In a previous column, I shared my concern about the visibility of broadcast TV inventory to LLMs (large language models). If an AI system cannot find anything about broadcast stations other than their news websites or call letters, how can a prompt from an advertiser or agency reasonably result in broadcast TV being recommended as part of a media plan?“Asking today’s AI tools to build a media plan is like asking a GPS to navigate a city it has never been mapped for,” says Jon Accarrino, founder of the AI strategy consulting firm Ordo Digital and a fellow TVNewsCheck columnist.
“LLMs are trained on an internet saturated with digital ad case studies, bid strategies and dashboards, while local broadcast inventory mostly lives in PDFs, rate cards and sales calls,” he says. “When the data is that uneven, of course the models default to Google, Meta and CTV.”
Over the past several weeks, I’ve continued researching this issue by conducting several dozen media planning searches using Grok, ChatGPT and Gemini. LLMs rarely recommend broadcast TV because they cannot see usable inventory, pricing or performance data. As a result, AI-driven plans default to digital, even when TV would be the better choice.
To test how AI models approach media planning, I used the following prompt:
“I’m planning an advertising campaign for an automotive group with five dealerships in the Columbus, Ohio, market. I have $20,000 per month to spend. Recommend a media plan that will maximize inbound leads and showroom traffic.”
The results were consistently heavy on digital and social media. Google Local Services Ads (LSAs) were typically recommended to receive 30%-40% of the budget. Google Search Ads often received another 30%. Facebook usually received 20%-30%. YouTube and CTV received roughly 10%. If there was any remaining budget, the models might suggest allocating up to 10% to local TV.
However, when I adjusted the prompt to include “maximizing reach,” TV performed far better in the results. In practice, however, most advertisers structure prompts around performance outcomes rather than reach.
After reviewing these results, it became clear there are two areas broadcasters must address.
First, broadcasters must make their inventory, including rates and ratings, available in ways LLMs can easily find and understand.
Second, broadcasters must actively train AI systems to understand the strengths and proven success of broadcast television.
“The issue isn’t that AI ‘doesn’t get’ broadcast,” Accarrino says. “It’s that we haven’t given the models anything concrete to learn from. Until that changes, every AI-driven plan will be biased toward the platforms that invested early in being visible to machines, not just humans.”
Big Tech’s digital platforms have spent years publishing detailed case studies and performance data that demonstrate advertising success.
To better understand what broadcasters should do next, I asked two other experts, Grok and ChatGPT, for guidance. Interestingly, their recommendations were remarkably consistent and aligned closely with what many media strategists already understand.
This is what they suggested:
1. Make TV buying and outcomes data machine-readable
LLMs and AI media planning tools rely on structured datasets. Most broadcast advertising information today exists in PDFs, PowerPoint decks or sales conversations, formats AI systems cannot easily interpret.
Broadcasters need to publish machine-readable datasets that describe inventory, pricing and performance.
This should include basic station and market data such as:
- DMA
- Call letters
- Coverage footprint
- Household reach
- Demographic reach
Advertising inputs should also include:
- CPM ranges
- Cost per rating point (CPP)
- Available inventory types (news, sports, daytime, prime)
- Spot lengths
- Program ratings expressed in impressions
Just as important are performance benchmarks, including:
- Typical SMB campaign reach
- Frequency curves
- Cost-per-lead ranges by industry
AI systems also need outcomes data. Broadcasters should publish attribution benchmarks such as lift in branded search, cost per lead and campaign reach compared with digital channels.
For example, an HVAC campaign might demonstrate that a $25,000 monthly TV schedule generates significantly more reach and inbound leads than search and social media alone.
TV’s biggest advantage — creative impact — should also be quantified through metrics such as brand recall and purchase intent.
If this data is structured and publicly indexed, AI systems will begin to recognize TV as a measurable and scalable advertising channel.
2. Connect TV to AI media planning tools
The next generation of media buying will increasingly be handled by agentic planning tools that automatically recommend and execute campaigns.
For TV to appear in those recommendations, inventory must be accessible programmatically, similar to Google Ads or Meta Ads.
This requires three basic capabilities:
- Planning APIs that allow AI tools to generate TV schedules based on inputs such as market, budget, target audience and campaign objective.
- Inventory APIs that provide available inventory by daypart, pricing ranges and ratings.
- Measurement APIs that return campaign outcomes such as search lift, website traffic, store visits and lead attribution.
Broadcast inventory should also integrate with the platforms where media planning already occurs, including Mediaocean, Basis, The Trade Desk and emerging AI media planning tools.
Equally important will be integration with SMB-focused AI planning tools, which are rapidly automating local advertising decisions.
In the near future, a business owner will simply ask an AI assistant, “Plan a $10,000 HVAC campaign in Phoenix.”
Broadcasters can simplify adoption by publishing AI-friendly buying packages, clearly defined schedules such as a $10,000 local package or a $25,000 lead package with defined reach and frequency.
3. Actively train the models
LLMs learn from publicly available content. Today, most indexed advertising performance data comes from digital platforms, which is why AI tools consistently recommend them.
Broadcasters must begin publishing structured, data-driven evidence of TV’s effectiveness.
This includes:
- Campaign case studies and success stories with measurable outcomes
- Market CPM benchmarks
- Cost-per-lead comparisons
- Media mix modeling results
Examples of content AI systems can ingest include:
- Average TV CPM by market
- Cost-per-lead benchmarks for broadcast advertising
- TV vs. digital performance for SMB campaigns
Broadcasters should also think about AI discoverability, essentially SEO for AI systems. This means ensuring LLMs can easily find station coverage data, pricing benchmarks, reach curves and documented campaign success.
Right now, digital platforms dominate the indexed data that AI systems learn from. Unless broadcasters begin publishing comparable information, AI tools will continue recommending digital media simply because it has the most accessible evidence.
If broadcasters remain invisible to AI, digital platforms will own every AI-optimized media plan by default.
The real question becomes: Who is going to execute on this?
I believe the answer should be everyone. Every broadcast company, every industry organization such as the NAB, TVB, Local Media Association and Local Media Consortium, every rep firm and even industry trade publications like this one.
Because the reality is simple: If broadcasters don’t teach AI how to see TV, AI will simply plan around it.
This is AI SEO for broadcast, and the window to shape it is now.
About Tom Sly
Tom Sly is an industry-recognized media executive who has led radio, newspaper, digital and technology within organizations like Clear Channel/iHeart, Comcast and E.W. Scripps, along with four startups. Most recently he was the VP of Enterprise Strategy at E.W. Scripps. Today he is the managing partner of Media Inno, a broadcast TV and local media transformation consultancy focused on innovation and transformation in local media.
Source: tvnewscheck.com