Content velocity: how many pages a small site can add without harm

Google's spam policies, updated in March 2024, name scaled content abuse: producing many pages primarily to manipulate rankings rather than to help people, whatever produces them, automation, humans, or both.

• The number of pages is not the test. Large sites publish thousands of pages a day without issue.
• The producer is not the test either. The policy explicitly covers human-written pages made at scale.
• The test is whether the pages exist to help someone, and whether each holds value of its own.

The same guidance is clear that Google does not penalise AI assistance as such. A hundred pages written with a model, each holding real data and a real answer, is not what the policy targets. Ten pages spun from one template with the nouns swapped is.

Before publishing a batch, check every page against this. If a page fails all four, it should not ship, whatever the batch size.

Some families are legitimately large because the underlying questions are legitimately many. They share one property: the difference between two pages in the family is real.

• One page per error message or status, where each has a different cause and a different fix.
• One page per term, where each definition is more precise than what ranks today.
• One page per tool comparison, where each holds a judgement about who should choose what.
• One page per task in documentation, where each is something a user must actually perform.
• One page per data slice, where you hold the data and nobody else publishes it.

The families that do not earn their scale are the ones built from a grid: every city crossed with every service, every tool crossed with every integration, when you have nothing to say about most of the cells. Those are the doorway pages the policies describe.

There is no published limit, so the constraint is practical rather than algorithmic. A batch is too big when you can no longer check it, or keep it accurate.

1. 1

   Read three pages of the batch in full

   Not skim: read. If you find an invented figure, a paragraph that says nothing, or a claim you cannot source, the batch is not ready and the sample was random.

2. 2

   Check the batch against the four tests

   Method, example, data, check. Pages failing all four come out of the batch.

3. 3

   Publish and stop

   Ship the batch, then wait for the index to judge it before shipping the next. A month is enough to see the pattern.

4. 4

   Size the next batch by what you can maintain

   Every page joins the quarterly review. A hundred pages is a hundred pages to keep accurate, forever.

After a batch, you are watching for a judgement about the family rather than about individual pages. Give it three to six weeks, then read three reports.

1. 1

   Stop publishing that family

   Before anything else. Adding more pages to a family Google has judged makes the signal louder.

2. 2

   Decide page by page

   Keep the ones with impressions or links. Merge the near-duplicates into the strongest URL with a 301. Remove the rest with a 410.

3. 3

   Take them out of the sitemap

   The sitemap should list the pages you want indexed, not everything you ever published.

4. 4

   Wait a full cycle

   Recovery from a quality judgement is slow and often waits for a broader update. Do not undo the cleanup because nothing moved in two weeks.

The library that survives this is the one where every page would have been worth writing on its own. That is the whole rule, and it is simpler than any threshold.