What contextual targeting can do in DV360 that audience lists increasingly cannot
Every bid request carries context. The OpenRTB specification published by IAB Tech Lab defines a site or app object in the request with the domain, the page, content categories and the publisher, and the real-time bidding guide lists those fields and what a bidder does with each, from inventory lists to contextual targeting and brand-safety checks. Recognising the person is a different matter: it needs an identifier the platform has seen before. The programmatic ecosystem guide warns that a large share of impressions arrive with none and tells planners to plan contextual targeting for them rather than discover a reach shortfall later.
Several forces push the same way. Browsers restrict third-party cookiesCookies set by a domain other than the site visited, used to recognise a browser across sites., mobile identifiers depend on permission, and the fundamentals hub notes that without consent for personalised advertising a DSPSoftware advertisers use to buy programmatic inventory. may still bid on the context while audience targeting and frequency capping lose their footing for that impression. The guide to audiences in DV360 describes the same shrinkage from the list side: every list, including those built from the advertiser's own first-party dataData a business collects directly from its own customers, visitors and users., is reduced by consent and browser limits. And the comparison of first-party and third-party data gives a further reason to look at context: bought third-party segments are often less accurate than they appear, and contextual targeting is the preferred alternative where it meets the need.
What context cannot do is set out in the glossary entry: it cannot recognise a past customer, cap frequency on a person or follow anyone across sites, and context is a proxy for interest rather than proof of it. So the plan is not context instead of audiences. It is context where audiences cannot reach, on subjects that genuinely signal the product, judged by the same outcome as every other line item.
From what the audience reads to a context map
The weak way to plan contextual targeting is to open the category list in the platform and tick whatever sounds related. The stronger way starts outside the platform with a written context map: where the people the plan wants to reach spend their reading time, in their own words. Its sources are ordinary research material, and the tutorial on running desk research shows how to gather it in order before anything is bought:
- What existing customers read, from surveys, interviews, customer service notes and the referral sources in the site's analytics.
- The subjects the product is used for, as customers describe them, which are often narrower and more specific than the platform's category names.
- Publishers the business already knows reach its audience, from past direct buys, sponsorships, PR coverage and the placement reports of earlier campaigns.
- The words people use when they write about the subject, including the ones that also appear in unrelated or unsuitable contexts.
- Contexts to avoid, from the brand's own suitability decisions, written down rather than assumed.
Group the map into a handful of themes. Each theme should be a subject the audience cares about that also says something about the product: a page about the subject should make the ad more relevant, not just reach someone who might be interested. Write next to each theme which publishers carry it, which words describe it and which adjacent subjects could pull in the wrong readers. That map, not the category list, is the plan; the platform's options are how it gets built.
Categories, keywords and channels: which form for which job
The DV360 account structure guide lists the targeting a line item holds, including keywords, categories, channels of URLs and apps, environments and brand safety settings, and the inventory and deals guide explains channels as reusable lists of URLs and apps that can be included or excluded, category targeting as the selection of content topics and environment targeting as the split between web, app and connected TV. The Display & Video 360 Help Centre is the reference for the current category list and for any limit on keywords or channel size; this guide does not reproduce them. Advertisers coming from Google Ads will recognise the forms from its content targeting, topics, placements and keywords, which the display targeting guide covers; in DV360 they sit on the line item rather than the ad group.
| Form | Selects on | Best job | Watch for |
|---|---|---|---|
| Categories | The topic assigned to the page or app | Reaching a subject at scale when the subject maps cleanly to a category | Categories are broader than most themes; read results by site before widening |
| Channels of named sites and apps | A list the trader controls | Publishers the context map names, and a quality floor for the rest of the plan | Reach is limited to the list, and the list needs maintenance as sites change |
| Keywords | Words found on the page | Narrow subjects with distinctive vocabulary that no category captures | Words that appear in unrelated or unsuitable contexts; pages with little text |
| Keyword exclusions | Words found on the page, removed | Removing a specific confusion the map predicted | Over-blocking: the same word appears in safe and unsafe contexts |
| Deals with contextual publishers | The publisher's own inventory, negotiated | Guaranteed or preferred access to the best context in the map | Brand safety layers on the line item that exclude the publisher's own inventory |
Categories give scale and channels give control, and most plans need both: a channel for the publishers the map names and a category line item to find more of the same. Keywords earn a line item only where the subject has vocabulary distinctive enough to separate it from everything else. Where the map names a publisher worth paying for access, a deal through the marketplace, covered in the inventory guide, is the contextual buy with the most control of all.
Layering context with audiences, inventory and the suitability policy
Targeting on one line item narrows: an impression must satisfy every layer. That has two consequences for a contextual plan. First, a contextual line item that also carries an audience list is no longer a contextual line item; it reaches only recognised people on the right pages, which is the reach problem the plan was meant to solve. Keep audience layers off contextual line items unless the aim is explicitly to reach known people in a context. Second, every inventory and safety filter removes volume from a context that may already be narrow, so each filter should be there on purpose.
The order that keeps results readable is the same as for any line item: inventory first, then context, then the brand safety floor, then any audience exclusion. Exclude the advertiser's own converters from contextual prospecting in the same way the audiences guide recommends for every prospecting line item, so the budget is not spent re-reaching customers.
Brand safetyControls that keep ads away from content that would harm the advertiser's reputation. and context interact more than in audience buying, because a contextual line item chooses pages by their content. The brand safety and viewability guide separates the safety floor, which the platform's content labels and sensitive category exclusions cover at advertiser level, from suitability, which is the brand's own judgement. It also warns that keyword exclusions over-block and that treating news as unsafe is the commonest suitability mistake. A contextual plan should take its exclusions from the brand's written suitability decisions, set out as the guide to writing a brand suitability policy describes, not from a word list that grows each time someone sees a screenshot, and check each layer against the publishers the map names before launch.
Testing contextual line items against audience line items
A contextual plan is a hypothesis that context can reach the right people at a cost the business accepts. The fair test runs a contextual line item and an audience line item side by side for the same weeks, under the same insertion order, with the same creatives, the same bid approach and the same frequency settings, and judges both on an outcome per unit of spend: cost per order, cost per lead or another outcome the measurement plan already uses. Judging context on click-through rate alone, or audiences on view-through conversions alone, settles nothing.
Keep each contextual form on its own line item while it is being tested, so the result for categories, channels and keywords can be read separately. Expect the contextual line items to reach more people and fewer past visitors than the audience line item, and note that difference when reading the outcome, because an audience line item that reaches mostly recent visitors will look efficient for reasons that have little to do with targeting. Where the measurement plan allows it, a holdout or geographic split answers the incremental question better than any platform report.
The experimentation loop gives the test its shape: write the observation and the hypothesis, decide the measure and the reading date before launch, run it, and record what was learnt before the next change. The impression calculator helps check before launch that each line item's budget and expected price buy enough impressions to produce a readable result in the time available.
Reading results by context and refreshing the plan
Review contextual line items on the same rhythm as the rest of the plan, and read them at two levels. At the line item level, compare each contextual form with the audience line item on the agreed outcome. Below it, read the sites and apps that received impressions, where the report types listed in the Help Centre offer that breakdown, and look for patterns: categories that delivered mostly to a few sites, keywords that matched pages nobody intended, channels whose publishers barely delivered. The DV360 reporting guide explains the report surfaces, the dimensions and why IDs should travel with names; the DV360 hub maps delivery symptoms to the guide that diagnoses them.
Each review feeds the context map. Sites that performed well become candidates for a channel or a deal; categories that delivered mostly to unsuitable or low-value sites get an exclusion or are replaced by a channel; keywords that never separated the subject from noise are dropped. Themes the audience has moved away from are retired and new ones added from fresh research. A contextual plan left alone drifts as publishers change, so the map should carry a review date and an owner like any other part of the plan.
A sheet music retailer plans contextual line items (illustrative)
Plimzarve Sheet Music plans contextual line items beside an audience test
Plimzarve Sheet Music, an invented online sheet music shop, finds its remarketing lists have shrunk by about 40% over a year while its prospecting budget of £12,000 a quarter is unchanged. Customer interviews and the site's referral sources give a context map with 3 themes: choir and singing, piano teaching, and school concerts.
| Line item | Targeting | Budget over 8 weeks | Cost per order |
|---|---|---|---|
| Category | Music and performing arts categories, web only | £2,000 | £31 |
| Channel | A channel of 46 music education and choir sites from the context map | £2,000 | £27 |
| Keyword | 18 keywords such as "sight reading" and "accompaniment", with 6 keyword exclusions for unrelated uses | £2,000 | £52 |
| Audience | Google in-market audience for musical instruments and accessories | £2,000 | £33 |
All 4 line items use the same creatives, a fixed £4.50 CPM bid and the same weekly frequency cap, with existing customers excluded. The keyword line item delivers only 210,000 impressions against about 440,000 for each of the others and matches several pages about computer keyboards. At the review, Plimzarve keeps the channel and category line items on cost per order, drops the keyword line item, adds 9 sites from the category line item's best performers to the channel, and sets the next review for 6 weeks later.
Common mistakes
- Starting from the category list. The platform's menu is how the plan is built, not where it comes from.
- Putting audience lists on contextual line items. The result reaches only recognised people, which is the problem context was meant to solve.
- Mixing forms on one line item. Categories, channels and keywords on the same line item narrow each other and report as one figure.
- Long keyword exclusion lists. They over-block, and most of what they remove was suitable.
- Judging context on clicks. Contextual and audience line items should be compared on the same outcome per unit of spend.
- Forgetting converter exclusions. Contextual prospecting reaches existing customers too unless they are excluded.
- Treating the map as finished. Publishers and subjects change; the map needs an owner and a review date.
