What can go wrong on the open web
The open auction offers a buyer a very large number of sites and apps, most of which the buyer has never heard of and none of which it has visited. Several things go wrong at that scale. An ad appears beside content that is illegal, hateful, violent or simply at odds with the brand: a holiday advert on a page about an air crash, a children's product beside adult material. A domain is spoofed, so the bid request claims a premium publisher and the impression serves elsewhere. A publisher's ads.txt fileA file on a publisher's site listing the companies authorised to sell its ad inventory. exists so that buyers can catch this kind of misrepresentation. Traffic is not human: bots load pages and ads at volume, or a hidden frame stacks several ads where one is visible. An ad is served into a slot far below the fold that the reader never reaches, or into an auto-refreshing unit that counts a new impression every few seconds while the reader looks elsewhere. A site is built with the sole purpose of selling ad impressions, with content scraped or generated to carry them and visitors bought through cheap clickbait, so that an advertiser's budget buys attention from people who arrived by accident and leave at once.
None of this can be caught by a person reviewing placements, because the placement list for a modest campaign runs to thousands of domains and changes daily. The controls have to be rules the DSP applies before it bids, verification the platform applies before and after the serve, and a review discipline that reads the placement report by where the money went rather than by what looks odd. The fundamentals hub shows where in the journey of an impression each control acts; this guide takes them in turn.
Brand safety controls
The controls fall into three groups: where the ad may run, what the impression must satisfy, and who checks. The table lists them with where each lives and what each costs.
| Control | Where it lives | Acts before or after the bid | What it does | Trade-off |
|---|---|---|---|---|
| Inclusion list (allowlist) | DSP line item or advertiser settings | Before | Permits only the named domains, apps or channels; everything else is excluded by default | The strongest control and the largest reach cut; needs maintenance to add new sites |
| Exclusion list (blocklist) | DSP, often shared across advertisers | Before | Removes named domains, apps or placements from an otherwise open buy | Reactive; a site not yet on the list is still bought |
| Sensitive category exclusions | DSP, using the platform's own classification | Before | Excludes content classified into categories such as tragedy, weapons or adult content | Classification is the platform's; the categories are broad and can remove news wholesale |
| Content labels and ratings | DSP; in DV360 the documented digital content labels | Before | Limits delivery to content rated for a general or mature audience | Unrated content is treated according to a setting the buyer must choose deliberately |
| Keyword exclusions | DSP or verification vendor | Before | Blocks pages whose text contains listed words | Over-blocks heavily: the same word appears in safe and unsafe contexts |
| Deals with chosen publishers | SSP and DSP, by deal ID | Before | Narrows inventory to a negotiated set of properties | Price premium; the publisher's own pages still vary in suitability |
| Authorised-seller checks | DSP, using ads.txt and sellers.json | Before | Bids only on inventory offered by sellers the publisher has authorised | Removes spoofed inventory; does nothing about the authorised publisher's content |
| Pre-bid verification segments | Third-party vendor integrated into the DSP | Before | Bids only on impressions the vendor classifies as safe, suitable, viewable or not made-for-advertising | Per-impression fee; adds processing time to the bid; coverage varies by inventory |
| Post-bid blocking and monitoring | Verification tag wrapped around the creative in the ad server | After | Measures each served impression and can block the creative from rendering on a failing page | A blocked impression may still have been paid for; monitoring without blocking only reports |
Google's DSP documents each of the first-party controls in the table in its help centre, and the DV360 inventory and deals guide explains where each is set in the hierarchy and how settings at the advertiser level cascade to line items. Google Ads display offers a simpler version of the same controls, which the display targeting options guide covers; a buyer moving from Google Ads to a DSP should expect the same ideas with finer grain and more responsibility.
Suitability vs safety
Brand safety and brand suitability are different questions, and conflating them produces both over-blocking and embarrassment. Safety is the floor: content that no advertiser should appear beside, such as illegal material, terrorism, explicit content and incitement. The industry's shared vocabulary for this came from the brand safety floor and suitability framework that the World Federation of Advertisers' GARM initiative published; the initiative has since closed, but verification vendors and platforms still use its categories and risk tiers as a common language. The floor is not negotiable and the controls for it should be on by default for every advertiser.
Suitability is the brand's own judgement about everything above the floor. A news publisher's coverage of a war is safe in the industry sense and unsuitable for some brands, essential context for others and neutral for most. Alcohol, gambling, dating, political debate, crude humour and tragedy are each acceptable to some advertisers and not to others. Suitability is therefore set per brand, usually in tiers of risk tolerance, and it should be written down with the people who own the brand rather than inferred by whoever sets up the line item. The guide to writing a brand suitability policy sets out what that document holds, from the floor and the topic stances to exceptions and the review.
The commonest suitability mistake is to treat news as unsafe. Keyword exclusions and broad category settings remove a large share of quality journalism from a plan because it reports on difficult subjects, which pushes budget towards the low-quality inventory that reports on nothing. A better approach is to use the verification vendor's suitability tiers at page level, to keep an inclusion listA list naming the sites and apps a programmatic buy may run on, excluding everything else by default; sometimes called an allowlist. of trusted news publishers regardless of topic, and to accept that a brand will occasionally appear beside hard news on a reputable site. The reputational risk of that is small; the reputational risk of funding content farms is not, and it is less visible. The academy exercise on handling a brand safety incident for a car seat maker works through a complaint about an ad beside a news report, from the evidence to the narrowest stop.
Viewability as a buying filter
An impression that is served but never on screen cannot do anything for the brand. The Media Rating Council defines a viewable impression by a documented share of the ad's pixels in view for a documented continuous time, with a longer time for video and a different pixel share for very large formats; the exact thresholds are in the MRC's guidelines rather than repeated here. Google's Active View measures against that standard and is accredited to do so, and the third-party verification vendors offer their own accredited measurement.
As a metric, viewability is the subject of the viewability and attention guide. As a buying filter, it works in two ways. Pre-bidVerification applied before the bid: a vendor's segments, checked by the DSP, decline impressions classed as unsafe, unsuitable, unviewable or invalid., the DSP or a verification vendor predicts the viewability of an impression from the placement's history and bids only where the prediction clears a threshold the buyer sets; DV360 documents this as viewability targeting using Active View's predicted rate. Alternatively the buyer optimises towards a viewable cost per thousand, paying for impressions in the ordinary way but letting the DSP favour placements that are measured as viewable.
Both have costs. A viewability threshold removes the inventory that falls below it, and the inventory immediately below a strict threshold is often ordinary editorial placements rather than junk. The prediction is made from history, so new placements and new publishers are under-bought until they have one. And viewability can be gamed: sticky units, auto-refresh and ad stacking all score well, which is why a viewability filter without a suitability filter drifts towards exactly the inventory the next section describes. Set the threshold from what the objective needs, not from what is possible, and check the placement report for what the filter kept as well as what it removed. If a running campaign's viewable rate has already fallen below target, the troubleshooting page for low programmatic viewability works out whether the slot, the bid goal, the creative or the measurement base is the cause before the threshold is touched.
Verification and pre-bid
Verification vendors sit outside both sides of the trade and measure what happened. Their tags, wrapped around the creative in the advertiser's ad server, record whether each served impression was viewable, on what page and in what context, whether the traffic looked human, and whether the domain matched what the bid request claimed. That record is independent of the DSP and the exchange, which is its main value: it can be shown to a client, an auditor or a publisher as the common ground for a dispute.
Pre-bid is the same intelligence applied earlier. The vendor classifies pages and apps continuously and makes the classification available to the DSP as segments: safe against the industry floor, suitable at a given risk tier, predicted viewable above a threshold, not made-for-advertisingA site built to sell ad impressions rather than serve readers: thin or scraped content, a high ratio of ads to content and traffic bought rather than earned., not invalid trafficAd impressions and clicks not caused by genuine human interest: bots, fraud and accidents.. The DSP checks the segment before bidding and declines impressions that fail. DV360 documents integrations with the major vendors for pre-bid targeting, and other DSPs have equivalents. The segment is charged per thousand impressions, which means a buyer pays the fee on every impression it bids on through the segment, won or not, on some pricing models, and should check the vendor's terms.
Post-bid blocking is the last line. When the verification tag detects a failing page at serve time it can prevent the creative from rendering, so the brand does not appear there; depending on the exchange and the contract the impression may still be billed, so blocking is a reputational control rather than a financial one. Monitoring without blocking reports the same failures and lets the buyer act on them in the next inclusion list revision.
A sensible sequence for a new campaign is to monitor first, read a week of verification data to learn where the plan's inventory actually fails, then apply pre-bid segments for the failures that matter, and keep post-bid blocking for the safety floor only. Applying every segment at once on day one removes inventory the buyer never saw and makes the reach shortfall impossible to attribute to any one control.
Made-for-advertising inventory
Made-for-advertising sites are built to sell impressions rather than to serve readers. Their characteristics are consistent: content that is scraped, thin or machine-generated; a very high ratio of ads to content; auto-refreshing and stacked ad units; traffic bought from clickbait and social placements rather than earned; and domains that are new, numerous and quickly replaced. They are not fraudulent in the sense of fake traffic, which is why invalid traffic filters do not catch them. The visitors are real people who arrived by accident, see a wall of ads and leave.
They get into campaigns through the open auction and, more insidiously, through optimisation. An algorithm told to maximise viewable impressions at the lowest price will find made-for-advertising inventory on its own, because it is cheap, highly viewable and plentiful. Placement reports that rank domains by impressions show the pattern: a long list of unfamiliar domains with similar names, each taking a small share, together taking a large one.
The platform-level defence is an inclusion list, which excludes these sites by never admitting them; the tutorial on building a programmatic inclusion list builds one from the audience's reading, checks each domain's sellers and sizes the reach it leaves. The vendor-level defence is a made-for-advertising segment, which the major verification vendors now classify and offer pre-bid; Google's DSP does not document a single native switch for the category, so in practice the controls are inclusion lists, vendor segments, exclusion lists built from the placement report and a preference for shorter supply paths. The buying-level defence is to stop optimising towards cheap viewable impressions alone. An objective that includes an outcome on the advertiser's site, engaged visits, time on page or conversions, does not reward made-for-advertising inventory, because its visitors never produce one. When a placement report already shows most of the spend on such domains, the diagnosis of programmatic spend going to low-quality sites works back from the spend share through the list, the goal, the verification, the supply path and the inherited settings.
The reach and cost trade-off
Every control described above removes inventory, and the arithmetic compounds. An inclusion list removes most of the open web; a suitability tier removes a share of what remains; a viewability threshold removes some of that; a pre-bid segment removes more. Each removal lowers the number of people the campaign can reach (ReachUnique people who saw an ad.), raises the price per thousand as the DSP competes for a smaller pool, and lowers the win rate as more bids go unmade. A plan that stacks every control will reach fewer people at a higher price than it forecast, and the shortfall will be blamed on the market. When a running buy's price comes in above the plan's, the diagnosis of a programmatic CPM higher than planned reads stacked controls beside five other causes before anyone blames the market.
The right question is not how clean the inventory can be but what the objective needs. A reach campaign for a mass-market brand can tolerate a broad inclusion list, a moderate suitability tier and a viewability threshold set at the level the format usually achieves. A campaign for a brand with real reputational exposure, or one measured on outcomes rather than impressions, should accept the smaller pool and the higher price because the removed inventory would not have produced the outcome anyway. In both cases the comparison to make is cost per impression that met the standard, or cost per outcome, rather than raw cost per thousand, which always looks better without controls.
Tightening a reach campaign's inventory
A home-insurance brand runs an open-auction display campaign for a month with only sensitive category exclusions. It spends £29,000 on 10 million impressions at a £2.90 CPM across 3,800 domains; measured viewability is 52 per cent, and the placement report shows 41 per cent of impressions on sixty domains with auto-refreshing units and a bounce rate above 90 per cent in analytics. Engaged visits, defined as a session with a quote started, number 2,100, a cost of £13.80 each. For the second month the team applies an inclusion list of 400 publishers, a pre-bid made-for-advertising exclusion and a viewability threshold at the level the first month's editorial placements achieved. Spend stays at £29,000 but buys 6.5 million impressions at a £4.46 CPM; viewability rises to 71 per cent, reach falls by about a third, and engaged visits rise to 3,400, a cost of £8.53 each. The CPM report looks worse and the business result is better, which is the usual shape of the trade.
The discipline is to decide the controls before launch from the objective, record the reach and price the plan expects under them, and review the placement report monthly by spend share. Controls added in reaction to a screenshot tend to be too broad; controls designed from the objective tend to be exactly as broad as they need to be. When a running buy is reviewed, the programmatic advertising audit scorecard scores these controls against the brand's written tolerance, beside Floodlight, deals, frequency and reconciliation. Deals with chosen publishers, covered in the deal types guide, are the other way to narrow inventory, and the same trade-off applies to them: certainty about the environment in exchange for a smaller pool at a higher price. The DSP's own role in all this is to apply the rules it is given; the supply-side platform (SSPSoftware publishers use to sell inventory.) and the publisher behind it decide what is offered in the first place, and a buyer who wants better inventory than the open auction shows should be talking to them.
To see what viewability does to delivery, enter an all-in viewable CPM in the impression calculator set up for programmatic deals and read the result as viewable impressions.
Common mistakes
- Blocking news by keyword. It removes quality journalism and leaves the content that reports on nothing; use suitability tiers and an inclusion list of trusted publishers instead.
- Treating viewability as a safety control. The most viewable inventory on the open web includes the least suitable; filter both, separately.
- Optimising to cheap viewable impressions. The algorithm finds made-for-advertising sites on its own; give it an outcome on the advertiser's site to optimise towards.
- Stacking every control on day one. Reach falls, price rises and nobody can say which control did it; monitor first, then filter for the failures that appeared.
- Reading the placement report by impressions rather than spend. Long tails of unfamiliar domains hide where the money went.
- Assuming a deal with a trusted publisher needs no controls. It narrows the inventory; it does not verify the page.
- Reporting raw CPM after adding controls. The right comparison is cost per impression that met the standard, or cost per outcome.
