Claude's /goal Command Is Automating SEO—Who's

When a piece of software removes the human from the loop entirely—not just assists them, not just speeds them up, but literally operates while they are away from their computer—that is not AI assistance. That is delegation. The question regulators haven't caught up to yet is: delegation to whom, and accountable to what?

That's what's actually at stake in SEO consultant Julian Goldie's recent walkthrough of Claude Code's new /goal command. The video is pitched as a practitioner tutorial, and the technical demonstration is genuinely instructive. But read it from where I sit—tracking how policy frameworks designed for human-generated content are straining against systems that were never contemplated when those frameworks were written—and a different story surfaces.

What `/goal` Actually Does

The mechanism is straightforward enough. You invoke /goal in Claude Code followed by a completion condition. Claude begins working toward that condition across multiple turns without requiring further input. A separate lightweight evaluator model checks after each turn whether the condition has been satisfied. If not, Claude keeps going. If yes, it stops and surfaces the output.

Goldie's demonstration prompt is a useful illustration of scale: a complete SEO-optimized website package including a homepage, five landing pages each with unique H1s and FAQ sections, a 20-article blog structure targeting specific keywords, a full internal link map, schema markup, and meta tags for every page—all with the instruction "don't stop until every page, every article outline, every meta tag, every internal link has been completed and documented."

The evaluator architecture is what separates this from prior automation approaches. /goal differs from /loop (which runs on a timer until you stop it) and from auto mode (which approves tool calls within a single turn but doesn't self-reinitiate). Goldie describes it plainly: "Auto mode won't just keep going on its own if the job isn't done. Whereas with goal, it will keep going over and over again until the evaluator says yeah, this is ready to go." Claude Code can also spawn parallel sub-agents handling different portions of the task simultaneously, which means a single session can produce content at a volume that would previously have required a staffed team.

The token costs are real—Goldie flags this as a constraint worth managing—but the operational model is otherwise close to what the pitch claims: set a verifiable condition, walk away, return to completed output.

The Enforcement History Goldie's "It's 2026" Framing Skips

Goldie dismisses concerns about AI content ranking by dating them to 2022: "The belief that AI content doesn't rank comes from 2022, but it's 2026 now. Google ranks helpful content. AI can produce helpful content at scale."

That is a selective telling. Google's March 2024 core update—the most significant algorithmic intervention in years—specifically targeted what Google's documentation described as "content that seems to have been created for ranking purposes" and "scaled content abuse," language that search quality analysts widely interpreted as aimed at high-volume, low-differentiation AI-generated sites. The update hit a number of AI-heavy publishers hard, with some reporting traffic losses of 50 to 90 percent within weeks. Several of those sites have not recovered.

The video was published recently, and Goldie's own Google Search Console screenshots do show upward traffic trajectories across multiple sites. That evidence is real, as far as it goes—though it comes from Goldie's own promotional material and cannot be independently verified by anyone watching the video. His claim that one piece of content ranked first for a target keyword within 24 hours of publication is striking enough to note without a corroborating source: extraordinary results sourced only to the person selling the method should be read accordingly.

The more honest framing is that AI-assisted content can rank, that some practitioners are achieving it, and that Google's enforcement posture has been inconsistent enough that the risk calculus varies significantly by niche, domain authority, and how closely the output resembles what Google's quality raters are flagging. "It's 2026 now" papers over an enforcement history that matters to anyone making actual business decisions based on this approach.

The Regulatory Surface Nobody in This Conversation Is Discussing

Here is where I have to depart from the practitioner frame entirely, because this is the question my readers need to understand.

When a business deploys an autonomous pipeline that produces hundreds of AI-generated web pages and publishes them without disclosure, it is operating in a regulatory environment that has not formally resolved whether that constitutes a deceptive practice. The FTC's guidance on AI-generated content in marketing contexts—particularly its 2023 policy statement on deceptive endorsements and its ongoing enforcement focus on undisclosed AI—has not been applied to pure content-generation scenarios in any definitive ruling. But the agency's stated principle is clear: consumers are entitled to know when the information they're receiving has been produced by automated systems, particularly when those systems are being used to influence commercial decisions.

Goldie's use case is primarily commercial SEO—driving traffic to landing pages, service offers, affiliate content. A pipeline producing hundreds of optimized pages for a business that sells products or services, with no disclosure that those pages were autonomously generated, sits in territory the FTC hasn't formally mapped but has signaled it considers within scope. The EU's AI Act, now in force, imposes disclosure requirements on AI-generated content in ways that U.S. law has not yet matched, which means businesses operating internationally face an asymmetric compliance surface.

This isn't an argument that the tool shouldn't be used. It's an observation that the industry is, characteristically, running several laps ahead of the regulatory framework designed to govern it, and that the practitioners being taught this workflow in videos and Discord communities are not being told that.

The Wider Question

Goldie's framing of SEO, GEO, and AEO as "all the same thing"—"it's are you ranking, are you getting traffic"—is operationally sensible. For someone running content operations at scale, collapsing the distinctions is a reasonable efficiency move. But the flattening carries a cost that isn't felt by the person running the pipeline.

The web that autonomous content loops produce at scale is a web where ranking is the primary signal of value, where the evaluator checking quality is the same system generating the content, and where human editorial judgment has been explicitly removed as a friction point. That is a coherent business model. It is also a description of how information environments degrade—not through any single bad actor, but through the aggregate effect of thousands of pipelines all optimizing for the same verifiable conditions.

I don't think Julian Goldie is building a disinformation operation. I think he's doing exactly what the tool enables and what his audience is asking him to show. The problem isn't intent. It's that no one in the room—not Goldie, not Anthropic, not the FTC, not Google—has fully reckoned with what happens when every small business owner with a Claude subscription is running a /goal loop while they sleep.

The evaluator model checks whether the schema markup is complete. No one has built the evaluator that checks whether any of this should have been written at all.

By Samira Okonkwo-Barnes, Tech Policy & Regulation Correspondent