Beyond the hype: Teaching the machines about TV

Opinion

If we don’t fix AI bias, autonomous agents will systematically defund premium media in favour of performance marketing. Jon Block explains how.

The television industry faces an invisible crisis. The AI systems that will soon make billions in media-planning decisions don’t seem to favour TV.

Not because the data says so, but because the data is biased. If we don’t fix that fast, we’ll watch autonomous agents systematically defund premium media in favour of performance marketing – not because it’s more effective, but because the data structure favours digital.

This was the urgent message from a panel I moderated at The Future of TV Advertising Global in London last month. The practitioners on stage – Pippa Scaife (Sky Media), Nadim Samara (MBC Media Solutions), Maria Ingold (Mireality), Simon Rigby (Reigning Media) and Julien Boyreau (Broadpeak) – are implementing these systems. And they’re concerned.

The bias in the machine

AI models are only as intelligent as the data they’re trained on. The vast majority of foundation models – ChatGPT, Gemini, Claude and every AI assistant being deployed in agencies – are trained on the open internet.

The internet is dominated by digital metrics. It understands clicks, conversions, last-touch attribution and the entire language of performance marketing intimately.

What it understands far less well is the language of television: brand lift, long-term memory encoding, shared viewing experiences, and the halo effect of premium content.

If you ask a generic AI agent to plan a media campaign today (try it for yourself!), it will almost invariably skew towards digital channels. Not because digital is more effective. But because the data density for digital is higher.

Digital marketing has flooded the internet with case studies, attribution models and success stories – and that’s what AI systems are trained on. When the medium documents itself, the machines inherit its worldview.

The question isn’t whether AI will shape media planning. It will. The question is who architects that logic. Broadcasters and agencies can structure their effectiveness data, encode their institutional expertise and ensure AI systems understand television’s value. Or they can let foundation models trained on internet bias make those decisions by default.

Synthetic data isn’t enough

The panel emphasised the need for synthetic data to address this gap. Because TV measurement is inherently sparser than digital – we don’t have a tracking pixel on every viewer’s retina – the industry needs advanced data science to create representative models. Effectiveness studies, econometric models, and brand-tracking data structured for AI ingestion.

This is necessary. But it’s not sufficient.

The real opportunity goes further. We’re entering an era in which AI agents no longer consume static training data; they actively query knowledge sources in real time. Emerging protocols like the MCP (Model Context Protocol) and A2A (Agent2Agent) enable AI systems to access structured, proprietary knowledge dynamically during decision-making.

This changes everything. The broadcasters and agencies that structure their effectiveness, data, inventory intelligence, and compliance frameworks as a queryable knowledge architecture will find their institutional expertise woven directly into AI decision-making. Those who don’t will see their value proposition eroded by models trained on the open internet.

In an environment where everyone has access to the same foundational models, the competitive moat is your knowledge architecture. Not your AI model. Your structured institutional knowledge.

Where AI is actually working today

When I asked the panel which departments were using AI most effectively, they cited Data Analytics, Ad Ops, Finance, and Activation. The engine room.

The winning organisations have empowered operational leaders to identify specific friction points and apply AI to address them. They’re building AI assistants that understand their business logic by first structuring their knowledge. Tagging content. Codifying compliance rules. Documenting workflows. Creating knowledge architecture that makes AI useful rather than dangerous.

AI is only as good as the knowledge you give it. Teams moving from data entry to data strategy are structuring institutional knowledge so AI can augment judgment rather than replace it with internet-trained guesswork.

The democratisation dilemma

AI is changing access to television. For years, TV barriers were high: creative agency, production shoot, compliance clearance, complex planning. Weeks of work and significant capital risk.

Generative AI is compressing that timeline from weeks to hours. If AI can generate broadcast-compliant assets from existing brand materials, we open the door to a “long tail” of advertisers who previously found TV inaccessible.

This is happening now. Broadcasters are deploying solutions that enable smaller businesses to rapidly generate broadcast-ready creative. The democratisation of production is real.

But here’s the problem: if those same businesses use AI agents to plan their media mix and those agents are trained on internet data that deems TV irrelevant, the newly accessible inventory won’t be bought. We’ll have solved the production bottleneck whilst the planning bottleneck routes the budget elsewhere.

Democratisation only works if the AI planning it understands TV’s value.

The window is narrow

This is the battleground for the next three years. It isn’t about which AI model you use. It’s about the context you provide it.

If broadcasters and agencies don’t structure their effectiveness data, inventory intelligence, and compliance frameworks as queryable knowledge, the AI will default to internet biases. And, like it or not, the internet’s bias is that TV barely exists.

We know that isn’t true. Brand-building research, effectiveness studies and long-term ROI analysis tell a different story. But the machines don’t know that yet. Once autonomous agents make budget decisions at scale, the patterns will be set.

The opportunity is immediate. Structure your knowledge. Make your effectiveness data queryable. Codify your institutional expertise. Build a knowledge architecture that AI can access dynamically, not just PDFs lost on your internal SharePoint drive.

The TV industry has distinct advantages: trust, attention and premium content. AI offers tools to eliminate the inefficiencies that have historically held the sector back. But only if we remain the architects of the logic, not passive consumers of it.

The machine will learn from someone. Make sure it’s you.

Jon Block is the founder and principal consultant at Syllepsis, an AI innovation consultancy and knowledge engineering firm.