Who Owns What AI Says About Your Brand?

There is a question underneath the measurement question that nobody in the marketing industry seems particularly eager to ask: when an AI tool tells a consumer that your brand is "a budget option with mixed reviews," and that characterization is wrong — or outdated, or simply synthesized from stale sources — who is accountable for that?

Right now, the honest answer is: no one, specifically. And companies are quietly scrambling to manage a reputational risk that has no established legal remedy, no regulatory framework, and no correction mechanism.

A recent Semrush walkthrough by content strategist Chris lays out the practical problem with unusual clarity. "Everyone knows that AI visibility matters," he opens, "but so many teams still have no idea how to measure it. And right now, that gap is costing brands real opportunities." The video offers three tracking methods of escalating sophistication — manual prompt-testing with a spreadsheet, free tools like Google Search Console and Bing Webmaster Tools, and Semrush's own AI visibility toolkit — and it's a genuinely useful operational framework. But read it through a policy lens and the subtext is striking: the entire industry is currently building internal instrumentation for a phenomenon that regulators haven't yet decided how to classify.

That classification problem is not abstract. It will determine, eventually, whether AI-generated brand descriptions constitute commercial speech, whether the platforms producing them carry defamation exposure or FTC disclosure obligations, and whether consumers have any right to know when a purchasing recommendation was shaped by training data that predates a company's last product launch.

What the Measurement Framework Actually Reveals

The three-method approach Chris describes is worth taking seriously on its own terms, because it maps the problem's architecture precisely.

The manual method — querying ChatGPT, Perplexity, Google AI Mode, and other platforms with non-branded prompts like "best organic coffee subscription under $25," then scoring mentions and citations in a spreadsheet — gives brands their first unfiltered look at how AI systems characterize them in competitive contexts. Chris is candid about its limitations: "these results are only from a tiny sample size of one, i.e. you manually entering the prompts," and personalization effects can skew even that small sample unless you're running queries through temporary, unlogged sessions.

The free-tools layer adds evidentiary grounding. Google Search Console, filtered with a regex expression Chris provides to isolate queries of eight words or more, surfaces long-form queries that real users are actually typing — a meaningful distinction from the prompts a brand team might guess at. Bing Webmaster Tools goes further, offering an AI performance tab that tracks citations in Copilot-enabled tools and labels actual user queries as "grounding queries." Chris cites Microsoft's most recent earnings report as the basis for Bing's scale, describing it as reaching 1 billion monthly users — though that figure, drawn from Microsoft's own communications, may aggregate Bing search activity with Microsoft Edge and Copilot usage more broadly, and should be read as a directional indicator rather than a precisely comparable metric to Google's search numbers.

The third tier, Semrush's AI visibility toolkit, is where the measurement becomes genuinely sophisticated — and where the regulatory implications snap into focus. The platform's perception tab doesn't just count mentions; it surfaces the specific phrases AI tools are using to describe a brand in response to category queries. In the demo, Patagonia is shown being described as "a global leader in environmental activism" and "widely considered the industry leader" in recycled materials — accurate characterizations, in this case, and ones Patagonia has clearly earned through consistent public positioning. But Chris names the other scenario explicitly: "If you perform this analysis for your own brand and you're seeing inaccurate or outdated descriptions, then you can understand exactly where you need to get to work to improve your content, your product descriptions, and your overall messaging to change that."

The framing is operational. The implication, sitting just beneath it, is rather more unsettling.

The Regulatory Gap Nobody Has Closed

What Chris is describing — a brand discovering it is being mischaracterized in AI-generated purchase-intent responses, and responding by optimizing its content to influence future model outputs — is currently the only available remedy. There is no dispute resolution process. There is no right of correction. There is no disclosure requirement that would inform the consumer that the AI's characterization was drawn from training data of uncertain provenance and unknown vintage.

This is a live regulatory gap, and it sits at the intersection of several active but unresolved policy debates.

The EU AI Act, which entered into force in August 2024 with staggered implementation deadlines, contains transparency obligations for certain AI systems — but its provisions are most developed around high-risk categories like biometric identification and credit scoring, not around commercial recommendation systems. The question of whether an AI model that steers purchasing decisions through brand characterizations should be classified as a general-purpose AI system with attendant transparency obligations, or as something else entirely, remains genuinely open. The Act's recitals acknowledge the challenge of regulating systems whose outputs influence commercial markets, but the specific mechanisms are still being worked out in implementing regulations.

The FTC's posture is more immediately relevant to U.S. companies. The Commission has been expanding its enforcement theory around algorithmic systems that influence consumer decisions — its 2023 report on AI and dark patterns, its updated endorsement and testimonial guides, and its ongoing attention to "digital deception" all point toward a regulatory frame that could, plausibly, reach AI-generated brand descriptions that consumers encounter in purchase-intent contexts. Whether a chatbot's characterization of a brand constitutes an "endorsement" under the FTC Act, or whether the platform producing it bears disclosure obligations, has not been adjudicated. But the Commission's direction of travel is not ambiguous.

The deeper consumer protection question is this: existing frameworks assume that commercial speech is identifiable — that a consumer can distinguish an advertisement from editorial content, a paid placement from an organic result. Answer engine optimization is already straining that assumption in the context of AEO practices broadly. AI-generated brand characterizations strain it further. When a consumer asks an AI which hosting provider offers the best value and the tool responds with a confident synthesis that places one brand over another, there is currently no mechanism requiring disclosure of how that characterization was generated, what sources it drew from, or whether those sources reflect the brand's current reality.

Building the Evidentiary Record

What Semrush's toolkit effectively enables — and what the manual and free-tool methods approximate at lower cost — is an evidentiary record. A company that has been systematically tracking how AI platforms describe it, what queries trigger those descriptions, and how those descriptions have changed over time has something that will matter considerably when regulators do develop correction and accountability frameworks: documentation.

The pattern in tech regulation is consistent enough to treat as a working assumption. Companies that have already built internal accountability structures when a regulatory moment arrives — that can demonstrate they identified the problem, tracked it, and took reasonable steps to address it — are positioned differently than companies that have no record at all. GDPR compliance teams understood this. The early movers on algorithmic impact assessments understood this. The AI visibility tracking problem is currently framed as a marketing optimization challenge. It will eventually be reframed, at least partly, as a compliance one.

The measurement gap Chris identifies is real, and the framework he offers is genuinely useful for closing it. But the more consequential gap — the one that no spreadsheet or third-party tool can resolve — is that brands currently have no mechanism to compel correction of inaccurate AI-generated descriptions, and consumers have no mechanism to know such descriptions exist or to evaluate their reliability. That is not a measurement problem. It is a governance problem, and right now the brands building the most careful evidentiary records are the ones best positioned to argue their case when someone in Brussels or Washington finally decides to define who owns responsibility for what AI says about you.

---

Samira Barnes covers technology policy and regulation for Buzzrag.