT and Perplexity. Note the cited sources.
  2. List the facts they all share. These are commodity. Keep a short version for completeness, but they won't win you the citation.
  3. Write down what you know that they don't. Client results, your own numbers, a process you run, a mistake you've watched people make, a date or price that changed recently.
  4. Put the new information near the top of a section, stated plainly in one or two sentences, so it can be quoted on its own.
  5. Source it. Name where the number came from and when. An unsourced statistic looks invented, and engines and readers both discount it.
  6. Cut the padding. Paragraphs that restate the obvious make the useful sentence harder to find.

Mistakes That Look Like Information Gain but Aren't

  • Made-up statistics. A number with no source is a liability, not an asset. If you can't back it up, leave it out.
  • Rewording a competitor. Paraphrasing the top result adds no new information. Google's spam policies call out "stitching or combining content from different web pages without adding value."
  • Length for its own sake. A 4,000-word guide that repeats the same ten facts has less information gain than a 900-word page with one original finding.
  • Opinion with no reasons. A contrarian take only helps if you show why.

Frequently Asked Questions

Why is information gain important for getting cited by AI?

AI answers combine several sources into one response and cite the pages that support specific claims. If your page repeats what the other sources say, the engine can cite any of them instead. A page with original data, testing or specific, dated facts is the only possible source for those claims, so it earns the citation.

Is information gain a Google ranking factor?

Google hasn't confirmed one. The term comes from a Google patent granted in 2022, and patents don't prove what runs in Search. Google's helpful content guidance does ask whether a page provides original information and "substantial value when compared to other pages in search results," which is the same idea.

How do I measure information gain?

There is no public score. A practical test is to list every fact on your page and mark which ones also appear in the top five results. If almost everything is marked, the page has little information gain. Then check whether AI tools cite you for the prompts you care about.

Does AI-written content have information gain?

Not by default. A model writing without new inputs mostly recombines what is already published. AI can help draft and structure, but the new information has to come from you: your data, your tests, your clients and your judgment.

How much original information does a page need?

Enough that the page is the best source for at least one specific claim. One original data point, worked example or tested finding per main section is a good working target for pages you want cited.

Indexed Research

Written by

Indexed Research

Research team, Indexed · Reviewed by Anjan Luthra

The Indexed research team tracks how search and AI answer engines behave, tests what actually moves visibility, and publishes the reference material behind our client work.

Share

One email a week

How to turn search into revenue.

One idea a week from live client work: what we changed, what it was worth, and what you can take from it.

One email a week. Unsubscribe in one click.