Pre bid brand safety: what the filter sees before it bids
Part of Brand suitability
The short answer
Pre bid brand safety is a filter inside a buying platform that decides, before a bid is placed, whether an impression's page or app fits the advertiser's rules. It reads the bid request, mainly the page URL, domain and category fields, and looks up a stored rating of that page. It is fast, but only as good as that data.
Key takeaways
- A pre bid filter decides inside the auction's time limit, using the bid request and classifications made earlier, not a fresh reading of the page.
- The fields a filter reads are the page URL, the domain or app bundle, the category arrays with their taxonomy version, and the content language.
- Filters miss unsafe pages when the URL is missing or generic, the classification is stale, the risk is in images or video, or the page is new.
- The setting for unknown or unclassified pages can remove reach on its own, so measure it first.
- Each pre bid setting should trace to one line of the written policy, with a note on what the setting cannot see.
What is pre bid brand safety?
Definition
- Pre bid brand safety
- A rule applied by a buying platform to each bid request before it bids, which skips impressions whose page, app or content fails the advertiser's safety or suitability settings. Nothing is bought, so nothing is paid for the skipped impression.
The idea is simple: avoid a bad placement instead of paying for it and blocking it afterward. The IAB's December 2020 guide (opens in a new tab) calls determining suitability before matching a brand to an impression opportunity the ideal outcome for all parties, and pre bid avoidance the first step to achieving it. The same guide warns that pre bid avoidance relies on the accuracy of bid stream data that may be unclear, incomplete or inaccurate.
A pre bid filter need not be built by the buying platform alone. IAB Tech Lab's implementation guide for Content Taxonomy 2.2 (opens in a new tab) describes the design it expected. The platform gives buyers a screen to set a risk tolerance for each of the 11 brand safety categories grouped under Sensitive Topics, passes the content URL it receives from the publisher or exchange to a classification service, and uses the response to decide whether the content matches that tolerance. Those calls run through proprietary connections, not through a published standard, so each platform's details differ. The ad verification guide covers what happens after the bid.
What data does a pre bid filter use to decide?
The bid request is the filter's only fresh input. In OpenRTB 2.6 (opens in a new tab), the fields below describe where the impression will appear. A filter combines them with classifications stored earlier, keyed by URL or domain. These fields are filled in by the publisher or the exchange, and none of the fields in the table is required by the specification.
| Field | Specification wording | What can go wrong |
|---|---|---|
| site.page | URL of the page where the impression will be shown | Missing, truncated, or a generic URL shared by several articles |
| site.domain | Domain of the site | Only the domain is known, so the whole site gets one rating |
| site.cat, sectioncat, pagecat | IAB Tech Lab content categories of the site, section and current page | Self declared by the seller; site level labels say little about one page |
| cattax | The taxonomy in use; if absent, Content Taxonomy 1.0 is assumed | Old default; AdCOM lists 1.0 and 2.0 as deprecated and 3.1 as value 9 |
| content.url | URL of the content, for buy side contextualization or review | Optional, so it may be absent on video and apps |
| content.language | Content language using ISO 639-1 alpha 2 | Missing or wrong for bilingual sites |
| app.bundle and app.cat | The app's identity and its categories | An app has no page URL, so one rating covers all its screens |
| tmax | Maximum time in milliseconds the exchange allows for bids to be received | The lookup must finish inside this limit or the bid is lost |
The taxonomy code matters more than it looks. The AdCOM 1.0 list of category taxonomies (opens in a new tab) gives value 7 for Content Taxonomy 3.0 and 9 for 3.1, and recommends against 1.0 and 2.0 because they lack SCD flags. A request that sends category IDs without cattax is read as Taxonomy 1.0, so a filter can misread perfectly good labels.
Why can pre bid filters miss unsafe pages?
A pre bid filter judges a page it has not just read. It relies on what the request says and on a classification made earlier, and each of those can be wrong in a predictable way. The MRC's 2018 supplement (opens in a new tab) names several of them, and notes that in certain pre bid situations the buying platform's own data can have a material impact on context and brand safety determinations.
| Blind spot | Why it happens | What a buyer sees |
|---|---|---|
| Stale classification | The page changed after it was rated; the MRC requires stated refresh and time to live policies for this reason | A news page rated before a story broke still passes |
| No page URL | The request carries only a domain or an app bundle | One rating for a whole site, safe and unsafe pages alike |
| New page | The page has not been classified | It is treated as unknown, then blocked or allowed by default |
| Images, video and audio | Classification read only text or keywords | A graphic image on a page with mild text passes |
| Self declared categories | The seller fills in cat and pagecat itself | Labels that describe the site's hopes, not the page |
| Sampled decisions | Some pre bid approaches rely on samples, which the MRC says must be built to 99.7 percent confidence with errors disclosed | Confidence varies by site; ask whether it is shown to the buyer |
How do pre bid filters affect reach?
Each rule can remove impressions before the auction, and the removals do not show up in delivery reports, because nothing was bought. Reach loss can come from three places: categories set stricter than the brand needs, unknown pages blocked by default, and whole domains rated on their worst pages because no page URL was sent. The third affects news and small publishers whose requests carry less detail.
The trade is real in both directions. The IAB guide puts it plainly: broad specifications allow scale but may allow non ideal contexts, while narrow ones may limit scale but also limit delivery outside the targeted context. Google says the same of its own account level control: Limited inventory is intended to exclude as much sensitive content as possible and may limit your reach and performance. Neither statement gives a size for the effect, so test it on the campaign at hand.
Four steps to measure what a pre bid setup removes
- Ask for avoided request counts. Request the number of bid requests skipped by each rule, split by category, unknown and missing data, for one week.
- Run a short split. Launch two line items with identical targeting, one with the pre bid setup and one with only the safety floor, and compare eligible requests.
- Sample the skipped pages. Open a sample of URLs the stricter line item avoided and record how many the brand would actually have rejected.
- Adjust one rule at a time. Change the unknown setting or one category level, then measure again, so the effect of each change is visible.
How should pre bid settings match a written policy?
A pre bid screen is a translation of the policy, and translations can drift. Write the mapping down so a reviewer can follow any setting back to the sentence that justifies it, and so a change to the policy shows which settings need to move. The brand suitability profile guide covers writing the policy itself.
Policy to pre bid mapping check
- Each category setting names the policy line it implements and the taxonomy version the filter uses.
- The unknown and unclassified setting is written in the policy as a decision, with an owner.
- Each setting notes what it cannot see, such as images, app screens or requests with no page URL.
- Campaign specific avoid topics are applied as separate rules and expire with the campaign.
- The same tolerance is set on every buying platform in the plan, with differences recorded.
- Approved inclusion lists and deals are checked against the filter so it does not quietly remove approved sites.
- A dated record shows who changed each setting and why.
What does a pre bid review look like on a real brief?
Fictional example
A rural internet provider in Wyoming finds its missing reach
A fictional provider of home internet for rural households in Wyoming runs a display campaign on local news, ranching and community sites. Delivery falls well short of the plan. The planner asks the buying platform for avoided request counts by rule.
The largest share of avoided requests comes from the unknown setting, not from any category. Several small local news sites send only their domain, with no page URL, and their pages have not been classified, so the filter treats them as unknown and skips them. The client's written policy accepts local news and says nothing about unknown pages; the strict default was not a decision anyone made. The planner proposes allowing unknown pages on the approved list of local sites only, keeping the unknown block for open exchange supply, and adding post bid monitoring on those sites. The marketing lead approves the change in writing, and the mapping document records the new line.
How does PlacementGraph prepare the list a pre bid filter works with?
Frequently asked questions
Does pre bid filtering replace post bid monitoring?
No. The IAB's 2020 guide says that because bid stream data can be unclear, incomplete or inaccurate, buyers should use blocking and monitoring after delivery to account for shortcomings pre bid. It describes mapping unsuitable URLs found post bid to their seller and source IDs, then blocking those sources pre bid in future.
Do private deals still need pre bid brand safety?
A lighter version can do. A deal limited to approved domains has already done much of the filtering, but the IAB guide notes that a seller in a direct deal may not have every control needed to match a brand's settings. Keep the safety floor and the unknown setting, and review whether category rules duplicate the list.
Can a publisher send its own suitability rating in the bid request?
Not per category in standard OpenRTB, according to IAB Tech Lab's 2020 guide, which said a general suitability level could be passed and an extension for per category ratings was being worked on. Publishers can send category IDs in cat and pagecat, which buyers may verify independently.
Why does the same page pass on one platform and fail on another?
Each platform connects to its own classification service, stores ratings for different lengths of time, and may set unknown pages differently. The settings screens can look alike while the underlying data differs. Compare the taxonomy version, refresh policy and unknown setting before comparing results.
Sources
- OpenRTB 2.6 specification (opens in a new tab)IAB Tech Lab
- AdCOM 1.0, List: Category Taxonomies (opens in a new tab)IAB Tech Lab
- Implementation Guide for Brand Suitability with IAB Tech Lab Content Taxonomy 2.2 (opens in a new tab)IAB Tech Lab
- Enhanced Content Level Context and Brand Safety, Ad Verification Supplement (September 2018) (opens in a new tab)Media Rating Council
- Brand Safety and Suitability Guide (December 2020) (opens in a new tab)IAB
- About content suitability (opens in a new tab)Google Ads Help
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.