Hey there!

Here's a test you can run on your own blog in about ten minutes.

Take your best-performing article, open the pages currently getting cited for that topic in ChatGPT or AI Mode, read them, and then ask what's in your piece that isn't in any of theirs.

Most of the time, the honest answer is actually nothing. The writing might be better, the structure might be cleaner, but the information is the same, just reorganized.

That's the gap information gain describes, and that’s something that matters in AI search.

The term gets thrown around a little too much, though, so I'm going to be specific about what Google has actually said, what a patent everyone quotes talks about, and where the honest limits of both sit.

Then, I will share four things that create information gain, because the concept does nothing for you on its own.

Let's go. 🚀

Original research topped the content-quality factors in a survey of 131 SEO professionals. Cyrus and Dawn Shepard at Zyppy Signal asked 131 practitioners to score more than 100 possible ranking factors on a seven-point scale, generating around 13,665 data points. Search Intent Match rated highest overall, and original research with first-party data topped the content-quality factors. Scaled AI content with minimal value scored badly, though heavily reviewed AI content paired with unique site material didn't. It's a survey of expert belief rather than a measurement of Google, so read it as where the industry's head is at. [Source: Search Engine Land]

Google is testing exact and phrase match Search campaigns inside AI Mode. The test covers those match types on queries with clear, direct intent. Ads in AI Mode aren't new, since Google announced testing back in May 2025, so this is a widening of eligibility rather than a first move. Still worth noting if your AI Mode plan assumed the surface would stay organic, because the direction hasn't changed. [Source: Search Engine Land]

A recovered Google Maps binary surfaced 72 signal entries (but with a warning attached). Researchers pulled 72 Oyster Rank signals out of Geostore, Google's internal system for geographic entities. Twenty-five are explicitly marked deprecated, the recovered names come without weights, and the author is blunt that "turning the 72 Oyster Rank signals into a checklist of 72 Google Maps ranking factors would miss most of the architecture." Useful read, and a good example of a finding that might get misreported within a week. [Source: Search Engine Land]

Let’s start with what Google has actually published. Its guidance on creating helpful content asks you to answer these honestly:

Google doesn't use the phrase "information gain" anywhere in that document, and I'm not going to pretend it does. What it does is ask the same question in plain language, on a public page, with no interpretation needed. That's enough to act on.

Then there's the patent, which is where people go wrong in both directions. Google holds a patent called Contextual estimation of link information gain (US11354342B2), filed in October 2018 and granted in June 2022. It describes scoring a document by how much new information it holds compared to documents the person has already read. A score near zero suggests little additional information.

I've seen it described as applying only to voice assistants. That's an overcorrection. The patent covers both, and it's explicit about search: references "may be reranked and/or one or more documents may be excluded" from search results based on updated information gain scores after someone has viewed a document.

What it doesn't do is establish a general originality ranking signal, and Google has never said it's running in Search. So ignore anyone calling information gain "Google's number one ranking signal in 2026." A patent shows you what a company has thought about, not what it has actually shipped.

The part I find interesting is the shape of the mechanism. Something reads documents in sequence, tracks what's already been covered, and prefers the next one that adds something. That's a reasonable analogy for what an answer engine faces when it builds a response from several sources at once.

So, let’s look at four ways to create information gain.

You can't add to a conversation you haven't listened to.

Run your target question through ChatGPT, Perplexity, and AI Mode. Write down every source each one cites, then open them. In my experience, a handful of domains recur across the engines, and they often aren't the pages ranking on Google's first page for the same query.

Then build two lists. What do nearly all of them say? That's the consensus. Covering it again is fine if your reader needs it, but it won't be the reason anyone cites you. What does none of them say? That's your brief.

I'd run this before committing to anything substantial. Occasionally, it tells you the ground is already well covered, and you have nothing to add, which is a cheap thing to learn before you write four thousand words rather than after.

Models can generate explanations, but they can't generate your data.

Nothing needs to cite you to explain what a Bitcoin wallet is. Your own numbers are different, because you're the only available source for them. That's not a guarantee of a citation, but it's the strongest reason to give one.

There's a wide spread in the numbers here. Adam Gnuse looked at 10 sites and around 150,000 pages and reported trends and analysis posts at a 78% citation rate, data-based year-in-review posts at 61%, and educational how-to content at 12%.

But be careful with those figures because the methodology measures GA4 referral sessions over a single month in March 2026, which counts referred visits rather than citations that nobody clicked, and the denominator isn't clearly defined. It's one practitioner's analysis of ten sites, not a controlled study. I'd treat 78 against 12 as a direction worth acting on rather than a number to put in a deck to show your CEO.

You already have data. Anonymized client results, your own pricing outcomes, a survey of your list, the questions your sales team answers every week. Most companies sit on all of it and publish a glossary page instead.

The fastest route to a genuinely new claim is checking an old one.

Every field has advice that gets repeated because it got repeated. Someone said it, it sounded right, and now it's in forty articles with no source behind any of them. Pick one, check it properly, and publish what you find, whether it confirms the advice or kills it.

Ahrefs did this with structured data. They identified 1,885 pages that added JSON-LD between August 2025 and March 2026 and compared each against three matched control pages that never did. ChatGPT moved 2.2% and AI Mode 2.4%, neither statistically distinguishable from zero. AI Overviews showed a 4.6% decline that was statistically significant but small in absolute terms, around 12 daily citations per page.

Worth being precise about what that is, since precision is the whole point of this section. It's a matched observational study. The authors flag that pages adding JSON-LD often change other things at the same time, and that it can't tell you anything about pages not already visible to AI systems. What it does show is no meaningful citation uplift for pages that already had visibility.

You don't need their sample size. Testing something across your own twenty pages and publishing the honest result still puts you ahead of everyone theorizing.

Read your category's top pages and notice where they go quiet. That could include things like pricing, who the product is wrong for, what it costs when it goes badly, etc.

Vagueness in a category is usually legal caution or sales nerves, and it leaves a gap. The page that names the number or states the limitation is answering the question the buyer actually asked, which is a better reason to be quoted than anything you can add to your markup.

This is also the hardest one to get signed off internally, so pick your fight. One specific admission per piece is plenty.

Use this week’s prompt to find the information gap before you write, rather than judging the piece afterward.

The Scenario: You're about to write on a topic that's already well covered, and you want to know what's genuinely missing before you spend a week on it.

The Prompt:

You're helping me find an information gap. Search the web before answering anything. Only use pages you can open right now, and give me the link for every claim. Don't tell me about your training data or how you rank things. If you can't check something, say so.

Topic: [topic] The exact question my reader would ask: [target question] Who they are and what they're deciding: [reader and decision]

Work through these in order.

Be clear that this is based on the sources you opened today, not a prediction about what will get cited.

Run it on the same topic across ChatGPT, Perplexity and AI Mode separately. The sources won't match, and the questions none of them answer are the ones worth your week.

On measuring this. A tracker like LLMRefs will tell you how often you're appearing across a sampled set of prompts, which is worth having. What no tool can do is prove originality caused the change. If you want a rough read, watch whether pages where you published something original pull ahead of pages where you explained something, checked monthly on a fixed prompt set in a clean browser. And if they don't pull ahead after two quarters, the honest response is to reassess everything: the data, the topics, whether the pages are discoverable at all, and whether the idea holds.

These are the most popular AI tools we use at Rise Up Media. If you're not using them already, they're worth a look.

Full disclosure: some links above are affiliate links. If you sign up, we'll earn a small commission at no extra cost to you.

The uncomfortable part of information gain is that it isn't a technique. There's no markup for it and no tool that produces it. It comes down to whether you've done something or learned something the other pages haven't.

That's bad news if you wanted a checklist. It's good news if you run a real business, because you're sitting on results, numbers, and customer conversations nobody else can publish.

Read the pages already getting cited, find what they all skip, and then go and be the only source that answers it.

Until next time, keep exploring the horizon. 🌅

Alex Lielacher

P.S. If you want your brand to show up in Google AI Mode, ChatGPT, and Perplexity, reach out to my agency, Rise Up Media. That's what we do!