The same native unit produces very different results depending on where it sits within an article. The variation is driven by what the reader is doing at that point rather than by visibility alone.
Readers at the top have not committed
A unit placed above or immediately after the opening paragraphs reaches everyone who arrived, including the large share who will leave within seconds.
That audience has not yet decided whether the article is worth reading, so a recommendation competes with the content the reader came for.
Volume is high and quality is low, which produces impressive impression counts and weak downstream behaviour.
The end of an article is a decision point
A reader who finishes a piece has a genuine moment of choice about what to do next, and recommendations at that point answer a question the reader is already asking.
Fewer people reach this position, so inventory is scarcer, but the response rate and the quality of the resulting visit are both markedly better.
This is the structural reason end-of-article units command higher rates despite delivering a fraction of the impressions available at the top.
In-content placement interrupts
Units inserted between paragraphs are seen by engaged readers, which is valuable, but they arrive mid-thought rather than at a natural pause.
Performance depends heavily on whether the insertion respects the article's structure, and automated insertion by paragraph count frequently does not.
Publishers who place these positions manually or by section boundary generally see better results than those inserting at a fixed interval.
Position interacts with article length
On a short article, the top and the end are only a screen apart, and the distinction between them largely disappears.
On a long feature, the reader who reaches the end has invested considerable time and represents a much more selected audience.
Applying one placement policy across a mixed archive therefore produces inconsistent results that look like creative variance but are structural.
Density erodes every position
Adding more units raises impressions and reduces the response to each one, because attention is finite and repetition within a page is immediate.
The revenue curve rises and then falls, and the turning point differs by audience, by device and by the type of content being read.
Publishers who measure revenue per session rather than per impression find that point, while those measuring per impression rarely notice they have passed it.