Information Gain in SEO is the score for how much new knowledge your page adds

Information Gain in SEO is Google’s way of measuring how much genuinely new information a page contributes on top of everything that already ranks for a query. Not how complete it is. Not how long. How much it adds that the other results do not already say. A page that restates the consensus scores near zero, no matter how polished it is. A page with a number, a method, or a take nobody else has scores high, and that is increasingly the page that wins.

It has quietly become one of the most important ideas in ranking, and almost nobody outside of technical SEO is building for it.

A page that only tells the model what it already knows adds nothing. Information Gain is Google putting a score on whether you actually said something new.

Why Information Gain matters more now: the machines already know the basics

Here is the mechanism, and once it clicks the whole thing makes sense. Google’s systems and the large language models behind AI search are trained on the web. They already contain the commodity version of almost every answer: the definitions, the “what is X” intros, the same ten-tool listicle, the same three stats everyone cites. When your page repeats that, you are handing the model something it already has. You add no signal, so you earn no reward.

What these systems actually crave is information they do not already hold. New data. A first-hand result. A contrarian but defensible argument. That is the only input that changes what the model can say, which is why it is the only input that earns a citation or a ranking bump. Information Gain and first-party data are not SEO jargon dreamed up to sell content. They are the direct consequence of how these systems learn. Give them what they lack, and you become worth quoting. Give them a rerun, and you are wallpaper.

This is also why the Great Decoupling and AI Overviews hit commodity content the hardest. The model can answer the basic query itself, right on the results page, with no click for you. The only content that survives that is the content it cannot generate from what it already knows.

Why AI has to reward Information Gain, or it eats itself

There is a deeper reason this is not a passing trend. LLMs are increasingly trained on content that was itself written by LLMs, and that is a problem for them, not just for you. A model trained on its own output is a copy of a copy of a copy. Go far enough down that chain and something breaks: by the fifth generation of restated, synthesized, no-new-information content, the quality degrades and meaning gets lost. If the whole internet were AI rehashing AI, these systems would get worse fast, because there would be no new signal left anywhere to learn from.

So the models, and the search engines built on them, have a survival interest in finding and rewarding genuinely original, first-hand content. It is the only thing that keeps the well from going stale. That is why Information Gain is not a temporary algorithm quirk. It is structural. For an LLM or an AI Overview to stay useful at all, it has to prioritize content that actually adds something.

Which lands on the least clever, most durable advice in SEO: make content that genuinely satisfies the person who searched. Answer the query so well they stop looking. The most reliable way to do that is to hand them a real insight they cannot get anywhere else, and that, not by accident, is exactly what scores high on Information Gain. Satisfy the user and you satisfy the machine, because the machine is trying to satisfy the user too.

The Information Gain patent, and the core updates that woke it up

The idea is not a theory someone floated on a podcast. Google filed the Information Gain patent in 2018, and it was granted in June 2024. It describes scoring documents by the additional information they provide relative to what a user has already consumed, so that a search session surfaces new material instead of ten pages saying the same thing in a slightly different order.

For years it sat mostly quiet as a ranking influence, and here is my read on why. Until recently, Google had no cheap way to actually enforce it. Think about what Information Gain requires: reading dozens of existing articles on a topic, understanding them, and judging whether a new page adds anything they do not already contain. Ten years ago that was not remotely feasible at scale, in real time. AI is what changed it. The same leap that made AI Overviews possible also gave Google the ability to dissect a whole topic’s worth of content and score a new page against it in an instant. The patent could finally be enforced, and the 2026 core updates are what that enforcement looks like.

The March 2026 update alone reshuffled roughly 80% of top-three results and knocked about one in four top-ten pages out of the top 100, and the pattern underneath the volatility was consistent: thin, restated, coverage-for-its-own-sake content lost, while pages with a genuine point of view and original data held or climbed. On an observability SaaS account I gave an early read on that update, the split was exactly this, and it is the same story I have watched repeat since. So what: if a site got hammered and nobody can explain why “good, comprehensive” content dropped, this is usually the answer. Comprehensive is table stakes now. New is the differentiator.

What high Information Gain looks like, in B2B SaaS terms

Picture two pages targeting “best project management software.” The first lists ten tools with feature tables pulled from each vendor’s own site and a paragraph of description you could copy off any competitor. It is complete. It is also pure restatement, and its Information Gain is close to zero, because every fact on it already lives on fifty other pages the model has read.

The second names the same tools but adds what no one else has: which one a mid-market firm actually chose after a migration, the intake benchmark that made the case, the one integration that broke and how. That page has Information Gain, because it carries knowledge that exists nowhere else, and it is the one a model pulls from and a buyer trusts. This is why the ten-item listicle stopped compounding. On a legal SaaS account I run, three of its number-one BOFU listicles started slipping as the format lost its edge, exactly the are-listicles-dead pattern. Length and keyword coverage did not save them; the absence of anything proprietary did them in.

Here is the heuristic I actually use to judge it, and it is brutal but fast. If a sentence could appear unchanged on a competitor’s page, it does not belong on yours. Read a draft and mark every line that fails that test. What is left, the parts only you could have written, is your Information Gain. If almost nothing is left, you have a formatting exercise, not a page worth ranking.

There is a blunter version of the same test: if your content could be produced by anyone with a computer and an LLM, it is not worth posting. That is the same idea from the other direction, and it is exactly what Information Gain measures. An LLM can only recombine what already exists, so the moment your page is nothing but a recombination, its Information Gain is zero by definition.

How to create Information Gain: first-party data and a POV you will defend

You cannot fake new knowledge, so you have to generate it. Four sources, in order of durability:

  1. First-party data from your own operation. Customer outcomes, internal benchmarks, before-and-after numbers, a metric only you can see. One real figure (“firms that switched cut intake time 40%”) outweighs a page of borrowed stats, and it is the thing others start citing.
  2. Original research on a cadence. Run a survey of your market, publish the findings, and repeat it quarterly or annually so it compounds. Slice one dataset by industry and company size and you get many citable cuts from a single effort. Proprietary data others cannot replicate wins twice: it lifts rankings and makes you the source AI names.
  3. First-hand test results. Change one thing, measure it, and report what happened. A single documented experiment (“we shipped schema and nothing else, here is the lift”) is information the model has never seen, because you just created it.
  4. A defensible point of view. Not a hot take for its own sake. A specific, argued position you will stand behind, drawn from doing the work. Models quote a distinct argument and discard the mush.

Commodity stat dumps, the Gallup-SHRM-McKinsey triple that opens every competing article, do the opposite of Information Gain. They are the definition of information the model already has. Use them as seasoning if you must, never as the backbone.

How to tell if your content has any Information Gain

You can audit this without a tool. Pull up the current top five results for your target query, read them, and list every subtopic they all cover. That shared list is table stakes: match it, because a page that misses the basics does not get to compete on anything else. Then ask the only question that decides the outcome: what does my page say that none of those five do? If the answer is a first-party number, a named example, a method, or a genuine argument, you have Information Gain. If the answer is “mine is written better,” you do not, and better writing is not a moat.

There is also a symptom worth watching. When strong-looking pages come back crawled but not indexed, Google is often telling you it evaluated the page and decided it added nothing it did not already have. That is an Information Gain verdict in disguise. The fix is rarely more words. It is one real thing to say that no one else is saying, which is the same discipline behind E-E-A-T and sharp keyword research: cover the topic, then earn the ranking on what only you can add. For the full playbook on turning that into citations, the ranking factors that actually get you cited start from the same place. It all ladders up to the SEO fundamentals that hold in the AI era.

FAQ

What is Information Gain in SEO?

Information Gain in SEO is a measure, drawn from a Google patent, of how much new information a page adds relative to what already ranks for a query. It rewards pages that contribute something the other results do not already say, and gives near-zero credit to pages that only restate the consensus. In practice it is why original data and a real point of view now outrank longer, more polished versions of the same commodity content.

Is Information Gain a Google ranking factor?

It is best understood as a ranking concept Google has patented and appears to weight, not a single dial you can see in a tool. The 2018 patent (granted 2024) describes ranking documents by the additional information they provide, and the pattern across recent core updates, restated content losing while original content holds, is consistent with it being weighted more heavily. Treat it as a real influence to build for, not a metric to game.

What is the Google Information Gain patent?

It is a patent Google filed in 2018 and was granted in June 2024 that describes scoring documents by how much new information they add to what a searcher has already seen in a session. The goal is to avoid serving a page of near-identical results and instead surface material that advances the searcher’s understanding. It is the technical backbone behind the idea that “new beats comprehensive.”

How do you increase Information Gain on a page?

Add something the competing pages do not have: first-party data (customer outcomes, internal benchmarks), original research, a documented first-hand test, or a defensible argument you will stand behind. Then cut the lines that could appear unchanged on a competitor’s page. The goal is not a longer article; it is a page that carries knowledge available nowhere else, because that is the only part Google and AI cannot already generate themselves.

What is an example of high Information Gain content?

A page that includes something the competing results do not: original survey data, a customer’s real before-and-after numbers, a documented first-hand test, or a specific argued position. A “best project management software” page that adds which tool a mid-market team actually chose, and the benchmark data behind the decision, has Information Gain. A page that only rebuilds everyone else’s feature table does not, no matter how well it is written.

Does Information Gain replace keywords and E-E-A-T?

No, it sits on top of them. You still need the right target keyword, clean structure, and the experience and trust signals behind E-E-A-T. Information Gain is what breaks the tie once several pages have all of that: among equally optimized, equally credible results, the one that adds new information wins. Treat it as the differentiator, not a replacement for the fundamentals.