Contextual Targeting

Deciding who sees an ad leaned for years on records collected by following people around the web.

That channel started closing. As the third-party cookie that links one site to another got blocked in browsers and privacy rules tightened, the accuracy of interest targeting — filtering by the topic labels a platform attaches to people — fell with it. Pagedeck watched the cost per signup double in six months on ads running against a productivity label, with no way to see what had changed.

Contextual targeting doesn't look at the person. It picks the slot from the content of the page alone, putting a collaboration tool ad beside an article about preparing for meetings. The premise is single: you may not know who is reading that article, but you do know what they're thinking about.

It runs in five steps.

1. Decide the context. Take a broad subject, narrow it with a word list, or name specific pages. Meeting notes, meeting prep and work documents make one set.
2. Build an exclusion list. The same words always catch some bad slots. Sit beside an article titled "meetings are hell" or a report on an accident and the clicks arrive while the brand takes damage.
3. Match the creative to the context. The value of this method shows up when the ad reads like the article's next sentence. Beside a piece on meeting prep, the first line opens with notes being finished before the meeting ends. Run one identical ad across every slot and there's no reason to use this method at all.
4. Break the results down by slot. With no person-level record, the unit of judgment is the page rather than the person. Separate which subjects produce signups from which produce only clicks, and drop the latter.
5. Convert the cost all the way through. Cost per click usually comes in cheaper — an interest condition at $0.80 a click can land nearer $0.45 here. The verdict still gets made on cost per signup. Cheap clicks that don't convert leave you worse off.

What it can't do is clear too. Without identifying anybody, counting how many times one person saw the ad is difficult, and there's no way to show it again to somebody who visited and left. If either is needed, another criterion gets layered on top.

"Interest targeting has been shaky since the cookie changes, so I want to run one set contextually."

Filling a gap left by record-based conditions with page-based ones.

"Add the exclusion keywords first. I saw one of ours next to an accident report."

The list for filtering out bad slots is empty.

"Broken down by slot, signups only come from the meeting-notes subject. Let's drop the rest."

Proposing to move budget with the page as the unit of judgment.

At Pagedeck, Ravi assembles the word sets and Noa writes a different first line for each subject. The first week, running one creative everywhere, performed no differently from the interest condition; it separated once the opening line was matched to the page.