Discussion
  • Technical SEO

Database-built ‘X vs Y’ pages: how much hand-written content did you need before they held rankings through updates?

Published 1 reply

Discussion starter: a programmatic SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    There is no word count that makes database-built comparison pages safe, and “how much hand-written content” is the wrong unit to plan with. The comparison pages that tend to hold through core updates share two traits. First, each indexable pair has a reason to exist: real search demand and a meaningful difference between the two items. Second, each page answers the question a reader brings to it (which should I choose, and why) rather than placing two spec sheets side by side.

    Hand-written content matters where it changes that answer: the verdict, the “choose A if, choose B if” reasoning, and the caveats the data cannot express. It matters most on the pairs with demand. Spread thinly across thousands of pairs, it becomes templated prose with a human in the loop, which does not fix the underlying problem. For most sites, the number of indexable pairs is a bigger decision than the amount of text on each one.

    What Google’s documentation actually sets as the bar

    Google is explicit that length is not the target. Its helpful content guidance asks, “Are you writing to a particular word count because you’ve heard or read that Google has a preferred word count? (No, we don’t.)” The same page asks whether content provides “original information, reporting, research, or analysis”, whether it provides “substantial value when compared to other pages in search results”, and whether you are “using extensive automation to produce content on many topics”.

    The spam policies define scaled content abuse as “when many pages are generated for the primary purpose of manipulating search rankings and not helping users”. One of the listed examples is close to a description of a weak X vs Y template: “Stitching or combining content from different web pages without adding value.” Automation itself is not the violation. A comparison page built from two of your own product records is, by construction, combining two pages. The question Google’s guidance keeps returning to is whether the combination adds anything.

    The helpful content system was folded into core ranking systems in 2024, and 2026 has already brought two completed core updates (March and May). That is why “held through updates” is the right test. A large programmatic section is judged repeatedly, not once at launch.

    Why X vs Y databases are riskier than they look

    The combinatorics work against you. A database of n items produces n(n-1)/2 possible pairs. Five hundred products give 124,750 pairs. Two thousand give just under two million. Search demand typically exists for a small fraction of those, concentrated on items people actually shortlist together.

    The pages also overlap with each other heavily. “A vs B” and “A vs C” share half their data. If the only differentiated element is the other item’s spec row, Google sees thousands of pages that are mostly the same, each assembled from pages that already exist on your site. That pattern is the programmatic version of what the helpful content questions warn about, whatever the template looks like.

    The layers that make a comparison page worth indexing

    It helps to think of a comparison page as a stack of layers, each one harder to produce than the last. The lower layers are cheap and look similar across every site with the same data. The upper layers are where pages become distinct.

    Layer Example Who can produce it Distinctiveness
    Side-by-side specs Weight, price, battery, dimensions Database Low: any competitor with the same feed has it
    Computed differences “B is 18% lighter”, normalised units, which item leads on each attribute Database with rules Low to moderate
    Proprietary data Test results, price history, compatibility checks, aggregated owner reviews, return rates Database, if you own the data High
    Context What an 18% weight difference means for someone who carries it daily Human, or rules written by a subject expert Moderate to high
    Verdict by use case “Choose A for travel, B if you need the larger screen” Human with product knowledge High
    Expert judgement and caveats Known faults, what the spec sheet hides, when neither is the right choice Human with real experience Highest

    A page built only from the first two layers is what most sites launch. A page with proprietary data plus a short, specific verdict can be genuinely useful at modest length. Several hundred words of generic introduction add length without adding a layer, and a reader can tell the difference immediately.

    Where to spend hand-written effort

    A tiered model keeps the editorial cost proportional to the value of each pair. The thresholds below are placeholders to set from your own demand data, not tested cut-offs.

    Tier Which pairs Treatment Indexing
    1 Pairs with clear search demand and commercial value Hand-written verdict and use-case guidance from someone who knows the products, plus any proprietary data. Reviewed when either item changes. Indexable, linked from both product pages and relevant categories
    2 Pairs with some demand and a meaningful difference Data-rich template with computed differences and proprietary data, plus a short rule-based verdict written by an expert once per product type Indexable if the page says something the two product pages do not
    3 Everything else On-site comparison tool generated on demand for users Noindex, or not generated as static URLs at all

    Two implementation details save a lot of trouble. Pick one canonical order for each pair (alphabetical is simplest) and redirect the reverse order, so “A vs B” and “B vs A” do not compete. And make sure tier 3 comparisons do not leak into crawlable internal links, or you will generate the pairs you meant to keep out.

    Hypothetical example:

    A retailer has 800 laptops, which is roughly 320,000 possible pairs. Keyword data shows demand for about 1,500 pairs, mostly within the same brand and price band. The team hand-writes verdicts for the 200 pairs that carry most of that demand, publishes 1,300 tier 2 pages built on its own battery and thermal test data, and leaves the rest to an on-site comparison tool. The indexable section is 1,500 URLs instead of 320,000, and every one of them contains something a manufacturer spec sheet does not.

    How to tell whether the section is holding up

    Rankings on a handful of head pairs will not tell you much. I would track the section as a whole, using a page filter or a dedicated Search Console property for the comparison directory:

    1. Indexed ratio. Indexed comparison URLs as a share of submitted ones. A rising count in “Crawled - currently not indexed” for the directory is often the earliest sign that Google has decided many pairs add little.
    2. Pages with any impressions. The share of indexed comparison pages with at least one impression in 90 days. A section where most indexed pages get nothing is carrying weight without earning it.
    3. Clicks per indexed page by tier. This shows whether tier 1 effort is paying back relative to tier 2.
    4. Behaviour around each core update. Compare the section’s clicks and indexed count in the weeks before and after each rollout, excluding the rollout window itself, and compare against the rest of the site.

    That last comparison matters because the risk is not limited to the comparison pages. Google has said its core systems work mainly at page level but also consider some site-wide signals. A very large mass of low-value pages can plausibly affect how a site is assessed overall. The public evidence on how much it weighs is thin, but a section with a low impressions share and a high “Crawled - currently not indexed” count is the one I would look at first after a sitewide drop.

    Testing whether the hand-written layer earns its cost

    You can test the value of a verdict layer, but not whether it protects you through the next core update. For the first question: take tier 2 pairs with similar demand and current performance, split them into matched groups, add a hand-written verdict to one group, and leave the other unchanged. Measure impressions, clicks and CTR over 8 to 12 weeks, and on-page engagement or assisted conversions as secondary metrics. Keep deployment to a single date, and treat any core update inside the window as a confounder that may invalidate the comparison.

    The result that should change the plan is clear either way. If the verdict group gains measurably, extend the tier 1 treatment further down the demand curve. If it does not, the constraint is probably the data layer or the number of indexable pairs, and more writing will not fix it.

    Watch for this:

    Hand-written does not mean distinct. If writers produce 150 formulaic words for each of 5,000 pairs, or an AI model generates “verdicts” with no underlying data, you have scaled content with extra cost. Stale data is the other quiet failure: a comparison showing last year’s price or a discontinued model damages trust faster than thin copy does. Show when the data was last updated, and remove or update pairs when a product changes.

    If this were my site, I would decide the size of the indexable section before deciding the amount of copy. I would index only pairs with demand and a real difference, put hand-written verdicts on the pairs that carry most of the demand, invest in data only I have, and keep everything else in an on-site tool. Then the question to ask after each update is not “was there enough text” but “did the indexed share and the impressions share of the section hold”.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.