What Is Information Gain? Why "Comprehensive" Content Stopped Working
Key takeaways
|
For about a decade, content marketing ran on a formula most of us could recite in our sleep. Search the keyword. Open the top ten results. Write something longer, with more headings, a better graphic and a few extra sections the others missed. Call it the ultimate guide and wait for it to climb.
It worked, for a while. Then everyone did it. The first page of Google filled up with ten versions of the same article, each one a little longer than the last and none of them saying anything the others didn't. We called it comprehensive. Readers called it skimmable. Search engines, it turns out, called it redundant.
Now there's a new reader in the room, and it has already read all ten. Ask an AI model the question and it will happily summarize the consensus for free, in a paragraph, without anyone visiting a single one of those pages. So the only content worth publishing is the part the consensus is missing. That part has a name. It's information gain, and I think it's the most important idea in content strategy right now.
What is information gain?
Information gain is the amount of new, useful information a piece of content adds beyond what a reader, search engine or AI model has already seen on the topic. Content with high information gain offers original data, first-hand experience, new frameworks or distinct perspectives instead of restating what existing sources already say. |
Where the idea comes from
The term comes from information theory, but in search it's tied to a Google patent titled "Contextual estimation of link information gain," granted in 2022. The patent describes a way of scoring documents by how much additional information they would give a user beyond what that user has already seen. In plain terms, if you've read three articles on a topic, the fourth one is only valuable to the extent it tells you something the first three didn't.
I want to be careful here, because a lot of SEO writing treats patents as proof. A patent tells you what a company thought was worth protecting, not what's running in its systems today. But the direction is consistent with what Google says in public. Its guidance on helpful content asks whether a page provides original information, reporting, research or analysis, and whether it provides substantial value compared with other pages in search results. That's information gain, described in Google's own words, without the patent language.
Why AI made it non-negotiable
Large language models are, in a sense, consensus machines. They've absorbed the average of what the web says about almost every topic you can name. If your article restates that average, a model has no particular reason to retrieve it, quote it or remember where it came from. It already knows what you're saying.
The citation data backs this up in a way that should make every content team sit up. Kevin Indig's analysis of around 1.2 million ChatGPT responses, covered by Search Engine Land, found that roughly 30 domains captured 67% of citations within a given topic, and that ChatGPT retrieved about six times more pages than it ended up citing. Citations aren't spread evenly across everyone who wrote something decent. A small number of sources win, and they tend to win because they offer something the others don't.
That's the real shift. In classic SEO, you could rank tenth with a me-too article and still pick up a trickle of traffic. In AI answers, there is no tenth place. You're one of the handful of sources the answer is built from, or you aren't in it at all.
The skyscraper hangover
I have real sympathy for teams caught in this, because the skyscraper method was genuinely good advice once. Find what ranks and build it taller. A lot of us, me included, built content programs on that idea, and they produced results.
But comprehensiveness stopped being a differentiator the moment AI could produce a comprehensive summary of any topic in seconds. Length without new information is now just a longer version of what the model already knows. And there's a cost to length that we rarely talk about. Every extra section of restated consensus pushes your genuinely original idea further down the page, where, as the citation research shows, it's less likely to be found and quoted at all.
So the question I now ask of every draft isn't "did we cover everything?" It's "what's in here that isn't anywhere else?" If the honest answer is nothing, the draft isn't finished, no matter how long it is.
Where information gain actually comes from
Not all originality is equally durable, and it helps to think about sources of information gain in order of how hard they are for a competitor, or a model, to copy.
The easiest is perspective, a clear, defensible opinion about what you'd do and why. Anyone can imitate a perspective, but it's hard to own one unless you hold it consistently and back it up. A step harder is synthesis, connecting ideas or data that nobody has connected before. Then comes the named framework, a model that organizes a messy problem into something people can use and remember, which has the useful side effect of attaching itself to your name. Harder still is experience, what actually happened when you did the thing, including the parts that didn't work. Then original data, meaning your own research, surveys or numbers from your own work. And hardest of all is access, meaning interviews, insider detail and sources nobody else can get.
Most strong pieces combine at least two. Perspective on its own is cheap. Data on its own can be dry. But a named framework built on real experience and backed by a few honest numbers is close to unbeatable, and it's exactly the kind of thing AI engines learn to associate with a specific brand. That's the core idea behind LLM SEO.
Finding the gap before you write
The process I use is less glamorous than it sounds. Before drafting anything, I collect the consensus. I read the top few results and ask ChatGPT and Perplexity the target question, then write down every point they all make. That list is table stakes. Readers need it, so I'll cover it, but briefly, because it isn't my contribution.
Then I ask what's missing. What would a practitioner add that a summarizer wouldn't? What's out of date? What does everyone repeat that I think is wrong, or at least incomplete? Those answers are where the piece actually lives. And I make sure that material shows up early. Indig's research found 44.2% of ChatGPT citations came from the first 30% of a page, so burying your best idea in the ninth section is a quiet way of hiding it. What is an AI snippet goes deeper on placement.
The last check is the one I'd recommend to any team. For every section, ask so what, and then what. So what does this mean for the reader, and then what should they do about it? Generic content answers neither question. Content with real information gain answers both without being asked.
You don't need a research department
One of the most common objections I hear is that original research is expensive, and only big companies can afford it. That's half true. Large studies are expensive. But some of the best sources of information gain are already sitting inside your company, unused.
Your sales calls are full of objections and questions that competitors never write about, because they never hear them. Your support tickets describe real problems in real customer language, along with the fixes that actually worked. Your own data, pipeline by channel, time to close, what changed after a launch, is information nobody else has, and it's more credible when it's honest than when it's impressive. Even a small survey of fifty well-chosen people, reported transparently with its sample size, is more useful than a borrowed statistic everyone has already quoted. Content gap analysis and competitive content analysis help you see where those openings are.
Where it goes wrong
Information gain can be faked, and the fakes tend to backfire. Novelty for its own sake is the first trap. Being different isn't the same as being useful, and a contrarian take nobody can act on is just noise. Manufactured data is the second and more serious one. A fake or inflated statistic is worse than none, because once it's found out it poisons everything else you've published, and it runs directly against the trust principles in E-E-A-T.
The quieter failure is gain without credibility, a bold claim with no author, no source and no explanation. Original ideas need more support, not less, precisely because nobody else is saying them. And the most common failure of all is simply skipping the basics in a rush to be original. Readers still need the definition. Give it to them quickly and clearly, then add what's new.
Information gain in the Engine Optimization Matrix™
Information gain sits at the heart of the Content column in the LLMO row of the Engine Optimization Matrix™, which calls for distinctive point-of-view content with citable phrasing and named IP. It's also what makes the GEO row work, because generative engines need a reason to choose your passage over the consensus.
You can't measure it directly, but you can see its effects in other sites citing the piece, AI answers repeating your framework or data, journalists asking about it and your brand mentions growing around the topic. For related reading, see what is generative engine optimization, vector embeddings and how to build authority through thought leadership.
Buyers are asking ChatGPT for a shortlist. The Demand Doctrine™ for GEO is how you get on it, with structured answers, the right off-site presence and the original ideas generative engines choose over the consensus. Out November 18. Want all four engines? Mastering AI Visibility.
Frequently asked questions
What is information gain in SEO?
Information gain is the new, useful information a page adds beyond what other pages on the topic already provide, such as original data, experience, frameworks or perspective.
Is information gain a Google ranking factor?
Google holds a patent on estimating information gain, granted in 2022, but a patent doesn't confirm current use. Google's helpful-content guidance does ask whether pages add original value compared with other results.
Why does information gain matter for AI search?
AI models already know the consensus, so they favor sources that add something new. ChatGPT citations are concentrated among a small number of domains within each topic.
How do I add information gain to my content?
Add perspective, synthesis, a named framework, first-hand experience, original data or access to sources others don't have, and put it near the top of the page.
Is longer content better for information gain?
Not by itself. Length without new information is just a longer version of what already exists.
Can small teams create information gain?
Yes. Sales calls, support tickets, your own data, small honest surveys and defensible opinions are all strong sources.
How do I find information gaps?
Collect what the top results and AI answers already say, treat that as table stakes and look for what's missing, outdated or wrong.
How do I know if my content has information gain?
Look for its effects, such as citations, links, AI answers quoting your ideas, media interest and growing brand mentions around the topic.
About the author. Harold Bell is the founder of MQL Magnet and has spent about 16 years in enterprise technology marketing, working with brands like Cisco, Rubrik and Nutanix in different capacities. He's a member of the Forbes Communications Council, the creator of the Engine Optimization Matrix™ and the author of Partner Over Product. |
