AI Agents Won’t Fix Bad Audience Data, They’ll Amplify It via @sejournal, @gregjarboe

AI agents are supposed to replace a lot of what human researchers used to do before a purchase decision got made, including pulling sources, spotting patterns, and building audiences.

This isn’t necessarily wrong; it’s that everyone drawing the “agents replace research” conclusion is skipping the actual mechanism. Agents don’t replace the need for good audience data. They run on it, at a scale and speed no human team can match, and whatever quality that data was, good or bad, comes out the other end louder.

To test this, I put some skeptical questions to Mallory Gray, creative director at Skydeo, an audience data company that says it draws on 1.4 trillion data points across more than 320 million people. Skydeo sells exactly the kind of data this argument puts at the center, so it has an obvious stake in the answer. The argument still has to stand on its own, and I think it does.

“A human researcher might look at several sources, identify patterns, form a hypothesis, and build an audience from there,” Gray told me. An agent works across “thousands of behavioral, purchase, interest, and intent signals” simultaneously, continuously revising as new information arrives. Humans still decide what matters and what the brand should do about it. The agent just expands how much raw material can realistically feed that decision.

Being Mentioned Isn’t The Same As Being Chosen

I think that cuts against most of what passes for AI marketing strategy in 2026. Marketers have spent the year racing to get cited by ChatGPT and Gemini, treating a mention as the finish line. I made almost the identical argument several weeks ago about agentic commerce: getting a product into ChatGPT’s results turns out to be the easy half, and whether the checkout underneath can actually process a machine-speed transaction is the part nobody tested. Gray’s argument is the audience-data version of the same mistake. Showing up in an AI answer and showing up for the right reasons are not the same accomplishment, and treating them as interchangeable is how a brand ends up automating its own blind spots.

That distinction matters more once you separate GEO from what Gray calls AI visibility. GEO, as most marketers currently practice it, is about making content easy for a generative engine to extract and cite. Clean structure, clear answers, credible mentions. Gray doesn’t dispute any of that works. Her argument is that citation frequency measures the wrong thing on its own. “A brand can increase its visibility in AI answers substantially without seeing the same improvement in qualified traffic or conversions,” she said, and the fix isn’t more optimization; it’s asking who is actually being served up your brand and whether that matches who your business needs.

Output Up, Results Flat – The Warning Sign To Watch

The warning sign she flagged for when this goes wrong is one that I’d bet half the SEO teams reading this column have already tripped. Output keeps climbing, more content, more variations, more campaigns, while engagement or conversion quietly flattens or slides. Volume looks like progress because it’s easy to measure and automation makes it cheap to produce. The harder question, the one Gray says teams stop asking, is whether anyone on staff can still explain why a particular audience was targeted or a particular message went out. Once the honest answer becomes “the AI chose it,” the feedback loop that used to catch a bad assumption is gone, and a strategy can run itself into the ground for months before anyone notices the number that mattered was never the one going up.

We’ve Been Here Before – The DMP Era

There’s a useful parallel here from the last time marketing got obsessed with a data platform. The DMP era of the early 2010s promised that enough third-party data, stitched together at scale, would out-target anyone still relying on first-party relationships. It mostly didn’t, because the third-party data was frequently wrong, and scale just meant the wrongness compounded faster. Cookie deprecation from Safari and Firefox forced the industry back toward first-party and declared signals, even after Chrome dropped its own plan to deprecate them.

Gray’s argument about AI agents is the same lesson with a new engine. Access to a powerful model is getting increasingly common. What a brand feeds into it isn’t, and that’s still where the advantage sits.

3 Checks To Run Before You Scale

There are three key ways to apply this to your strategy immediately:

AI agents are turning audience data quality into the thing that separates a brand that gets mentioned from a brand that gets bought. I said it about checkout infrastructure a few weeks ago and I’ll say it again here about targeting data. The agent isn’t the differentiator. What you fed it before it ever started working was.

More Resources:

Featured Image: Krot_Studio/Shutterstock