Is Streaming TV Using AI to Automate Its Own Waste?

Comcast Advertising just surveyed 216 senior advertisers. Eighty-two percent say they're using AI for audience targeting and segmentation.

We've recently covered Truthset's data showing that IP-to-household matching in streaming is only 13% accurate. Age targeting? Also just 13%. That means the machine learning is learning from garbage. The proverbial 'garbage in, garbage out' metaphor.

Fundamentally, this is a plumbing problem disguised as an AI and the math is brutal. Forty percent of every dollar in open programmatic Connected TV is wasted on inaccurate data. That will be approximately $7.4 billion in 2026 ad spend. And 82% of the industry just told Comcast they're feeding AI into that pipeline to make it run faster.

Faster waste is still waste.

'T' is for 'Trust'

Fewer than a third of advertisers say they trust AI to handle advertising tasks. Sixty-one percent say they haven't seen meaningful results. But 82% are using it anyway. That's hardly adoption and looks more like peer pressure with a budget line.

Here's what the smart money is actually doing.

The 30% of advertisers who named measurement and attribution as AI's top priority for 2026 are the ones worth watching. Instead of asking AI to target better, they're asking it to prove what worked. That's a fundamentally different question — and the only one that closes the loop between spend and outcome. The CIMM/4As study found 84% of advertisers see AI-powered measurement as the most consequential shift ahead. Companies like Lumen are already embedding into Netflix infrastructure to make it real.

Three things to do before your next AI-powered campaign:

First, audit your data inputs before you automate anything. If your audience segments are built on probabilistic matching below 50% accuracy, AI will scale the error, not fix it.

Second, stop measuring AI's value by efficiency. Faster buying means nothing if you can't prove it drove a result. Insights are the outcome.

Third, ask your partners one question: what is the verified accuracy rate of the audience data your AI is optimizing against? If they can't answer it, you have your answer.

Streaming doesn't have an AI problem. It has a truth problem, as television has since it's genesis, The promise is that AI will make it run faster, 'in which direction' remains the question to be answered.