Why contextual targeting gets better when it's a prompt

There’s a quiet dogma in ad tech that goes like this: audience targeting is precise, contextual targeting is approximate. Cookies and deterministic identifiers let you reach a specific person. Contextual only lets you reach someone who happens to be near some content. Audience wins; contextual is a fallback for the cookieless web.

That dogma made sense as long as “context” meant the URL of the page someone happened to land on. A person on a running-shoe review site might be shopping for shoes, or might be killing time, or might be researching for a friend, or might have clicked the wrong tab. The URL is a weak proxy for intent.

AI-native products break this. The “context” they hand to an ad auction isn’t a URL. It’s a sentence the user wrote themselves, moments ago, describing what they want.

The bandwidth of the signal

Think about how much information actually flows through a URL to a DSP. Domain, path, maybe a few IAB taxonomy tags scraped from the page content. Call it a few dozen bits of useful signal per impression. A DSP then layers audience data on top to fill in everything the URL doesn’t know: the user’s age, interests, purchase history, device history, recent sites visited.

Now think about what flows through a user prompt. “I’m training for my first half-marathon in June and my current shoes are giving me knee pain, what should I look at.” A human advertiser reading that knows more about the buyer’s intent than six months of cookie tracking would have told them. It contains:

• The purchase stage (actively shopping)
• The use case (half-marathon training)
• The timeline (June)
• The pain point (knee pain → probably needs more cushioning or better pronation support)
• The price sensitivity, implicitly (first-time runner, probably not looking at $400 carbon-plated racers)

A DSP bidding on this impression doesn’t need a cross-site profile to bid smartly. The prompt is the profile, and it’s better than any third-party-cookie-derived profile ever was.

Why audience data was always a workaround

Audience targeting became dominant for one reason: the surface the ad lived on didn’t give DSPs enough context to bid well. If all you know is nytimes.com/running/article/123, you can’t tell which user is about to buy shoes and which one is reading to procrastinate. Audience data was the patch — a way to smuggle the user’s state into an auction that otherwise only told you the page’s state.

When the surface itself carries the user’s state, the patch is no longer needed. This is the thing most ad tech veterans aren’t internalizing fast enough: the cookieless future most of the industry has been dreading isn’t a return to weak contextual targeting. It’s a leap forward in contextual targeting, on surfaces that expose a signal richer than any cookie ever was.

What this means for how you buy

The buying motions that work here aren’t the ones that dominate web display today:

• Audience retargeting breaks down, because there isn’t a persistent cross-session identifier to retarget against. This is fine; the moment-of-intent signal usually dwarfs retargeting in lift anyway
• Contextual bidding becomes premium inventory, not discount inventory — the opposite of what display buyers are used to
• Keyword-style buying comes back, but against intent rather than query strings. You can bid on “looking for a new laptop” and reach users regardless of which AI product they asked in
• Lift-based measurement replaces click attribution, because there’s no persistent ID to chain clicks to downstream conversions

The buyers who will do well on conversational inventory are the ones who treat it like search-era contextual — with a budget, a set of targeting rules, and a willingness to let the system optimize against outcomes rather than against cookie matches.

The industry should be paying more attention

The programmatic industry has spent the last five years in a grim conversation about cookie deprecation, identity graphs, and the decay of addressability. Meanwhile, a new surface is quietly opening up where the addressability problem was never a problem in the first place — because the user is telling you what they want, in plain language, at the moment they want it.

That’s not a fallback. That’s the strongest ad signal the industry has ever had.