Served, measurable, viewable: the counts of one impression

Six counts of the same campaign

Each count is a subset of the one above it. An impression drops out at the first stage it fails and never comes back:

Count	Of 10,000	What it means	Who reports it
Billed	10,000	The DSP won the auction and charged for it.	Your DSP, your invoice
Served	9,400	The exchange delivered the ad to the page.	The exchange or ad server
Rendered	8,460	The ad loaded and the verification code inside it ran.	Your verification vendor
Measurable	7,073	The code could tell where the ad sat on the screen.	Your verification vendor
Viewable	5,305	It met the MRC threshold: enough of it, on screen, for long enough.	Your verification vendor
Valid	5,146	Viewable, and not to a known bot or automated browser.	Your verification vendor

The gap between the first two lines and the third is what most people call the impression discrepancy. The losses below it are inside your verification report, so they never show up as a discrepancy at all — which is why they are so easy to miss.

Served impressions

An impression is served when the exchange or ad server sends the ad to the page. It is the count a publisher is paid on and the one closest to what a DSP bills. It says nothing about whether the ad loaded, where it landed, or whether anyone could see it.

Measurable impressions

A measurable impression is one where the verification code could work out the ad's position and size on screen. The usual reason it cannot is a cross-domain iframe: the ad is boxed inside a frame from another site, and the browser, rightly, will not tell code inside the box where the box is. Newer browser APIs can report visibility across that boundary, but not every page, browser or ad server lets them.

Unmeasured is not the same as unseen. An impression that could not be measured may have been perfectly visible; nobody can say. It is reported as unmeasured, not guessed at — a modelled number is indistinguishable from a real one the moment it enters a spreadsheet.

Viewable impressions

The Media Rating Council's standard for display ads: at least 50% of the ad's pixels in view for one continuous second. For large ads — 242,500 pixels or more, such as a 970×250 billboard — the threshold drops to 30% of the pixels. Video uses two continuous seconds.

Viewable is a subset of measurable. Only an impression whose position could be measured can be judged viewable or not.

The trap in the viewability rate

Viewability is almost always reported as a share of measurable impressions, not of billed ones:

Rate	Formula	This campaign
Viewability rate (as usually reported)	viewable ÷ measurable	75.0%
Measurable rate	measurable ÷ rendered	83.6%
Viewable share of what you paid for	viewable ÷ billed	53.1%

All three are true. The first is the one in the dashboard. The last is the one that matches the invoice. A campaign can report 75% viewability while barely half of what was bought could ever have been seen; the difference is everything that was never served, never rendered, or never measurable. Always ask what the denominator is.

Valid impressions

The last cut removes invalid traffic: impressions shown to software rather than people. The industry splits it in two. General invalid traffic (GIVT) is the kind a list or a rule can catch — declared crawlers, known bots, headless and automated browsers, pages that were only prerendered. Sophisticated invalid traffic (SIVT) is built to look human and needs behavioural and network analysis. Check which kind a report removes; many only remove the first. The glossary defines both.

Which count should you pay on?

That is a negotiation, not a fact — but you cannot negotiate over a number you cannot see. The minimum is to know all six for every campaign, so a gap is a decision rather than a surprise. Pixtru reports each billed impression at the stage it stopped, with the reason, and the stages add back up to the invoice; the method describes each gate.