Take five agencies publishing on the same HVAC maintenance topic this month. Strip the logos off all five and hand them to someone unfamiliar with any of the brands. Most people could not tell you which page belonged to which agency. That is not a writing problem. It is what information gain SEO is built to catch, and Google has a patent that explains exactly why it costs you citations.
Google’s Contextual Estimation of Link Information Gain patent (US11354342B2, granted June 2022, with continuations US11720613B2 and US12013887B2 extending it through 2024) describes a system that scores a document by how much new information it adds beyond what a user has already seen on the same topic. A document that repeats what is already covered scores near zero. A document that introduces facts, angles, or data the existing set does not have scores high, independent of how many links point at it.
Most agencies already sense this. Almost none have named it, and fewer still have built a workflow step around it.
Key Takeaways
- Information gain SEO measures redundancy, not just relevance. A well-optimized page that repeats what four competitors already say scores near zero, regardless of link count.
- The failure is often invisible in standard rank tracking. A page can hold its position and still lose every citation comparison, since Search Console shows relevance, not redundancy.
- The fix does not come from the fulfillment pipeline. It comes from one deliberate layer of client-specific knowledge added before publication, usually supplied by the account team.
- Snap SEO, an Edmonton agency, runs this check on every page, ensuring there’s sufficient information gain before marking it finished.
What the Patent Actually Describes
The patent frames information gain around a simple scenario. A user asks a question, is shown a first document, and views it. The system then scores a second set of unseen documents based on how much of their content overlaps with what the user already read. A document that restates the first one scores near zero gain. A document covering a genuinely different angle, one the first document did not touch, scores high.
The mechanism matters less than the implication. Ranking and citation are not just measuring whether a page is relevant to a topic. They are measuring whether it is redundant with what is already there.
Here is the part most agencies get backwards. Comprehensive and original are not the same axis, and treating them as the same thing is why so much agency content underperforms despite hitting every on-page checklist. A page that thoroughly covers a topic exactly the way four competitors already have is not comprehensive in the way that earns citation. Google’s system does not see thoroughness. It sees four documents saying the same thing, then a fifth one saying it again.
What Happens to the Page That Loses This Comparison
This is not an abstract scoring exercise. It shows up as a page that ranks fine on a traditional SERP crawl, shows reasonable traffic in Search Console, and still never appears when a client asks ChatGPT or an AI Overview the exact question the page answers. The page is relevant. It is also redundant with three other pages the system already has, so it loses the comparison every time a citation gets chosen. Agencies tracking rankings alone will not see this happening. Agencies tracking citation rate will see it immediately, and it usually traces back to the same root cause: the page was built to be complete, not to be different.
Where the Difference Actually Comes From
Production at scale requires structure. A fulfillment workflow serving agency partners across dozens of verticals runs on repeatable process: consistent formatting, consistent research method, consistent quality bar. That structure is what makes volume possible.
What that structure cannot supply is knowledge specific to one client’s market. An HVAC page written for a general audience and an HVAC page written for a client whose team knows exactly which local housing stock drives their emergency calls are not the same page in Google’s information gain sense, even when both are equally well written.
For sake oof argument we will pretend to be in a city in Canada, Edmonton. So, Picture two versions of the same Edmonton HVAC service page. Version one covers furnace maintenance the way most produced pages do: seasonal checklist, filter replacement schedule, generic warning signs. Version two adds one paragraph the agency’s account team supplied, noting that call volume in older Edmonton neighborhoods with original 1970s and 80s boiler systems spikes hard in the first cold snap of November, not January like the national data suggests, because those systems were never sized for the extended cold. That single paragraph is the only sentence on the page a fulfillment pipeline could not have produced without knowing this specific client. It is also the only sentence with a real shot at surviving an information gain comparison against the other four pages ranking for the same query.
The Test Before Anything Publishes
Before a piece of client content goes live, ask whether three direct competitors could publish the same page with their logo swapped in and nobody would notice. If yes, the structure is sound and the information gain is close to zero.
Passing that test does not require an enormous research budget. It requires one of a few things layered onto solid production work:
- A specific, sourced local data point the rest of the reference set does not have
- A documented client outcome with real numbers attached
- A practitioner’s stated disagreement with the common industry advice on the topic, with reasoning
- A regional or seasonal pattern only someone serving that specific market would know to mention
Note: That layer is a job for the agency’s account team, not the fulfillment pipeline, because it depends on knowing the client rather than the topic.
The Workflow Change That Actually Matters
None of this requires a new tool. It requires one new checkpoint before publication: someone who actually knows the client asks what this specific piece says that a generic version of the same topic would not. If the honest answer is nothing, the piece needs one more pass before it goes live.
Agencies that build this checkpoint into every retainer are the ones whose content keeps showing up in AI-generated answers months after publication, while structurally similar content from the same production pipeline quietly stops getting cited at all. Snap SEO, an Edmonton agency, runs this exact check before marking any page finished. Aurora Admin, an aurora forecasting platform the agency built and operates as a proof of concept for the approach, currently draws roughly 21,000 clicks per month and holds more than 100 first-position rankings across its location pages, each one built around forecasting detail specific to that location rather than aurora content repurposed across markets.
The patent did not create the gap between the pages that get cited and the pages that do not. It just gave agencies a mechanical explanation for why it exists, and a name for the discipline that closes it: information gain SEO.