Skip to content

Content decay

Pages people stopped reading while the machines did not.

A page that was doing well and is now doing badly is an old story. A page whose human traffic collapsed while agent traffic held steady is a newer one, and it usually means the page is still being read, through something else.

http
GET /api/content/decay

What it looks for

The report compares this period against the one before it, over one statement, so the two halves cannot come from two different scans of a moving table. A page appears when all three hold:

  • human pageviews fell by at least 30%
  • agent fetches held at 90% or more of their previous level
  • the previous period had at least 20 human pageviews

That last threshold is the one doing quiet work: without it, every page that had three visits last month and one this month would be a crisis.

Each row carries both periods for both audiences, the percentage change in each, and the answer-engine arrivals for the current period, so you can see immediately whether the lost readers came back through an answer instead of not at all.

Field What it is
human_pageviews, human_pageviews_prev Both periods.
human_change_pct Negative by construction.
agent_fetches, agent_fetches_prev Both periods.
agent_change_pct Absent when the earlier period had none.
answer_arrivals This period, for context.

Reading a row

Humans down, agents flat, arrivals up. The page is working differently rather than failing. Readers are getting the answer somewhere else and some of them are coming through afterwards. Worth checking that what an answer would quote is still correct.

Humans down, agents flat, arrivals zero. The page is being taken and returning nothing. This is the Citation Gap verdict extracted, seen through time instead of through a ratio.

Humans down, agents down too. Not decay in this sense: the report will not show it. That is an ordinary ranking or interest problem, and the pages screen is where to look.

What it is not

It is not a ranking tool, and it does not know why. It cannot tell a page that lost search position from a page whose answer got absorbed, because the referrer of a reader who never arrived does not exist. What it can tell you is that the machines did not lose interest at the same moment your readers did, which is a fact worth having and one that no tool with a single audience can produce.

Requirements

  • A previous period. With a range that has nothing before it, the report is empty.
  • Agent traffic, which means a shipped server log. Without it, every page shows agents at zero for both periods and nothing can pass the “held steady” test.