The word impression does the dirtiest work in advertising. It sits in one column of every report, adding together events that have almost nothing in common. A search ad read by someone who just typed the product category into a box. A full-screen video that owned a phone display for four seconds. A 300x250 that technically rendered in the far right rail of a page nobody scrolled. One word. One column. One sum at the bottom that means nothing at all.
I have watched that column decide budgets for twenty years. It deserves a closer look than it gets.
What each channel actually certifies
Start with what the machinery can honestly promise, channel by channel, because the promises are wildly different.
A paid search impression certifies more than any other unit in media. The person requested the page. They typed words describing a need, and the ad appeared inside the answer to their own question, in the exact spot their eyes were already going. Search impressions are also counted cleanly: the page either served or it did not, and there is no ambiguity about whether the ad was in view. When search "underperforms," the impression is almost never the suspect.
A paid social impression certifies possession of the viewport. The feed is full-bleed on a device held a foot from the face, one story at a time, and the platform knows to the millisecond how long your ad held the screen. That is a genuinely strong unit, with one giant asterisk: scroll velocity. Possession of the viewport for 400 milliseconds is possession in name only. The platforms know this too, which is why their own creative guidance is obsessed with the first second.
A standard display impression certifies that pixels were delivered. The MRC viewability standard, the one the industry fought about for a decade, asks for half the ad to be on screen for one continuous second. Half. One second. That is the ceiling of what the unit promises, and a meaningful share of bought impressions never clear even that bar. I wrote about where the rest of the money goes in a third of your programmatic budget never reaches an impression; this piece is about the impressions that do arrive, because even the delivered ones are not what the column implies.
CTV sits somewhere better and worse at once: the ad is full-screen and unskippable, but the unit is a household, the room may be empty, and the phone in the viewer's hand is winning the attention contest during your spot more often than anyone reports.
Banner blindness is not a metaphor
The reason a delivered, viewable display impression still is not a search impression has a name, and it is one of the oldest findings in web research. Banner blindness was documented in the late 1990s: users' eyes learn the shapes and positions where ads live and stop visiting those regions entirely. Not consciously skipping. Not seeing. Eye-tracking studies have replicated it for twenty-five years across layouts, devices, and generations, and the effect gets stronger with exposure, because it is a learned filter and the training never stops.
This is why display CPMs are what they are. The market has priced the blindness in. A two-dollar CPM is not a bargain version of a thirty-dollar social CPM; it is the fair price for a unit of maybe-glanced. Buying display is perfectly rational as long as you buy it as what it is: enormous, cheap, low-attention surface area that works through repetition and peripheral familiarity, plus the logistical value of retargeting pipes. The mistake is not owning display. The mistake is putting its impressions in the same column as search and reading the sum.
Frequency is a per-channel number, or it is fiction
Here is where the false unit does real damage. The industry inherited a folk rule, effective frequency of three, from television research done when there were three networks and one screen per household. Applied across modern channels as a blended number, it is meters added to gallons.
Think about what frequency is for. It exists to answer one question: how many exposures does it take for a message to register, and when do additional exposures stop adding and start subtracting? The answer depends entirely on how much attention each exposure carries, and attention per impression varies across channels by a factor of ten to a hundred.
Search needs no frequency management at all in the classic sense; exposure recurs exactly as often as the need does, and capping it means declining to answer someone's question. Paid social needs aggressive frequency discipline, not because users are overexposed to your brand but because each exposure is high-attention, so wear-out arrives fast. The decay is measurable and predictable, and ad fatigue has math you can use to schedule creative rotation before the curve turns. On a high-attention channel, frequency management is really creative-lifecycle management.
Display inverts the logic. Because each impression carries a sliver of attention, breaking through the learned blindness takes many more exposures, and each additional one is nearly free. Display frequency caps exist mostly to stop waste on the structurally unreachable, the person who will never look at the right rail no matter what, rather than to prevent fatigue that low-attention exposure rarely generates. Running display at social-style caps buys you the worst of both: not enough repetition to register, capped for a fatigue that was never coming.
CTV needs household-level thinking, a cap around two or three per household per week, and an honest accounting for co-viewing that most dashboards skip.
Run the same number across all of them and you have managed nothing. A blended frequency of 4.2 is an average of quantities measured in different units, and the metric going up is usually hiding the one going down: capped social depriving your best channel of its optimal exposure while uncapped display racks up invisible repetition against blind spots.
What the accurate channels earn
The premise in this post's title cuts one more way. Because search and social impressions are counted more accurately and carry more certified attention, they earn something display never can: the right to be evaluated per impression at all. It is fair to interrogate a social campaign's per-impression outcomes, because the unit means something. Interrogating display per impression flatters or damns it at random, because the unit is noise. Display should be bought and judged in aggregate, on reach into audiences the priced channels cannot cover and on incrementality measured with holdouts, never on the arithmetic of its impression column.
That distinction also explains a pattern every buyer has seen: shift budget from display into social, watch blended CPM triple, and watch results improve. Nothing paradoxical happened. You traded a large number of weak units for a small number of strong ones, and the report that treats all units as equal was the only thing surprised.
The plan that follows
Treat the impression column as a currency conversion problem. Set frequency rules per channel, derived from that channel's attention profile: none for search, creative-rotation-driven for social, generous with reachability filtering for display, household-capped for CTV. Report reach and frequency within channels and refuse to bless the blended number. Weight cross-channel comparisons by attention, even crudely; a rough attention multiplier beats a precise count of unequal things. And when someone presents the summed impression total in a quarterly deck, ask them what a single one of those impressions certifies. The silence is the beginning of a better media plan.
The column was never lying, exactly. It was counting. The lie was ours, every time we read a count of different things as a measure of one thing.