The quality threshold: when a generated page deserves to exist

Most programmatic projects fail on a decision made before any content exists: they generate a page for a row that never warranted one. The quality gate belongs upstream of the template, not in the editing pass afterwards.

The four-test gate

Every row in the dataset passes these four tests before generation. One failure removes the row from the batch.

1. Demand test. Is the query attested — measured volume, autocomplete, related searches, or existing impressions in Search Console? Extrapolating from the template is not evidence.
2. Format test. Does the page type ranking in the top five match what the template produces? A guide will not rank where product pages dominate, however good it is.
3. Data test. Does the page carry at least one verifiable fact that the current results do not display? If not, it is another duplicate.
4. Next-step test. Is there a logical action after reading? A page with no continuation generates no engagement signal at all.

The deletion test

The fastest way to judge a finished template is subtractive. Take one generated page and remove every sentence that also appears on another page in the set.

Running this test on twenty random pages of a batch takes an hour and predicts the indexation outcome better than any content score.

What "quality" resolves to, operationally

Quality is not a score you can assert. On a generated set it decomposes into four observable properties.

• Specificity — the number of facts on the page that are true only of this entity.
• Completeness — whether the page answers the follow-up questions the query implies, not just the query itself.
• Accuracy — whether the data is current. A stale price or a discontinued product turns a good page into a liability.
• Accessibility — whether the answer is reachable in seconds, or buried under three paragraphs of throat-clearing.

All four are measurable at generation time and can gate publication automatically, in the same way title uniqueness checks do. A batch that fails all four is not a programmatic site — see programmatic SEO vs the AI content farm.

Pruning: the half of the job nobody schedules

A generated set is not a shipped artefact, it is a population with a mortality rate. Pages that never earn an impression consume crawl budget and drag down the average quality signal of the whole domain.

every quarter:
  for page in batch:
    if age > 6 months and impressions == 0:
      merge into the nearest hub, or remove and return 410
    if age > 12 months and clicks == 0 and impressions < 50:
      rewrite with fresh data, or remove

Removing 20% of a batch routinely improves the indexation rate of the remaining 80%. The mechanism is explained in crawl budget on large sites.

Where the floor actually sits

There is no word count that makes a page acceptable and none that makes it thin. A 300-word page answering a narrow question with exact data outperforms a 2,000-word page that circles the topic. The floor is informational, not dimensional: does the reader leave with something they did not have, and could not have got from the result above yours?

That is also, almost verbatim, the test behind Google's scaled content policy.