The everyday name for this is banner blindness. People do not skip advertising because they dislike it — they skip whatever sits where advertising sits, including things that are not advertising, including things they went to the page to find.
The demonstration has to separate two explanations that look identical from the outside. Something you cannot remember might have been too small, too faint, or too far from where you were reading. Or it might have been perfectly visible, in a normal place, and simply filtered.
So every screen below carries two labels of exactly the same size, in the same typeface, at the same size and colour. One of them sits where banners sit. The other sits inside the content, near the thing you were asked to do. Neither is small and neither is faint. The only difference between them is the part of the page they occupy.
Your job on each screen is to count something. Do that, and let the rest look after itself — it will.
What is being counted, and what is not
The counting task exists to give your attention somewhere to be. It is not scored, it is not the point, and getting it wrong costs nothing — it is there so that the labels are not what you are doing. A demonstration that asked you to look at a screen and then tested what you saw would be a test of looking, and everybody passes that.
Afterwards there is a recognition test over twenty words: the five from banner slots, the five from content, and ten that were never on screen at all. Those ten are the control. Picking them is a false alarm, and a run with several of them is a run where somebody was clicking generously rather than remembering, in which case the comparison is suppressed rather than reported.
What it costs you
Position is a filter before it is a location. Anything you put in a banner slot, a right rail, or a full-width strip above the content inherits the fate of everything else that has ever lived there. That includes your cookie notice, your service message and your one genuinely important announcement.
Making it bigger and louder is the reflex, and it is the wrong lever. The filter is spatial, not a matter of contrast — a larger banner in a banner slot is a larger thing being skipped. Move it into the content and it becomes visible at half the size.
Anything shaped like an advertisement will be treated as one. A rectangle with a border and a headline is a shape people have learned to discard, whoever wrote it.
This is what makes "users didn't see it" so hard to argue about. The people who missed it were looking straight at it, so nobody involved has an accurate account of what happened — not the visitor, who is certain it was not there, and not the team, who can see it plainly in the screenshot.