Some of Those Impressions Were Never Seen by a Person
A case study lands on your desk claiming several million impressions. Before deciding whether that is impressive, settle a smaller question: how many involved a human being. Ad platforms do filter automated activity out of some numbers, and document it. What they do not promise is that every figure an agency can export into a slide went through that filter.
An impression is a count from an ad server, not a person
The word carries more authority than the measurement deserves. An impression is generally logged the moment an ad is served or a creative is requested. Nothing in that confirms a person was in front of the screen, which is why viewability is a separate standard.
That gap is where invalid traffic gets in. Bots, crawlers, scripted tools and data centre traffic all generate requests that look, at the log level, like any other, and the ad server records them because recording requests is what it does. Filtering it out happens afterwards, in the platform’s detection systems, and applies to some outputs rather than all. An unqualified impression figure has an undocumented relationship to that step, the problem in a case study number is only as good as what backs it up.
Platforms say they filter this, just not always from the number you’re looking at
Google Ads publishes a help page called “About invalid traffic”. It defines invalid traffic as clicks and impressions on ads that are not the result of genuine user interest, listing automated tools, bots and crawlers, known invalid data centre traffic, and impressions meant to artificially lower an advertiser’s clickthrough rate among the types. Advertisers are not charged for invalid clicks or impressions. Detected before invoicing, charges are adjusted. Detected afterwards, credits are issued instead of a refund.
Notice what that guarantees: billing. Google also documents an “Invalid clicks” column and an invalid activity credit report so advertisers can see the adjustment. None of it says “every number this platform shows you has had bots removed.” A billed impression count and a dashboard reach or impression figure are different metrics, and a disclosure covering one does not travel to the other. An agency quoting a headline impression number is usually quoting the second.
What gets filtered, and what a case study rarely says
| Metric | What the cited platform page states |
|---|---|
| Billed Google Ads impressions and clicks | Advertisers are not charged for invalid clicks or impressions; pre-invoice adjustments or post-invoice credits. |
| Dashboard impression or reach in an ad report | Not confirmed from the cited sources: the page documents billing adjustments and an invalid clicks column, not that every reported metric is post-filter. |
| YouTube public view, like and subscription counts | Algorithmically confirmed to come from actual humans rather than computer programs; counts may be slowed, frozen or changed and low quality playbacks discarded. |
| Paid video views counted as TrueView | Gives the thresholds that make a paid view billable and separates those from non-billable public views. Does not describe invalid traffic filtering for it. |
| Non-Google platform impressions or reach | Not covered by either source checked here. |
A hypothetical, so the shape of the problem is visible
The following is invented: neither the brand nor the numbers are real, and they exist only to make the mechanism legible. Picture a hypothetical outdoor gear brand, Kessloe Outdoors, running a display campaign. The ad server logs 4,200,000 impressions over the flight. After invalid traffic detection, 3,780,000 are treated as valid for billing. Both describe the same campaign, and a case study leading with 4.2 million reads differently than one leading with 3.78 million.
That gap is arbitrary, picked to be visible rather than typical. The real filtered share is not something these sources state and it varies by platform and campaign, so nothing here should travel as a rate. The point is that one campaign can honestly produce two impression totals, and the document quoting one rarely says which.
Questions worth asking before trusting the number
- Does the metric name match one the platform documents as filtered? “Billed impressions”, “impressions” and “reach” are not synonyms. If the case study says “impressions delivered” while the platform’s filtering language is about what you were charged for, the disclosure does not carry.
- Is there a methodology note on that specific figure? A line naming the platform, the report, the column, and whether invalid traffic was excluded. “4.2M impressions, Google Ads campaign report, exported 14 March” is checkable. “4.2M impressions” is not, the same failure as when the headline number is a vanity metric.
- Is the number called served, billed, or viewable? Served is rawest, billed carries the filtering disclosure, viewable is a claim about the viewport. A document using none of them has left the guess to you.
Where this stops being about the number
This is about reading a reported figure in a published document, not about auditing an agency’s ad account, billing or contract. Everything above needs only the case study and the platform’s public documentation.
It is also not double counting. Invalid traffic is non-human activity inflating a single count from a single source. Cross platform double counting is a real human view counted more than once, the same person meeting the campaign in several placements. Different mechanisms, different questions. A figure summed across platforms may carry both, which is when it helps to know how to read a results chart before you read the caption.
Read a few and the pattern gets obvious
Metric labels are easier to judge in bulk. Browse the case study library and read only how each figure is named, ignoring its size. If your published work says where its numbers came from, submit your own.
FAQ
Does Google filter bot traffic before charging advertisers?
Google Ads Help documents that it does. Its “About invalid traffic” page defines invalid traffic as clicks and impressions that do not come from genuine user interest, counts automated tools, bots and crawlers among the types, and states advertisers are not charged for it. The page gives no rate, so infer none.
If a platform filters invalid traffic from billing, is the case study number already clean?
Not necessarily. The guarantee attaches to what an advertiser is billed for. A case study normally quotes a dashboard impression, reach or view figure, a metric the same page does not claim gets identical treatment. Before assuming a figure was filtered, check whether the metric it names appears in the platform’s filtering documentation.
Sources
- Google Ads Help, “About invalid traffic”. Fetched 7 September 2026.
- YouTube Help, “How engagement metrics are counted”. Fetched 7 September 2026.
