By Jack Wagner, CEO, Hawk Digital
AI can accelerate planning, targeting and optimization. It can also automate your bad data assumptions at scale.
Everyone wants faster media planning: faster briefs, faster audience discovery, faster optimization, faster reporting. That’s the promise of AI in advertising. But speed isn’t the same as accuracy.
When AI tools enter the media buying workflow, they don’t just summarize notes or draft campaign copy. They shape the inputs real budgets depend on: audience definitions, contextual signals, targeting logic, creative variants, measurement windows and optimization recommendations. That raises a more practical question than whether AI will buy media: what happens when the model is wrong?
A hallucinated audience isn’t just a bad insight. In most business functions, a bad AI output produces a bad slide or a bad summary. In media, a bad assumption becomes your ad spend. A model can describe an audience that sounds plausible but doesn’t exist the way the plan assumes. The danger isn’t that the model makes things up dramatically. It’s that it makes something slightly wrong sound operationally useful.
01 — The risk is upstream, not downstream
If the audience definition is wrong, the model can optimize beautifully against the wrong target. If inventory context is misunderstood, the campaign chases cheap efficiency while missing brand objectives. If conversion signals are messy, the system rewards behavior that looks like performance but adds no incremental value. This isn’t a creative problem. It’s an infrastructure problem.
02 — Guardrails are now buying infrastructure
For years, guardrails meant compliance: brand safety, frequency caps, fraud filters and privacy controls. AI shifts the focus to inputs: what data the model can use, which signals are trusted or excluded, when an AI-generated audience needs human review, how recommendations are audited before budget moves, and which decisions require operator approval.
The companies that get this right won’t automate the most. They’ll know what should never be fully automated.
03 — Human oversight isn’t friction
In a high-stakes media environment, human oversight is what makes automation usable. AI can speed up the workflow around planning, trading, ops and measurement, but it shouldn’t replace the judgment that makes those decisions valuable. The near-term advantage isn’t fully autonomous media buying. It’s assisted media buying with clear accountability.
Field Guide: What Media Buyers and Operators Should Focus On Now
- Validate AI-generated audience definitions against first-party and verified data before budget moves
- Decide which signals the model may use, and which are excluded
- Require human review for new audiences and large budget shifts
- Audit recommendations and keep logs you can trust
- Measure incrementality, not just platform-reported performance
The Operator’s Perspective
The AI conversation in advertising is moving from content generation to campaign execution. That’s the right direction, and it’s also where the consequences get much larger. A bad headline can be rewritten. A bad audience model can burn budget before anyone realizes the logic was flawed.
AI media buying won’t fail because the models aren’t impressive enough. It will fail where teams confuse confidence with accuracy.
What We’re Watching
AI ad interfaces, emerging standards, and attribution and measurement, including OpenAI’s move into self-serve advertising, where the guardrail conversation is becoming a live product decision rather than a hypothetical.
Work With Hawk Digital
Hawk Digital helps publishers, platforms, brands and agencies connect revenue strategy with the operational work needed to deliver it, across CTV, retail media, commerce, ad operations and AI workflows.
Book a quick call with our team →
This article first appeared in Hawk Talk: “Voices of Innovation,” our LinkedIn newsletter. Read the original on LinkedIn and subscribe for new issues.