Information gain measures how much new information a page adds beyond what a searcher has already seen. It rewards content for being different, not merely better — and that distinction is the entire point. Most teams optimize for "better": more comprehensive, more words, more subheadings. Comprehensiveness is now the cheapest thing on the internet to produce, which means it has stopped being a differentiator. What's still scarce is information nobody else holds.

Four properties of the mechanism drive everything downstream:

A fair caveat: no live "information gain score" has ever been confirmed. The mechanism is described in a granted patent, but many patents never ship. What is certain is that the underlying preference is real and consistently rewarded — quality guidance repeatedly asks whether content is original and adds substantial value versus other results, sites that win core updates reliably over-index on proprietary data and tools, and engines already re-rank mid-session on real behavior. So the goal isn't to chase a score. It's to be the source of something that isn't consensus yet. Value sits at the edges of a topic, because the core is already saturated.

No category on the web is more duplicative than product content. Retailers paste manufacturer spec sheets verbatim, resell identical SKUs with identical descriptions, and mirror each other's category copy word for word. Measured semantically, most of the shopping web sits directly on top of itself.

That cuts both ways. The downside: the sameness penalty falls harder here than anywhere else. The upside: in a field where almost everyone publishes the same thing, the brand that publishes something genuinely original stands out by a margin no content-saturated vertical can match. And the stakes are commercial — "best [product] for [use case]," "[brand] vs [competitor]," "is it worth it," "which size am I" are increasingly answered by engines that cite a handful of sources and drop everyone else. Whether your brand is named depends almost entirely on whether you hold information the consensus doesn't.

This is the step most teams skip, and skipping it wastes the data. Two things must be true before original data pays off: the page has to already cover what a searcher expects (topical authority is the cost of entry), and you have to know precisely what the results set already says (so your data fills a real gap instead of repeating one).

Map intent and find the underserved one. Pull the results page and cluster ranking pages by intent and the rough traffic share each captures. If one intent absorbs the overwhelming majority, that's the consensus — still covered, but the opportunity is the underserved intent, or one not represented at all yet. Ahrefs' SERP Overview (Identify Intents) does this clustering for you, or Keyword Insights if you're working a large keyword set.

Run a content gap analysis to build the authority layer. Grade your draft against ranking pages and work the lowest-scoring topics up first. Hold this distinction firmly: matching what competitors say raises your coverage score and gets you ranked; adding what no competitor has is what creates gain. Get credible on the topic first, then layer the novel data. That sequence governs the whole process. Ahrefs' AI Content Helper, Surfer, or Clearscope will score your draft against the ranking pages and expose exactly what each competitor covers.

Harvest questions nobody has answered. Filter your keyword tool to questions, sort by 3/6/12-month growth, then narrow to questions the page-one competitors don't rank for. For ecommerce, the valuable ones cluster around fit, longevity, compatibility, real-world performance, and regret — the pre-purchase anxieties product pages rarely address. Use Ahrefs Keywords Explorer (Matching terms → Questions) for the growth data, and AlsoAsked to map the branching "People Also Ask" trees where the unanswered follow-ups hide.

Intercept community discussion. When forum threads rank on page one, read it as a signal the other results aren't satisfying intent — and that real shoppers are stating, in their own words, exactly what the product pages never told them. That framing is often precisely what your data should answer. A plain site:reddit.com [topic] search is the durable method here, alongside mining your own reviews and marketplace reviews for the recurring objection no page quantifies.

The output isn't a page. It's a specific, evidenced gap — and its nature decides which method comes next.

Your structural advantage is that you sit on data no competitor, analyst, or model can reproduce, because it's generated by your own transactions. Before commissioning anything new, inventory it: returns and exchange reasons (the richest and most under-used source — why people send things back is a question every buyer has and almost no page answers), reorder and subscription timing, review and support text, on-site search queries, attach and bundle behavior, and variant/size/colour distribution. Each can yield a figure that exists nowhere else — which is the definition of the fringe.

Because gain is measured on meaning, the data has to be genuinely new in semantic space, not an old number rephrased. Pick the method by the kind of gap discovery surfaced.

Use when the fringe is a question about behavior, preference, or opinion no public dataset answers. Frame one headline question the category argues about but has never measured, add four to six segmentation questions so a single result can be sliced into several figures, distribute cheaply, target a few hundred responses, and disclose sample size and dates.

Example — commuter cycling brand: Discovery surfaces a growing, unanswered question about how people actually use an e-bike after buying — something the aspirational category marketing never addresses. A survey of 600 customers finds the large majority of rides are under five kilometres and overwhelmingly for commuting, with a sharp split by city size. It reframes the whole "which bike should I buy" conversation around real usage instead of aspiration.

Use when the fringe is a pattern or trend already sitting in your systems. This is the highest-leverage method for ecommerce — the data is unique by construction and costs nothing new to collect. Start from your inventory, look for a pattern-with-a-number the outside world finds interesting (not an internal KPI like AOV), anonymize and aggregate, then validate against a second cut so you don't publish a fluke.

Example — denim returns data: "Which size am I?" is the most-asked, least-answered question in apparel, and community threads outrank retailer pages on it. Analyzing tens of thousands of orders and exchanges, the brand publishes true-fit data by body measurement — how a given waist-and-inseam combination actually fits across each cut, and which styles run large or small. It answers the exact question the category dodges, lowers the brand's own return rate, and becomes the page every fit query has to reckon with.

Use when the fringe is a claim everyone repeats but nobody has verified, and you have neither a list nor rich internal data. Identify the unverified adjective ("durable," "long-lasting," "gentle"), design the smallest test that produces a countable number, log method, sample size, conditions, and dates as you go, and include named competitors where legitimate — a comparative result is far more citable than a solo one.

Example — luggage durability test: Every brand claims "durable" and none proves it. A documented stress test — repeated drop, crush, and zip-cycle counts across the brand's own range and several named competitors, published in full including where its own product underperformed — converts an empty adjective into a defensible set of numbers, and produces exactly the comparison an answer engine reaches for.

That's the foundation — the concept, why ecommerce is uniquely exposed, and the three ways to produce data that actually fills a gap. The part that decides whether any of it earns citations comes next in the full SOP: how to package a study so it survives being lifted out of context by an engine — the canonical hub URL, the self-contained headline claim, the key-findings box, question-based H2s pulled from discovery, the methodology section, and the exact schema stack for each page type — plus the per-method publishing patterns (research hub vs. fit hub vs. methodology page), the quarterly refresh workflow that keeps gain from decaying into baseline, and the three signals that tell you it's compounding into named citations.

The complete SOP — packaging checklist, per-method AEO publishing patterns, refresh workflow, and measurement framework. My gift to you :)

🖼️ Hedra - Turn an image plus audio into talking, expressive AI avatar videos.

📹 Runway - Pro-grade AI video generation and editing for creators and studios.

🤖 Synthesia - Avatar-led explainer and training videos in 140+ languages.

🎧 Decagon - AI support agents that resolve refunds, account and order tasks end to end.

😍It was great, all I needed

🙂Good, not great

🙄It sucked