Contextual Targeting Built on Page Meaning, Not Keywords

On a news-heavy Indian buy, the exclusion list a brand safety team hands over usually removes about a third of the reachable quality inventory before a bid is placed, and most of what it removes is a headline word sitting in an article about something else. Rebuilding that list around page meaning rather than string matching tends to return more than any contextual vendor you could add this year. Where the identifier problem actually bites: Chrome kept third-party cookies. The inventory with no stable user identifier is still large, and it is specific: Safari and Firefox traffic, users who declined consent, in-app supply with a reset or absent advertising ID, and connected TV. Sizing that share against your own log-level delivery comes first. A plan designed for a 70 percent problem when you have a 25 percent problem overspends on classification fees and underspends on the deal work that carries the result. Classification, and what it can actually read: Vendors are tested on your pages and your target sections, not on a coverage deck. The question that decides it in this market is language. Most semantic classifiers read English well and Hindi, Tamil, Telugu, Marathi and Bengali poorly, so a plan can quietly confine a national brand to English inventory. That should be a decision somebody made, not a side effect. I map the buy to the IAB Content Taxonomy at page level and then measure what share of your intended supply the vendor classifies at all, because unclassified pages are the line nobody reads and the place spend leaks. Deals over open exchange: Most of the outcome sits in deal construction rather than in a targeting toggle. Curated deal IDs assembled at the SSP, publisher-direct where a section earns it, and supply paths checked against sellers.json and ads.txt so you are not buying the same impression through four hops. Made-for-advertising domains are removed by looking at the path and the page, not by trusting a vendor's flag. What is handed over: A targeting plan with the taxonomy nodes, negative rules and deal IDs written out. A rebuilt exclusion list where every surviving term has a reason next to it and every removed term has the inventory it was costing you. A supply path shortlist with the hops you should refuse. And a designed test: contextual cohort against a matched control, with the reporting query written so your analyst re-runs it without me. Not included: Creative and dynamic creative production. Vendor resale of any kind. You contract Peer39, IAS, DoubleVerify or GumGum directly, see the rate card, and own the relationship. Ongoing list maintenance beyond the agreed scope. Brand safety policy sign-off, which belongs to your legal and communications teams and should never sit with the person buying the media. Who this is not for: Buyers who want contextual to replace retargeting reach at the same cost per acquisition. It will not, and a plan that promises it is guessing in front of your finance team. Teams that cannot hold a control group for four weeks, because then nothing here is measurable and you are buying an opinion. Medtech advertisers who want contextual used as an indirect route to a diagnosed audience by targeting symptom and treatment pages: that is condition targeting with a different label on it, and I decline it. Vertical notes: Education buys turn on exam-season pages and course-comparison sections whose value collapses the week results are published, so flighting matters more than the taxonomy. Travel demand reads inspiration and itinerary content weeks before it reads a booking page, and the two want different frequency.