What is Placement?
A placement is the specific surface an ad appeared on — a feed, a story, an in-stream break, a partner site, an app. Prices and response differ enormously between them, so a figure blended across placements describes none of them.
Where the ad actually was
A placement is the specific surface an impression appeared on. A main feed, a stories tray, a reels-style vertical surface, an in-stream video break, a search results page, a partner website, an app in an audience network, a marketplace listing.
These are not variations of one product. They differ in what the person is doing, how much of the screen the ad occupies, whether sound is on, whether attention was already engaged elsewhere, how easy the ad is to skip or scroll past, and what it costs to appear.
The consequence is that a campaign's headline figures are a weighted average across several unrelated markets, with the weighting chosen by the delivery algorithm rather than by the advertiser. The average describes none of the markets and can move substantially because the mix moved rather than because anything performed differently.
The differences that matter
Price. Cost per thousand impressions varies by multiples across placements in the same campaign. Feed inventory competes with everything; audience-network inventory frequently does not.
Viewability. A full-screen placement is seen. A unit below the fold on a partner page may never enter the viewport. Two placements at the same impression price can differ by half in what they actually cost per impression that had a chance of being seen.
Attention. An ad interrupting content somebody chose to watch is in a different position from one passed at speed in a scroll.
Format fit. A vertical full-screen surface and a square in-feed unit are different canvases. Creative designed for one and delivered on the other loses to cropping, to text that no longer fits, and to a composition that no longer reads.
Accidental interaction. Dense layouts with small tap targets produce clicks nobody intended, which inflates click counts and destroys the conversion rate of the traffic.
Why manual selection is usually the wrong move
The instinct on seeing a placement breakdown is to switch off the underperformers and concentrate on the rest.
Three reasons to resist it as a first move.
The delivery model is already doing this, continuously and at a granularity a weekly report cannot match. A manual selection is a coarser version of a decision the platform makes per auction.
The report is frequently too thin to support the conclusion. Splitting a campaign's conversions across a dozen placements leaves most of them with too few events to distinguish from noise.
And placements interact. Cheap upper-funnel inventory can contribute to conversions credited to a feed placement later, and excluding it removes a contribution that does not appear in its own row.
What manual control is genuinely for
Exclusion, on grounds the platform cannot evaluate.
Brand suitability. Environments a business will not appear in are a judgement about the business, not about performance.
Format incompatibility. Where the creative cannot work on a surface — a detailed static image on a fast vertical video placement — excluding is better than serving something that will fail.
Persistent waste with no upstream story. A placement consuming meaningful spend with no conversions and no plausible assisting role, over a long enough period to be more than noise.
Accidental-click inventory. Where destination analytics show traffic from a placement arriving and leaving instantly at scale, the exclusion is justified by evidence from outside the ad platform.
The category worth naming specifically is audience-network and partner inventory, which appears in most accounts, is cheap, and is where most placement complaints originate. Excluding it is a common and often correct decision — and worth testing rather than assuming, because in some accounts it is contributing.
Report separately, do not split structurally
The right response to placement variation is a breakdown, not a campaign per placement.
Splitting structurally fragments conversion data across entities that each need volume for the bidding to work, which usually costs more than the control gains. The exception is where a placement genuinely needs different creative and different messaging, in which case the split is about the creative rather than about the media.
The reporting habit worth building: read the placement breakdown before concluding anything about creative or audience. A creative test that looks decisive is frequently a placement-mix difference, because the delivery gave one variant more of a favourable surface.
Creative is the real placement lever
The most effective response to placement variation is not exclusion but making creative that suits each surface.
Vertical full-screen assets for vertical full-screen placements. Captions everywhere, because sound-off is the default on most feeds. A composition that survives cropping to several aspect ratios. A hook in the first moments for surfaces people scroll.
Platforms increasingly generate cropped variants automatically, and automatic cropping is a fallback rather than a solution — it protects against the worst outcome and does not produce a good one.
An account supplying properly formatted assets for each surface it buys can leave placement selection to the delivery model with far more confidence, because the model is no longer choosing between a good ad and a broken one.
Watching the mix rather than the average
Because the delivery model chooses the mix, the mix moves without anybody deciding, and the campaign's blended figures move with it.
This produces a recurring confusion. A campaign's cost per thousand impressions falls sharply, and the account concludes it is buying more efficiently. What has usually happened is that delivery shifted toward cheaper inventory, which lowers the average price and frequently lowers the outcomes with it.
The tell is the pair: a falling impression price with flat or falling conversions is a mix change, not an efficiency gain. Reading the placement breakdown confirms it in a moment.
The same applies to click-through rate, engagement rate and viewability, all of which vary far more by placement than by creative. Any of those moving without a creative change is a mix question first.
An honest default
For most performance accounts the workable position is narrow and stable.
Leave placement selection to the delivery model. Supply properly formatted creative for every surface the campaign can appear on. Maintain a short exclusion list built on brand suitability and on evidence from destination analytics rather than on a thin performance report. Read the breakdown regularly to understand what is happening, and act on it rarely.
That arrangement gives up a control that was worth more when delivery models were weaker, and it keeps the two things placement management is genuinely good for: refusing what the business should not appear in, and making sure the ad works wherever it lands.