They underperform in rankings despite looking substantial on a word-count basis, because Google’s quality evaluation, described directly in its “creating helpful, reliable, people-first content” guidance, assesses whether a page adds information not already well covered elsewhere for that query, and word count isn’t a proxy for that assessment. A programmatic page can hit 1,500 words and still be judged low-value if those words simply restate data or explanations that are already widely and better covered across the web, because length measures how much text exists, not whether any of it is genuinely new or useful relative to what a searcher could already find.
Why length and information gain are different axes entirely
Traditional content guidance, and a lot of legacy SEO practice, treated word count as a rough quality proxy: longer content historically correlated with more comprehensive coverage of a topic, and comprehensive coverage tends to rank well. That correlation was always incidental, not causal, and Google has been direct that length itself isn’t evaluated as a quality signal. Google’s helpful content guidance explicitly deprioritizes length in favor of whether content demonstrates genuine expertise and provides value beyond what’s already available for a given query.
Programmatic content is where this distinction becomes concrete and costly, because it’s entirely possible to build a template that generates long pages without generating informative ones. A page assembled from a data table, restated in prose form, padded with generic contextual paragraphs (industry background, generic definitions, boilerplate calls to action) can easily reach traditional length thresholds while containing almost nothing that a searcher couldn’t already get from a dozen other sources covering the same underlying data. The page is long because the template adds narrative padding around the same core facts, not because it contains more actual information than a much shorter page stating the same facts directly would.
This is conceptually connected to discussions around the leaked Google API documentation and Google’s own helpful content framing, both of which point toward evaluating whether a page’s content is genuinely additive relative to the existing corpus of information already available for a query, rather than being a purely mechanical density or length calculation. It’s important to be precise here: there’s no publicly disclosed, named “information gain score” that Google exposes as a literal 0-to-100 metric. It’s a conceptual evaluation criterion embedded in ranking systems and helpful content guidance, not a specific, checkable number a page either passes or fails.
What this looks like at scale
The failure pattern shows up most clearly across large programmatic page sets built from the same underlying dataset: city-by-city service pages, product variant pages, or data-lookup pages that repackage the same source information across many URLs with only superficial rewording between them. Each individual page might read as reasonably complete on its own, long enough, structured with headings, covering the expected fields, and still collectively represent a large template output where none of the pages add anything a competitor’s page, or a more authoritative single source, doesn’t already provide. Google’s evaluation, especially at a pattern or template level rather than a single-page level, can recognize this as low aggregate information value even when individual pages look substantial.
A worked comparison: two templates producing the same word count, different outcomes
Consider two programmatic templates both generating roughly 1,200-word pages for a “best X in [city]” style query set. Template A pulls a handful of local data points (business count, average rating drawn from a data provider, a couple of neighborhood names), then wraps those facts in generative paragraphs of general industry context, a boilerplate explanation of how the category of service works, and a closing call to action, repeated across every city with only the city name and the pulled data points swapped in. Template B pulls the same underlying local data points but uses them to drive genuinely city-specific structure: which specific neighborhoods the data points concentrate in, how the local figures compare to a broader regional baseline, and a short section addressing a city-specific practical consideration (permitting quirks, seasonal factors, local regulatory context) that doesn’t appear on the other cities’ pages in the same set.
Both templates hit the same length target and both technically incorporate real data. The difference Google’s evaluation is sensitive to is that Template A’s word count comes almost entirely from content that’s identical in substance across every page in the set, meaning a searcher landing on any single page from that template gains nothing they wouldn’t gain from any other page in the same set or from a generic explainer elsewhere on the web, while Template B’s additional length comes from content that’s genuinely different per page and tied to information not trivially available elsewhere in that combination. A pattern-level evaluation of the two templates, looking across many pages from each rather than one page in isolation, surfaces this difference clearly even though a single page from either template might look comparably substantial in isolation. This is why auditing a programmatic template for information gain has to happen across a representative sample of its output, not just by reading one exemplar page and judging it complete.
The comparison also clarifies a common misdiagnosis: teams that notice a programmatic template underperforming often respond by adding more sections to the template, an FAQ block, a related-topics section, an extra paragraph of context, on the assumption that more comprehensive-looking output will resolve the issue. If the added sections are themselves generated from the same limited underlying data and repeat the same generic framing across every page in the set, this makes the page longer without changing which axis it was failing on. The template gets closer to Template A in the comparison above, not further from it, because the new content is still substance that’s identical or near-identical across the set rather than substance that’s genuinely tied to what’s different about each individual page. Growing a low information-gain template usually produces a longer low information-gain template, not a fixed one, unless the added content is specifically the kind of per-page-distinct material that was missing in the first place.
Practical implication
Don’t use word count as a proxy for whether a programmatic template is producing valuable pages. Instead, evaluate each template against a direct question: if a searcher landed on this page, would they learn something they couldn’t get faster or more completely from an existing, better-established source, or does this page just restate available data in more words. If the honest answer is that the page’s length comes from narrative padding around commonly available facts rather than from genuinely additive analysis, unique data, or synthesis not available elsewhere, the fix isn’t adding more content to the template, it’s identifying what would make the underlying information genuinely useful (unique analysis, a genuinely differentiated angle, data not available elsewhere, practical synthesis a searcher can’t easily get by piecing together several existing sources) and building that into the template, even if the resulting pages end up shorter than the current output. A template redesigned around genuine information gain, even at reduced length, is a stronger long-term bet than a template optimized to hit a length target while adding no new value to what’s already indexed.
