If your audience data is flawed, an AI agent will not correct it. It will process that data thousands of times faster than any human team, then push the results out at scale. That is the central argument Mallory Gray, creative director at Skydeo, made in a recent interview reported by Search Engine Journal, and it reframes one of the most common assumptions driving marketing strategy in 2026.
What do AI agents actually do with audience data?
AI agents (software that autonomously executes research, targeting, and content tasks without step-by-step human instruction) work across “thousands of behavioral, purchase, interest, and intent signals simultaneously,” Gray told Search Engine Journal, continuously revising as new information arrives. A human researcher might review a handful of sources, form a hypothesis, and build an audience from there. An agent compresses that process and runs it at a scale no team can replicate. The output quality, however, is bounded by the input quality.
Gray’s employer, Skydeo, draws on 1.4 trillion data points across more than 320 million people, according to figures the company provided to Search Engine Journal. That scale illustrates both the opportunity and the risk: when an agent processes audience signals at that volume, errors compound at the same rate as insights do.
Is getting cited by ChatGPT or Gemini enough?
No, and the gap between citation and conversion matters more than most marketers have acknowledged in 2026. GEO (generative engine optimization, the practice of structuring content so AI tools can easily extract and cite it) has dominated strategy conversations this year. Gray does not dispute that clean structure and credible mentions increase citation frequency. Her argument, as reported by Search Engine Journal, is that citation frequency is the wrong primary metric 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.
Being mentioned and being chosen by the right audience are two different outcomes. A brand appearing in ChatGPT results for prompts its actual customers never ask has visibility with no business value attached.
What does the warning sign look like in practice?
Gray described a pattern that Search Engine Journal’s Greg Jarboe noted many SEO teams may have already encountered. Output keeps climbing: more content, more variations, more campaigns. Engagement or conversion quietly flattens or slides. Volume looks like progress because it is easy to measure and automation makes production cheap.
The harder question Gray says teams stop asking is whether anyone on staff can still explain, in plain language, 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. A strategy can run itself in the wrong direction for months before anyone notices the number that actually mattered was never the one going up.
Has the industry seen this mistake before?
Yes. The DMP (data management platform, a system for collecting and combining audience segments at scale) era of the early 2010s made a similar promise: enough third-party data stitched together at volume would out-target anyone still relying on first-party relationships. It largely did not deliver, because third-party data was frequently inaccurate, and scale compounded the inaccuracy rather than averaging it out. Cookie deprecation by Safari and Firefox eventually pushed the industry back toward first-party and declared signals (data users explicitly provide or that is observed directly on your own properties), even after Chrome dropped its own deprecation plan.
Gray’s argument, as reported by Search Engine Journal, is that the AI agent era is the same lesson running on a newer, faster engine.
What this means for AI-search visibility
The sharpest way to state the practical implication: spending 2026 optimizing for AI citations solves half the problem. The other half is whether the right people are being served those citations.
A concrete example. A local bakery could appear in ChatGPT answers for “best pastry shops in the city” every time the query runs. If the audience data feeding its visibility campaigns skews toward people who never buy in-store or who live 300 kilometers away, those mentions produce no foot traffic. The GEO work is sound. The audience layer underneath it is not. The agent amplified the wrong signal cleanly and at speed.
This distinction matters for how brands interpret AI-search metrics. Citation volume and Share of Model (how often an AI tool mentions a brand relative to competitors) are useful signals, but they are proxies for the outcome that matters. The more diagnostic question is whether those citations appear in response to prompts that match actual customer intent, and whether they convert at a rate that justifies the optimization spend.
The DMP parallel Gray cites is worth taking seriously as a structural pattern. The industry tends to follow a specific cycle: a new data infrastructure arrives, adoption outpaces scrutiny of data quality, scale amplifies errors, and a correction follows. AI agents are positioned to follow the same arc. The brands that build feedback loops between AI output and real business outcomes before scale becomes a factor will be better prepared for that correction.
One dimension the source does not address directly: the quality problem is not uniform across data types. Declared data (what a user explicitly told you), observed behavioral data (what you tracked them doing), and inferred data (what a model estimated about them based on lookalikes or proxies) carry materially different reliability profiles. Agents that mix all three without weighting them differently are likely to amplify the weakest signal at the same speed as the strongest. Building that hierarchy into any AI-agent workflow, before volume becomes the dominant variable, is where the real quality control sits.
Five checks before scaling AI agents in your marketing stack
- Map which audience signals currently feed your GEO and AI visibility tools. Separate declared, observed, and inferred data, and assess what proportion of your targeting rests on each category.
- Review whether the prompts your brand appears for in AI tools match the questions your actual customers ask. A mismatch points to an audience quality gap, not a content production gap.
- Check conversion rates on AI-referred traffic. If citation volume has risen but conversions have not moved, the audience data question is already worth investigating.
- Establish a standing review where someone on the team explains, in plain language, why a specific audience or message was selected. If the answer is “the model chose it,” pause before adding more volume.
- Apply the same skepticism to third-party data sources powering your AI-agent workflows that the industry learned from the DMP era: scale amplifies error rates, it does not average them out.
What to watch next
The tension Gray identifies between optimizing for AI citation frequency and optimizing for audience data quality is likely to sharpen as agents take on more of the research and targeting workload. How measurement frameworks evolve to capture the difference between “mentioned” and “chosen by the right person” is the question worth tracking through the rest of 2026.
Source: Search Engine Journal, Greg Jarboe, September 2026
Lam Nguyen · Hingewise
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.