First, read the placement report by spend share and name the domains nobody chose
Pull the site and app report for the period and sort it by spend, not impressions. The brand safety and viewability guide describes the pattern that reading by impressions hides: a long list of unfamiliar domains with similar names, each taking a small share, together taking a large one. Add up the share of spend on domains that nobody on the team chose or recognises, then read the outcome columns beside it, engaged visits, time on site, conversions or whatever the plan's measure is. The guide lists the marks of a made-for-advertising siteA 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.: content that is scraped, thin or machine-generated, a very high ratio of ads to content, auto-refreshing and stacked units, traffic bought from clickbait rather than earned, and domains that are new, numerous and quickly replaced. Their visitors are real people who arrived by accident and leave at once, which is why such sites pass invalid-traffic filters and why their outcome columns read close to nothing; the glossary entry on made-for-advertising sites keeps the definition short.
Goodhart's law explains why the money went there. A measure loses its value as a measure once people, or algorithms, are rewarded for moving it, and the brand safety guide applies that directly: an algorithm told to maximise viewable impressions at the lowest price finds made-for-advertising inventory on its own, because it is cheap, highly viewable and plentiful. The goal was met. The plan's purpose was not. So before any control is changed, write down what the line item was told to pursue and compare it with what the business wanted, because the fix for a drifting goal is a better goal, and no inclusion list can substitute for one.
Then put the same budget through the CPM calculator twice: once at the price the open buy cleared at, and once at the price the plan expects after an inclusion list and a verification segment remove most of the open web. The second figure buys fewer impressions, and the comparison the brand safety guide asks for is cost per impression that met the standard, or cost per outcome, rather than raw cost per thousand, which always looks better without controls. If the symptom is a Google Ads display campaign rather than a DSP line item, the diagnosis of display ads on low-quality placements reads that platform's placement report and settings; and if the trigger is one screenshot of an ad beside unsuitable content rather than a report, the tutorial on responding to a brand safety incident is the procedure for that day. This page starts from the spend share and works back through the list, the goal, the verification, the supply path, the targeting and the inherited settings.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Most spend on domains that did not exist when the plan was written, many with similar names | The line item's inventory targeting: open auction with exclusion lists and category settings only, no inclusion channel | Open auction bought with exclusions only, so sites that did not exist when the plan was written are admitted | Build an inclusion list from where the audience reads and assign it as a shared channel |
| Viewability high and cost per thousand low while outcomes on the advertiser's site are near zero | The line item's goal and bid strategy: what it was told to maximise or beat | A goal that rewards cheap viewable impressions, which an algorithm satisfies on made-for-advertising inventory | Move the goal to an outcome on the advertiser's own site and keep viewability as a separate filter |
| The failures were found in a monthly report rather than during the flight | Whether any pre-bid segment is applied and whether a verification tag measures served impressions | No pre-bid verification segment and no post-bid monitoring, so failures are found in the placement report weeks later | Monitor first, then apply pre-bid segments for the categories that appeared, keeping blocking for the floor |
| Trusted publishers appear in the report at prices and viewability that do not match a direct path | The seller path per domain: ads.txt entries, sellers.json and the supply chain object, and delivery by exchange | Long supply paths through resellers that carry fees and obscure the seller | Prefer the shortest path to each publisher and remove exchanges that add hops without inventory |
| The audience line items spend on pages with no relation to the product while contextual siblings do not | Which line items carry audience lists and what context, if any, limits them | Audience targeting that follows people onto any page rather than context that names the page | Add context from the context map, or move prospecting to contextual line items the plan names |
| Settings differ between line items and nobody can say why any was chosen | The advertiser-level defaults and each line item's brand-safety layer against a written tolerance | Brand-safety settings inherited from whoever built the first line item, with no written tolerance | Write the suitability policy, set the floor as advertiser defaults and trace every line item setting to a line of it |
Fix, verify and prevent each cause
Open auction bought with exclusions only
The tutorial on building a programmatic inclusion list draws the distinction that decides this cause. An exclusion list removes named domains, apps or placements from an otherwise open buy, which makes it reactive: a site that is not yet on it is still bought. An inclusion listA list naming the sites and apps a programmatic buy may run on, excluding everything else by default; sometimes called an allowlist. works the other way round, since the DSP only bids where the list allows, so a newly registered made-for-advertising domain is kept out without anyone having to notice it first. The brand safety guide calls it the strongest brand safety control and the largest reach cut, and the tutorial names its cost in the same breath: every domain left off is inventory the buy can no longer reach, so reach falls and the price per thousand rises as the platform competes for a smaller pool. A buy that ran open with sensitive category exclusions only was protected against the domains someone had already seen and against nothing else.
Fix: build the list from the objective rather than from a borrowed file: gather candidates from where the audience reads and from the domains in past placement reports that earned outcomes, check each domain's authorised sellers, judge each against the written suitability policy, score and cut, estimate the reach the list leaves before launch, and build it once as a shared channel assigned to the line items it governs.
Verify: the next placement report, read by spend, shows the budget on listed domains, the outcome per unit of spend rising, and the reach and price close to the estimate written into the plan.
Prevent: keep the register of why each domain is on the list, review it from the placement report by spend on a monthly rhythm, retiring entries that take money without results and adding the next candidates when the list cannot absorb the budget, so the list is reviewed rather than merely grown. Where the list proves too short for the budget and the line item under-delivers, the diagnosis of a DV360 line item that is not spending covers that trade from the other side.
A goal that rewards cheap viewable impressions
The brand safety and viewability guide is explicit that made-for-advertising inventory enters campaigns through the open auction and, more insidiously, through optimisation, and that 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. The guide also explains why viewabilityShare of measured ad impressions that had a real chance to be seen. cannot stand in for suitability: the Media Rating Council defines a viewable impression by a share of pixels in view for a continuous time, with the thresholds in its own guidelines, and Google's Active View is accredited to measure against that standard, but sticky units, auto-refresh and ad stacking all score well on it, so a viewability filter without a suitability filter drifts towards exactly the inventory this page is about. Goodhart's law names the shape: the measure was rewarded and stopped measuring what it was for.
Fix: change what the line item is told to pursue. Give the automated strategy an outcome on the advertiser's own site as its goal, keep viewability as a separate pre-bid filter set from what the objective needs rather than from what is possible, and read cost per outcome rather than cost per thousand when judging the result.
Verify: over the following weeks the share of spend on outcome-producing domains rises, cost per outcome falls even as the cost per thousand rises, and the placement report's long tail shortens without a single domain having been blocked by hand.
Prevent: write every line item's goal into the plan as a decision, pair any efficiency target with a guardrail on outcomes, as the Goodhart page recommends for any target, and treat a viewable-impression or cheap-reach goal as a setting that needs an inclusion list beside it. Where measured viewability is itself the problem, the page on low programmatic viewability diagnoses that symptom separately.
No pre-bid verification segment and no post-bid monitoring
The brand safety guide's table separates the controls that act before the bid from the one that acts after it. Pre-bid segmentsVerification applied before the bid: a vendor's segments, checked by the DSP, decline impressions classed as unsafe, unsuitable, unviewable or invalid. from a verification vendor let the DSP decline impressions the vendor classifies as unsafe, unsuitable at a given tier, unlikely to be viewable or made-for-advertising, for a per-impression fee and some processing time; post-bid monitoring, a verification tag wrapped around the creative in the ad server, measures each served impression independently of the DSP and the exchange and can block the creative from rendering on a failing page. A buy with neither finds its failures in a placement report weeks later, after the budget has gone. The guide's sequence for a new campaign is the one to follow now: 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, because applying every segment at once removes inventory the buyer never saw and makes the reach shortfall impossible to attribute to any one control. The guide to writing a brand suitability policy is where the segments get their instructions: it turns each line of the policy into content labels, category exclusions, channels, viewability targeting and vendor suitability segments, and its review asks of every placement report whether anything ran that breaches the policy, whether the settings removed inventory the policy allows, and whether the world has changed in a way the policy does not yet address.
Fix: deploy the verification tag through the ad server on every creative, read a week of its data by spend, then apply a made-for-advertising segment and the suitability tier the policy names on the line items that showed the failure, and leave blocking on only for the floor.
Verify: the verification report for the next week shows the share of impressions failing the segment's category falling towards zero, and the placement report's spend share on unfamiliar domains falls with it.
Prevent: put the verification read on the weekly rhythm rather than the monthly one, so that a new category of failure is found in days, and record each segment applied beside the line of the policy that asked for it, so the next trader can see which controls were chosen and why.
Long supply paths through resellers
The programmatic ecosystem guide follows the money from an advertiser's budget to a publisher's revenue and finds a platform at every step taking a share: the DSP's platform fee, data, verification and ad serving charged per thousand, the exchange's share, and any reseller between the exchange and the publisher taking another. None of this is published in one place. Three standards make the path visible, and the guide names them: a publisher's ads.txtA file on a publisher's site listing the companies authorised to sell its ad inventory. file lists the companies authorised to sell its inventory and whether each is a direct seller or a reseller, each exchange's sellers.json declares the identities behind its seller identifiers, and the SupplyChain object, carried in the bid request's source field as the OpenRTB specification published by IAB Tech Lab defines it, records every hop an impression took to reach the DSP. A trusted publisher's name in the placement report at a price and a viewability figure that do not match its direct path is the sign of a long route, and a domain sold only through resellers deserves the closer look the inclusion list tutorial gives it before it earns a place.
Fix: read delivery by exchange, which the ecosystem guide notes most DSPs report, compare the paths to the same publishers, prefer the shortest path to each, and remove the exchanges and seller paths that add a hop without adding inventory; set authorised-seller targeting, which the DV360 inventory and deals guide describes as using ads.txt and app-ads.txt to limit buying to declared sellers, to the option the policy chooses knowingly.
Verify: the same publishers appear in the next report at a lower blended price and the same or better viewability, with the supply chain object showing fewer hops, and the share of budget reaching publishers rises.
Prevent: make the quarterly look at where the money went, by exchange, by seller type and by fee category, part of the reconciliation, and record the seller path for every domain on the inclusion list so that a change in how a publisher is sold is caught at the next review rather than in the next placement report.
Audience targeting that follows people onto any page
The guide to planning contextual targeting in DV360 states the difference in one line: contextual targeting places an ad by what is on the page, not by who is reading it. An audience line item does the opposite, following a recognised person onto whatever page they open, which on the open web includes the pages this report is full of; the person may be in-market, and the page may be one nobody would choose. The guide builds the alternative from a context map of what the audience reads, turned into category, channel and keyword line items kept separate so their results can be read, and it warns that layering an audience list onto a contextual line item reaches only recognised people on the right pages, so the two jobs should stay on separate line items. It also notes that brand safety and context interact more than in audience buying, because a contextual line item chooses pages by their content, so exclusions should come from the written suitability decisions rather than a word list that grows with each screenshot.
Fix: give the audience line items an inventory boundary, the inclusion channel at least, and move prospecting that does not need a recognised person onto contextual line items the context map names, with categories for scale, a channel of named publishers for control and keywords only where the subject's vocabulary is distinctive.
Verify: the placement report for the contextual line items shows spend on publishers the map named, and the comparison the guide asks for, each line item judged on outcome per unit of spend over the same weeks, shows which form reaches the right people at a cost the business accepts.
Prevent: keep the context map as part of the plan with an owner and a review date, feed each placement review back into it, and treat an audience line item without an inventory boundary as a build error on the launch check rather than a choice.
Brand-safety settings inherited with no written tolerance
The DV360 inventory and deals guide describes where the controls live: brand-safety settings sit on the line item, with advertiser-level defaults applied when a line item is created, covering digital content labels, sensitive category exclusions, keyword exclusions, viewability targeting and pre-bid segments from the vendors the platform lists, with the Display & Video 360 Help Centre as the documentation of each. Defaults cascade, which cuts both ways: a careful default protects every line item built in a hurry, and a careless one is copied silently into all of them. The suitability policy guide describes what happens when nobody writes the judgement down: the settings are chosen by whoever builds the first line item, copied to the next, loosened when a deal will not spend and tightened when someone forwards a screenshot, until after a year they reflect a series of moments rather than a decision, and nobody can say whether a given placement was inside the brand's tolerance or not. The programmatic advertising audit treats that as a gating weakness, scoring the settings against the brand's written tolerance and scoring a missing tolerance as missing however careful the settings look.
Fix: write the policy with the people who own the brand: the safety floor on for every buy and not open to exception, the stance on each sensitive topic, on news and on user-generated pages, who may approve an exception and for how long; then set the floor as advertiser defaults, trace every line item's settings to a line of the policy, and log the exceptions.
Verify: every line item's brand-safety layer can be read against the policy's table, the next placement report shows nothing that breaches it and nothing the policy allows removed by a setting stricter than it, and the exception log explains each deliberate difference.
Prevent: review the policy against what actually ran on a regular rhythm, re-read it whenever the product range, the audience or the agency changes, and put the policy under mandatories in the media brief so the next plan is written inside it rather than discovering it at launch.
Prevention: an inclusion list, a written tolerance and a monthly placement read by spend
Every cause above is a door that was open before the flight, and the fundamentals hub shows where each control acts in the journey of an impression: the inclusion list and the suitability settings before the bid, verification before and after it, the goal in the valuation, the seller checks against the request. The discipline the brand safety guide sets 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.
Three documents carry that discipline between flights. The inclusion list's register says why every domain is there, with its seller path and score, and is versioned at each monthly review. The suitability policy says what the brand will stand beside, who decides and who approves exceptions, and is the line every setting is traced to. The plan records each line item's goal, its filters and the reach and price it was expected to buy under them, so the next report can be read against an expectation rather than against a feeling. The CPM calculator belongs with the third: run the budget at the price the controlled inventory clears at before launch, so the smaller pool and the higher price are chosen rather than discovered, and compare the result on cost per outcome.
Then make the placement read a rhythm. Monthly, sort the site and app report by spend, name the share on domains the list did not choose, read the outcome columns beside it, and feed the result into the register and the policy: retire domains that take money without results, add the next candidates when the list cannot absorb the budget, apply a segment for any new category of failure and record an exception for any loosening. Weekly, read the verification data by spend, so a new kind of failure is caught in days. Quarterly, read where the money went by exchange, seller type and fee, and when several of these habits are missing at once, the programmatic advertising audit rates safety, deals, frequency, bidding, creatives and reconciliation in one pass and is the better place to start than a single diagnosis. Where the right answer is to buy a named publisher's inventory directly, the deal types guide explains what each kind of deal commits the publisher to and what the certainty costs; a deal narrows the inventory, and the same controls still decide which of the publisher's pages are acceptable.
Finally, keep the goal honest. A line item that is told to find cheap viewable impressions will find them, and no list or segment will stop it trying; a line item that is told to find people who start a quote on the advertiser's site has no reason to visit a page where nobody does. The programmatic troubleshooting hub lists the neighbouring symptoms, from a line item that will not spend once the list is in place to viewability that falls once the auto-refreshing units are gone, that a buy can show after this one is fixed.
A boiler cover provider finds where its display budget went (illustrative)
Kelvingarth Boiler Cover
Kelvingarth Boiler Cover, an invented boiler breakdown cover provider, reads its monthly site and app report by spend for the first time and finds that £10,700 of an £18,500 display budget, about 58 per cent, sat on 74 domains with near-identical names, auto-refreshing units and an engaged-visit rate below 2 per cent, against a rate above 9 per cent on the news and home-improvement publishers the team recognised. The prospecting line item ran on the open auction with sensitive-category exclusions only, optimised to a viewable-impression goal, and carried brand-safety defaults copied from the first line item an earlier agency had built; nobody could produce a written tolerance.
Over three weeks the marketing manager and the trader close four doors. They build an inclusion list of 260 domains from where policyholders read, home, money, local news and trade titles, plus the 38 domains in the report that had produced quote starts, check each domain's sellers and remove 19 sold only through resellers, and assign the list as one shared channel. They move the line item's goal to quote starts on the provider's own site and keep a viewability filter at the level the editorial placements had achieved. They deploy the verification tag, read a week of its data and apply a made-for-advertising segment for the category that appeared. And they write a one-page suitability policy, set its floor as advertiser defaults and log one exception for a regional news deal that needs unlabelled content allowed.
The next month the same £18,500 buys 4.7 million impressions at a £3.90 CPM instead of 8.8 million at £2.10, the share of spend on domains the list did not choose falls below 3 per cent, and quote starts from display rise from 1,900 to 3,100, a cost per quote start of £5.95 against £9.70. The placement read by spend is now a monthly line on the trader's calendar, with the register versioned at each review.
Common mistakes
- Reading the placement report by impressions. The long tail of unfamiliar domains hides where the money went; sort by spend share.
- Blocking this month's domains and calling it fixed. An exclusion list is reactive; the tail is rebuilt from new domains next month.
- Treating viewability as a quality signal. Made-for-advertising inventory is built to be viewable; judge suitability and outcomes separately.
- Leaving the goal at cheap viewable impressions. The algorithm will find the inventory that scores best on it, whatever the list says.
- Applying every verification segment on day one. Reach falls, price rises and nobody can say which control did it; monitor first.
- Trusting a publisher's name in the report. Check the seller path; a trusted title sold through three resellers is not the impression the plan meant to buy.
- Letting audience line items roam without an inventory boundary. They follow the person onto any page the open auction offers.
- Inheriting settings without a written tolerance. Nobody can say whether a placement was inside the brand's tolerance, because the tolerance was never decided.
