What Is Information Gain in SEO? (and How To Add It to Your Content)

Whether Google runs a literal information gain score is still unconfirmed. But whether or not it improves your ranking is almost irrelevant. You should care about information gain because it’s your differentiator. The extra effort you put into creating...

Troov

Aug 20, 2026 - 23:17

0 8

What Is Information Gain in SEO? (and How To Add It to Your Content)

Information gain is the additional information your page contributes beyond what already ranks for the query.

Whether Google runs a literal information gain score is still unconfirmed.

But whether or not it improves your ranking is almost irrelevant.

You should care about information gain because it's your differentiator.

The extra effort you put into creating useful, memorable content can equate to growing your owned audience, building a moat, earning mentions in AI answers, and becoming a name search engines actually recognize.

Below we’ll dig into what information gain actually means, what we know (and don't know) about whether Google uses it, and, most importantly, how to improve your content with information gain.

What is information gain in SEO?

Information gain is a measure of how much new information a page adds beyond what a searcher has already seen.

The term isn't originally an SEO one—it’s borrowed from machine learning.

The patent addresses a specific problem: search results where every page says essentially the same thing.

Take the keyword ”how to improve your credit score”.

There are only so many things that can actually change a credit score.

FICO, one of the early pioneers of credit scoring, laid out five core factors:

Payment historyAmounts owedLength of credit historyNew creditCredit mix

In other words, this is a finite topic. When multiple websites say the same thing, that's just... the answer.

This is where information gain becomes so important. Including the criteria on that list is the bare minimum.

Sites writing about this topic need to offer more.

Unfortunately, they don’t.

Of the seven organic search results ranking for this keyword (minus ads), all share the exact same information…

And, crucially, these sites offer nothing more than that...

Seven different institutions, including credit unions, bank trade associations, and mega-bureaus, converge on the same handful of topics, just with a different number in the title (5 tips, 7 tips, 10 things).

This is a missed opportunity. There's genuine room for differentiation and creativity here.

A bank could cite its own customer data on what actually moved scores, or an author could share a personal before/after of their own credit report.

The topic is crying out for originality, but no one delivers.

Google's proposed fix for this exact situation is a score:

"An information gain score for a given document is indicative of additional information that is included in the given document beyond information contained in other documents that were already presented to the user."

The information game patent reveals Google's intention to enrich the SERP with novel information.

In the words of our Director of Content Marketing, Ryan Law...

"Information gain is Google rewarding content for being different and not just better."

Ryan Law, Director of Content Marketing, Ahrefs

How information gain works—according to Google’s patent

Here's what you need to know about the mechanism Google's patent describes, including what each part means for your content.

1. The information gain score is calculated per-user, per-session

Information gain is context-dependent. It compares a document against the documents an individual user has already been shown.

That means a page visit from two different users can generate two totally different information gain scores.

Someone who's just read five near-identical "how to improve your credit score" articles will find a sixth one redundant.

Someone who typed that query for the first time, with zero context, will find the exact same page genuinely useful.

2. Information gain compares meaning—not wording

The information gain patent has a model that reads each document as "a semantic representation”—it converts what a page means into numbers, so two pages can be measured against each other even when they share no vocabulary.

"The cat is unwell" and "the feline is sick" have almost no words in common, but are recognized by the model as the same information.

A system that understands meaning treats paraphrased content as the same idea.

So, if you've just reworded a competitor's article, you haven't added anything new for it to reward.

Your rewrite will land in the same “semantic spot” as the thing you rewrote.

To create true information gain, you can’t just rewrite the SERP—you need to add new ideas.

This aligns with former founder of Clearscope, Bernard Huang’s interpretation of information gain.

"The concepts and entities on the fringe of Google's knowledge graph for the topic."

The core of any topic is usually saturated, since everyone has already written about it.

According to Huang, gain lives at the edges: the examples, data and perspectives the SERPs haven't picked up yet.

3. Information gain is dynamic

Information gain scores are calculated on-the-go throughout a user's session.

As the user consumes more pages, the remaining, unseen pages get rescored and reordered based on what's already been covered.

Google's patent calls this re-ranking, and it's different from the initial, one-time rankings every searcher sees on the SERP.

In other words, Google is reshuffling the results you'd normally see, based on what you've already read, rather than deciding the initial order (ranking).

Information gain: cause or effect?

One reading of the patent is that each page is scored individually, and those same results are then re-ranked based on their scores. But Bernard Huang sees it differently. He believes information gain feeds back into Google's broader understanding of the topic. In Bernard's view, Google:

Ranks you on authority and entity coverage: Authority refers to factors like your branded backlinks, mentions, and topical consistency, while entity coverage is how often you mention the right people, places, and things.Watches whether real users are satisfied: Based on classic Google proxies like CTR, dwell time, and pogo-sticking.Harvests what your page has said once it's proved itself: Once you've earned trust, Google begins extracting your claims.Feeds that back into its overall understanding of the topic: This feeds into the Knowledge Graph, or whatever sits behind search.

In this scenario, information gain is an output, not an input. Google isn't rewarding you for adding something new, it's absorbing what you added to use elsewhere—e.g. as raw material for future AI Overviews, another data point in an entity's coverage, or as the new baseline for what counts as "consensus" the next time it scores a page on this topic.

4. Information gain removes duplicate facts, not just duplicate pages

Long before AI Overviews existed, Google patented a way to stop repeating itself when it only had one answer to give.

Picture a voice assistant answering "what causes jet lag?"

Ten sources might all say the same three things: your circadian rhythm gets disrupted, light exposure resets it, symptoms fade after a few days.

Read all ten aloud back to back and you'd hear the same three facts nine extra times.

The score was designed to prune repetition, so the assistant picks out each fact once, from whichever source states it best.

The information gain patent explicitly extends to "chatbots," "conversational agents," and "personal voice assistants", which is confirmation that it goes beyond ten blue links.

Here’s how Google proposes it should work:

How a repeated fact scores for information gain

Imagine a document (source 1) contains facts A and B, which an assistant reads to the user. The system then finds another document (source 2), containing fact B plus something new (fact C).

Having determined fact B "has already been conveyed to the user," the assistant generates output that "conveys the third information element and excludes the second information element." Source 2 is used only for fact C. Its overlap (fact B) gets deleted before the user hears anything.

Here's what that could mean for your content.

Unoriginality could get you cut out of assistant answers

In a ranked list, even if you’re the fourth site saying the same thing, you’re still on the SERP, and still get a chance at a click.

But in a synthesized answer there is no position four.

Either you provide the information that hasn't been said yet, or your content gets cut.

Your content needs topical authority AND information gain

Covering what every ranking page already says is the cost of entry into the SERP.

But once you’ve addressed the expected topics, you could earn a place in the answer with a specific insight no one else has covered.

Is information gain officially part of Google’s algorithm?

There is no official confirmation from Google itself that it explicitly uses an "information gain score" in its algorithm or AI search features.

Google filed for, and was granted, a patent describing an information gain score.

But we don't know whether the mechanism in that patent ever shipped.

Google files hundreds of patents that never reach production, and it has never confirmed an information gain score as a live ranking or re-ranking system.

Why SEOs suspect that Google rewards information gain

Some SEOs still believe information gain is baked into Google’s system because:

Google’s best practice guidelines encourage information gainGoogle algorithm updates tend to reward original contentGoogle can already change what users see mid-session

1. Google’s best practice guidelines encourage information gain

Google's own helpful content guidance explicitly asks creators to assess whether content is original and provides substantial value when compared to other pages in the search results.

Google also reinforces “adding value” in their spam policy, confirming it's not just a ranking nicety.

And in its Search Quality Rater Guidelines—the actual document that Google gives human raters to review content, and arguably the most authoritative source on what Google wants its search results to look like—it repeatedly states that the main content (MC) needs to add value compared to similar pages already on the web.

What's more, in March 2024 Google said it had reduced "low-quality, unoriginal content in search results by 45%."

Originality is unambiguously something Google says it measures and acts on.

With such emphasis on original content, some SEOs have reasonably inferred that Google must have a way to measure additive content in its algorithm.¹

2. Google algorithm updates tend to reward original content

SEOs like Lily Ray, Marie Haynes, and Cyrus Shepard have made it their job to study the impact of algorithm shifts.

Individually, they have all noticed that sites benefiting from Google’s updates tend to over-index on original content ¹ ² ³ ⁴

For instance, in Cyrus Shepard's analysis of 400+ sites following the December 2025 core update, he found that “proprietary assets" like original data and tools were the third strongest predictor of which sites gained traffic.

This suggests originality is something Google's ranking systems are actively rewarding.

3. Google can already change what users see mid-session

Google's own VP of Search, Pandu Nayak, confirmed under oath in the 2023 DOJ antitrust trial that NavBoost, a system that re-ranks pages based on rolling click data, is one of its most important ranking signals.

Several other Google patents also describe re-ranking results within a single session, based on what a user has already clicked.

So Google already has the technical ability to change what a user sees mid-session, which is exactly how the information gain patent is described as working.

Why information gain matters more now than ever

Google's information gain patent was filed in 2018 and granted in 2022.

Shortly after, ChatGPT launched and commodity content exploded.

An "information gain" fringe trend has grown right alongside it, as people search for ways to stand out from AI-generated content.

Now that anyone can generate a "comprehensive guide" in minutes, comprehensiveness has stopped being a differentiator.

LLMs predict the most probable next token.

Which leads to the most probable content.

And it’s exactly that kind of “consensus content” that gets absorbed into AI answers without credit.

Information gain is, by definition, the part that isn't consensus yet.

It’s less susceptible to click-loss, but it also makes your brand memorable—and the brand an audience remembers is the brand they come back to; first for content and then for products or services.

So information gain is more than just a case of ranking or being mentioned in AI answers.

It's differentiation, brand building and—at risk of sounding a little dramatic—survival.

How to add information gain to SEO content

To optimize for information gain in SEO, you need to study what the top-ranking pages don't say, and fill those gaps with original data, first-hand experience, or a distinct angle they're missing.

This will help you better engage your readers and improve their overall experience of your brand.

But remember, the foundation of good content is understanding and serving the primary intent.

Information gain only works once that's in place.

Study search intent to find underserved topics angles

Search intent is about figuring out what information searchers are looking for.

Start by pulling up the SERP for your target keyword and reading every result as if you were the searcher, not the writer.

Most pages will answer the obvious version of the question, but you need to look for the one they've all skipped.

To save time, you can also get a quick read on overall search intent using the Identify Intents report in Ahrefs SERP Overview.

Our AI clusters ranking pages by what searchers are actually after, and shows roughly what share of traffic each intent captures.

Most SERPs serve several intents at once, unevenly.

If one intent already absorbs 80% of the ranking content, it represents the consensus.

That doesn’t mean you shouldn’t cover it—you absolutely should.

But the information gain opportunity is the underserved intent group, or potentially even in a group that doesn’t exist on the SERP yet (I'll give you tips on how to find these topics later on).

Run a full content gap analysis against top-ranking competitors

You can't add information the SERP is missing until you know what the SERP already says.

A content gap analysis gives you a map of everything the top-ranking pages cover, so you can see exactly where the unclaimed ground is.

Ahrefs AI Content Helper scores your draft's topic coverage in real time against the pages currently ranking for your keyword.

Enter a keyword, and it extracts the topics the top pages cover, then scores your draft—overall, and topic by topic—so you can spot gaps and enrich thin content.

The “Competitor” content pane shows you the full content body of each competing page alongside its Domain Rating, word count, and referring domains, so you can read exactly what your rivals have written about specific topic, and decide what to add.

To optimize for information gain, start with the lowest-scoring topics, open the competitor content for each, and ask "what can I say here that none of these pages does?"

If you get stuck, the AI chat can help. Paste in a weak section and ask what angles the ranking pages are missing.

Remember, adding content that matches what competitors say raises your Content Score—meaning you're closing a coverage gap.

But adding content no competitor has creates information gain.

So try to get your score up as high as possible to build the kind of authority and entity coverage that gets you ranked, before adding in genuinely novel information that drives real user satisfaction and engagement.

Fill gaps with unanswered questions

Questions that keep getting searched, keep surfacing in People Also Ask boxes, and keep turning up in forums—but go unanswered on page one—are a strong signal of underserved intent and a real opportunity to add information gain.

Here are a couple of ways to find them.

Spot trending questions using Ahrefs “Questions” report

Ahrefs Keywords Explorer shows you question keywords (what, how, why, does, can, should) related to the core keywords you enter in the search bar.

Just drop in your target keyword, open Matching terms, and toggle to Questions.

Then sort the queries by 3/6/12 month growth to find questions that may not yet have a definitive answer…

You can even filter to see questions that page one competitors have failed to answer using the following filter:

“Target > Show keywords target doesn’t rank for in top 10”

For instance, Digital Marketing Institute ranks in position two for the head term “SEO”, but aren’t yet covering the trending topics of “weekly SEO metrics” or “AI SEO” well enough to hold page one positions.

Obviously, this is all a balancing act.

You need to add enough novel information to stand out from the current SERP, but not stray so far that you lose the core search intent.

My advice would be: watch your rankings, and treat information gain optimization as an ongoing test.

Find and intercept questions being answered on Reddit

When Reddit is ranking prominently on page 1, it's often a signal that other sites in the SERP aren't fully satisfying intent.

Reddit is where searchers will say, in their own words, what the existing SERP content didn't tell them.

Here's an easy way to find and intercept relevant questions that are currently being fielded by Reddit.

Head to Keywords Explorer and either add in a stem keyword or…Add a category filter relevant to the topic you want to write aboutHit the “Question” present to filter for question keywordsDrop reddit.com into the Target filter, and select “Show keywords target ranks for in top 10”Lastly, sort by 3/6/12 month growth to find untapped questions that Reddit is currently answering

Include unique data

Of everything you could add to a page, original data is one of the most defensible forms of information gain.

In fact, there's early evidence that it's the strongest single correlate of originality.

On-Page.ai's study of 150 top-3 ranking pages found that unique data points correlated even more strongly than ranking position itself.

While originality barely separated the top 3 positions, averaging an information gain score* of 50–55 regardless of rank...

Bucket

Mean

Median

Mostly shared

Positions 1–3	51.4	52	24%
Position 4	46.5	47	40%
Position 7	49.3	50.5	37%
Position 10	44.5	48	38%

...pages with 15 or more unique data points averaged a score of 62, compared to just 40 for pages with just one or none, making original data the strongest single predictor in the study.

* On-page.ai defines their information gain score as "A 0–100 measure of how much a page adds beyond the live ranking cohort for its keyword, compared by meaning rather than exact wording."

As On-page.ai’s Eric Lancheres acknowledges, these results don't justify statistic-stuffing.

Padding a page with junk numbers just adds figures, not actual information.

Trying to game the score like that won't fly, since it's measured on meaning.

If you want to produce your own unique stats, you don’t have to run a million-datapoint investigation.

A LinkedIn poll of 200 SEOs, a survey of your own customers, or even an analysis of your Ahrefs data can all produce figures that don't exist anywhere else.

Take our customers at Visibility Labs. They used AI Overview data they already had access to in Ahrefs Keywords Explorer, and created a super interesting study analyzing 20.9 million shopping SERPs.

Keep your content fresh

Information gain has a time dimension: what counts as "new" is always relative to what's already been said, and the longer a fact has been in circulation, the more thoroughly it's been absorbed into the consensus.

Fresh information is, almost by definition, information the rest of the SERP doesn't have yet.

We've seen the impact of refreshing content firsthand.

Our “What are SERPs” blog, published in 2020, lost almost all of its traffic by mid 2025, but after a rewrite and republish in December 2026, traffic recovered by 50x.

Google runs a ranking system called "query deserves freshness", which prioritizes recently published or updated content—especially for time-sensitive queries.

Updates also feed into user signals: an outdated title earns fewer clicks, and click behavior influences how Google re-ranks content.

To find refresh candidates with the best effort-to-reward ratio in Ahrefs:

Head to the Top Pages report in Ahrefs Site Explorer.Drop in your site—either the domain or a specific subfolder. Below, I’ve gone with the Ahrefs Blog.Set up a traffic filter, and set that to “declining”. You want to choose a monthly traffic number that usually reflects your best, most visible pages.Set a low KD filter (e.g. up to 40). Content isn’t always the reason your posts stop ranking. Drops also occur when other pages have built better “link authority”. Setting a low Keyword Difficulty score will show you update opportunities in a less competitive SERP.Select the date range you want to monitor the decline over. For me, that’s one year.Sort by negative traffic change. So you’re basically zeroing in on the top pages on your site that have dropped off lately.Check Content Changes. This shows you which pages have already been updated over your date range so you can focus on outdated content only.

Then it’s just a case of covering the information a user expects to read, finding out what the SERP hasn't said yet, and becoming the page that says it.