Content marketing is measured badly almost everywhere, and the reason is a mismatch of timescales. Content compounds over years; reporting happens monthly. Judge a compounding asset on a monthly cycle and you will consistently reach the wrong conclusion about it.
Why the obvious metric fails
Pageviews per article is what most people track, and it is wrong in three ways at once.
It is measured too early. An article published in March may not reach its steady state until September. Judging it in April tells you about your distribution on launch day, not about the article.
It ignores who arrived. Two thousand views from an aggregator that sends nobody who cares is worth less than fifty views from people searching for exactly your product's problem.
It treats each piece as independent. Much of the value of a body of content is that it makes the rest of it findable and credible.
Measure cohorts, not articles
The single most useful change: stop asking "how did this post do" and start asking "how is the content published in Q1 doing now".
Group articles by publication month. Track each cohort's traffic over the following twelve months. You get a curve rather than a number, and the curve is the actual shape of the thing you are buying.
What it reveals:
Time to plateau. Most search-driven content takes three to nine months to reach a steady state. If you are cancelling programmes at four months, you are cancelling them before they have started working.
Whether it decays. Some content plateaus and holds for years. Some peaks and declines. The difference tells you which kind of thing you are good at making.
Whether output is improving. Compare the shape of the Q1 cohort at month six against the Q3 cohort at month six. That is the only honest way to know whether your content is getting better, and it takes a year of data before it says anything.
Three metrics per article that are worth it
Once an article is past its first ninety days:
Entry-page conversion rate. Of the visits that started on this page, how many converted — on any timescale you count. Entry page rather than any pageview, because that is the acquisition question.
Assisted conversions. How often the page appears in the journey of a converting visitor without being the entry. Much content works this way: read during evaluation, not at arrival. Ignoring it systematically undervalues middle-of-funnel writing.
Search position for the query it targets. A leading indicator. Position improving with traffic flat means traffic is coming.
Two things almost nobody measures
Whether AI reads it
Increasingly the thing separating content that compounds from content that quietly stops working.
An assistant reading your article and answering from it produces no pageview. If your analytics only counts humans running JavaScript, that activity is completely invisible — crawlers do not execute scripts, so they are absent rather than undercounted.
Two numbers per article are worth having: how often AI crawlers fetch it, and how many visits arrive from AI assistants that landed on it. The ratio tells you whether the piece is being read and cited, or read and absorbed.
The second pattern — heavily crawled, never cited — is the specific failure mode of content that is a competent summary of general knowledge. Which is also the content most exposed to AI Overviews. Same problem, same fix: write things that require attribution to repeat.
Internal linking effects
Publishing a new article that links to five older ones usually lifts those five. Almost nobody looks, and it is one of the highest-return activities available — often higher than writing the new piece.
What a sensible report looks like
Quarterly, not monthly. Content does not move fast enough for a monthly report to say anything, and a monthly report on a slow-moving asset invites premature conclusions.
- Cohort curves — each quarter's output, traffic over time since publication
- Top ten entry pages by conversions, not by traffic
- Assisted conversions from content overall, as a share of all conversions
- AI crawl and citation ratio for the top twenty pages
- One qualitative note — what you learned that the numbers do not show
The honest limitation
Content marketing attribution is genuinely hard, and anyone claiming a clean number is simplifying.
Someone reads three articles over six months, forgets where, searches your brand name and converts. That is recorded as direct traffic or brand search. The content did the work; the last click gets the credit. No attribution model fully fixes this, because the causal chain runs through a person's memory and nothing in your infrastructure can see it.
The practical response is not a better model. It is to hold both things at once: measure what you can attribute, and accept that the real figure is higher than the measurable one by an unknown amount. Then avoid the mistake this causes — cutting the content that produces no last-click conversions, which is usually the content doing the persuading.
Common questions
How long before content marketing shows results?
Three to nine months for search-driven content to reach a steady state, longer in competitive areas. This is why judging individual articles monthly produces bad decisions — you are looking at launch-day distribution rather than at the article. Group posts by publication month and track each cohort's traffic over the following year instead.
What is the best metric for content marketing?
Entry-page conversion rate, measured after the first ninety days, combined with assisted conversions. The first answers whether the piece attracts the right people; the second captures content that is read during evaluation rather than at arrival, which is most middle-of-funnel writing and is systematically undervalued when you only count entry pages.
How do I know if AI is using my content?
Two numbers per article: how often AI crawlers fetch it, from server or edge logs, and how many visits arrive from AI assistants and land on it. Your analytics will not show the first, because crawlers do not run JavaScript. A page fetched constantly and cited never is usually a competent summary of general knowledge — the fix is originality that requires attribution to repeat.