What happened
On August 7, 2026, Hank Green published a Vlogbrothers video called "Just Trying to Figure Out My Job." The video's public YouTube feed description lists a new personal AI policy: no portion of any script will be written, edited, or outlined by an LLM; the thesis of a video will originate with a human; no image or music in a video will be generated by AI; LLM outputs will not be trusted as a source; and informational or educational videos will include links to primary sources in the description.
Business Insider reported the same day that Green created the policy after addressing his earlier apology for relying too heavily on AI for research. The article also reported that he encouraged other creators to have an AI policy, even if they do not publish it.
The criticism started earlier. People reported on August 7 that viewers questioned Green's use of AI after a July 29 Complexly video and after Green acknowledged using AI to locate papers and other research resources. Axios framed the broader reaction on August 8 as part of a creative-industry climate where even suspected AI use can damage trust.
REC sees the useful signal in Green's move from a loose personal workflow to explicit operating boundaries because his audience needed to know where the human judgment came from.
Where AI still helps
This is not a claim that AI can never help with research, organization, editing logistics, transcription, captions, or distribution planning. Green's own discussion separated AI-assisted research from AI-written scripts, and the public reaction shows why that distinction needs to be clear before content ships.
It is also not a reason to turn creator workflows into legalistic theater. A policy that nobody can remember will not guide behavior under deadline pressure. A disclosure that arrives only after backlash will not repair the missing source trail.
For expert communication, the question is more practical: when a viewer reads, watches, or hears a claim, can the team explain which part came from the person, which part came from source material, and which part AI helped organize?
REC's read: boundaries beat vibes
The weak version of an AI policy is a vibe: we use AI responsibly, humans stay in the loop, trust matters. Those lines may be true, but they do not help an editor, founder, researcher, or creator decide what to do with a draft at 5 p.m.
The useful version is operational. It tells the team what AI is allowed to prepare. It names the parts AI cannot originate. It says which sources count. It identifies the person who owns the final meaning.
Green's policy is useful because it draws those lines in plain language. Scripts, outlines, and theses stay human. Images and music are not generated by AI for his videos. LLM outputs are not sources. Educational videos need primary-source links.
That is more actionable than a generic disclosure label. A label may tell the audience that AI was involved. A boundary tells the creator how to keep authorship intact before there is anything to label.
Why this matters for expert-led content
Expert content carries a different burden than entertainment formats built around experimentation or fiction. Expert content earns its value when a person with relevant experience, responsibility, or judgment makes a claim they can stand behind. Polish and clarity help the audience understand that claim.
AI can make that work easier to prepare. It can summarize supplied links, draft sharper interview questions, find repeated themes in a transcript, suggest clip candidates, or format a post for a channel. Those are real workflow advantages when they are used carefully.
The risk is that the tool quietly moves upstream. It stops preparing the conversation and starts shaping the thesis. It stops organizing source material and starts becoming the source. It stops helping the expert speak and starts sanding the expert into a generic voice.
That is where a research-guided interview helps. The person answers a specific question in their own words. The transcript becomes source material. The editor can check whether a clip, article section, or social post fairly represents the answer. AI can still assist around that source, but it does not become the origin of the claim.
A practical boundary note
Before publishing an AI-assisted expert asset, write a short boundary note. It does not have to be public in every context. It does have to be clear enough that the team can inspect it later.
First, state the human-origin rule. Which parts must come from the person: thesis, example, judgment, caveat, decision rule, recommendation, or final approval? If the asset has no human-origin part, do not present it as expert-authored.
Second, state the AI-assist rule. Name the jobs AI may do: research preparation, question drafting, transcript cleanup, theme grouping, clip suggestion, caption options, formatting, or channel adaptation. Specific verbs are better than broad claims.
Third, state the source rule. Do not let an LLM output serve as the evidence for a factual claim. Link the claim to a primary source, a reputable report, a product document, a recording, a transcript, or another inspectable artifact.
Fourth, state the review rule. Someone has to compare the final wording with the source and decide whether the meaning survived the edit. That person should be able to say what changed and why.
Fifth, state the disclosure rule. Decide what the audience needs to know, what the platform requires, and what belongs in the internal record. The answer may vary by medium, but the decision should not be improvised after someone complains.
Write the boundary before publishing
Green's policy shows that AI boundaries are becoming part of the craft of trustworthy publishing. Each creator can choose red lines suited to their own work.
For REC's world, the strongest boundary is simple. Use AI to prepare the expert, organize the source material, and make the review process easier. Keep the thesis, examples, judgment, sources, and final responsibility with a real person.
When an audience asks whether a piece of content is actually the creator's point of view, vague reassurance is weak. A practical boundary note is stronger because it connects the public asset back to the human source and the evidence behind it.