Skip to main content

Moderation Standards

How decisions get made. Public so everyone can see the process.

All features described on this page are implemented. Policy text is pending final legal review.

Why this page is public

A common pattern on other platforms is that moderation policies are vague, inconsistent, and applied behind closed doors. We don't think that's compatible with running a marketplace where livelihoods are at stake. The standards moderators apply are the same standards we publish here. If we change them, we change this page.

This means a buyer can see exactly what kind of evidence will get a listing removed. A seller can see exactly what counts as a successful defense. A reporter can see what we consider a useful report and what gets dismissed. We believe transparency is cheaper than surprise.

Principles moderators apply

  1. Humans review reports, not AI. The platform that protects human-made art will not use AI to judge what is and is not human-made. Tooling can help moderators triage and surface relevant evidence, but the decision is always made by a person.
  2. The seller has a right to respond. No listing is removed based on a report alone. The seller is notified and given a real opportunity to provide evidence before any action is taken, except in cases where the violation is obvious from the listing itself (e.g., a clearly Midjourney-style image with no claim of other process).
  3. The burden of proof scales with the consequences. A first removal of a single listing requires less than a shop suspension. Permanent action requires clear and repeated evidence.
  4. Reasonable doubt favors the seller. If a moderator is genuinely uncertain after reviewing evidence, the listing stays. We would rather miss a few edge cases than wrongly remove a real maker's work.
  5. False reports are tracked. Reporters who repeatedly file unfounded reports against the same seller, or who use the report system to harass competitors, will see their reports given less weight or rejected.
  6. Decisions are explained. Sellers who have a listing removed receive a written explanation referencing the specific policy section. Sellers who successfully defend a report receive confirmation that the report was reviewed and dismissed.
  7. Appeals go to a different moderator. The moderator who made the original decision does not review the appeal. This is a basic fairness check.

The review process, step by step

  1. Report received. A buyer, another seller, or an internal flag triggers a report. Reports must include a reason (selectable category) and may include free-text detail and links.
  2. Triage. Reports are checked for completeness. Vague reports ("looks fake") without any specific signal are returned to the reporter for more detail rather than acted on.
  3. Initial review. A moderator reads the report and looks at the listing. If the violation is obvious from the listing itself, the seller is notified and given 7 days to respond before removal.
  4. Seller response. The seller submits evidence per the evidence guide on the seller page. The most common successful response is a short video showing the work in progress on the seller's bench.
  5. Decision. The moderator weighs the evidence. The default is "stay" unless the evidence clearly supports removal. Sellers and reporters are both notified.
  6. Appeal window. Removed listings can be appealed for 30 days. The appeal goes to a different moderator who has not seen the original decision.
  7. Logged for pattern review. All decisions are logged so that patterns of behavior — repeat offenders, repeat false reports, unusual moderator outcomes — can be reviewed periodically.

How the moderation queue works

Reports land in an admin triage queue where a human moderator reviews each one. Reports are sorted by priority: multiple reports from different, reputable reporters escalate faster. The moderator sees the reported listing, the reporter's explanation, any C2PA watermark detection results from our automated scan, and the seller's history. Each report gets a status (pending → reviewed → resolved) and every decision is audit-logged. Sellers are notified when a report is filed against them and given 7 days to submit evidence before any action is taken. Our target is to review all reports within 3 business days of submission.

What gets removed and what stays

Removed

  • Listings where a clear Midjourney/Stable Diffusion/DALL-E-style image is presented as original art and the seller offers no counter-evidence
  • Shops that drop-ship from third-party fulfillment addresses while claiming handmade
  • Listings where the seller admits the work was AI-generated
  • Listings whose photos appear elsewhere as known generated content or stock
  • Disclosure-required categories where the disclosure is missing after warning
  • Repeat violations after a clear warning on the same category

Stays

  • Listings where the seller provides credible process evidence (video, in-progress photos, source files)
  • Listings reported on the basis of style alone without specific claims
  • Listings where the report appears to be from a competitor with a history of unfounded reports
  • Listings where the moderator is genuinely uncertain — reasonable doubt favors the seller
  • Disclosed generative / procedural / AI-narrated work that complies with disclosure rules

What we will not do

  • Use automated AI detection as the basis for removal. Current AI image detectors are not reliable enough to justify taking someone's listing down. They may inform a moderator's reading but cannot drive the decision.
  • Remove listings on style grounds alone. "It looks AI-generated" is not a reason. Reports must point to a specific signal — admission, identical image found elsewhere, drop-ship pattern, scale inconsistent with handmade, etc.
  • Suspend a shop on a single removal. Suspensions follow patterns, not individual incidents.
  • Use removal as a competitive weapon. Sellers who file repeated reports against direct competitors will be reviewed.
  • Hide our decisions. Both seller and reporter receive an explanation of what was decided and why.

Edge cases we publish on purpose

Some cases are genuinely hard. Rather than pretend they aren't, we've documented our reasoning publicly so you can see how the rule applies:

Style transfer applied to the seller's own photograph

Allowed. The photograph is the seller's; the AI tool is acting as a filter, not as a generator. The creative choice (subject, framing, moment) was human.

Comic colored or inked by AI from the seller's pencils

Prohibited. The inker and colorist are creative roles. Replacing them with an AI tool is closer to selling collaborative work as solo work without disclosing the collaborator.

AI-generated background behind a 3D physical product photo

Allowed. The product is real and handmade; the background is staging. Photo backgrounds in catalogs have always been constructed.

Procedural / generative art written in code by the artist

Allowed with disclosure. A 60+ year old art form. The artist is the author of the system that produces the image. Disclosure required so buyers understand the production method.

Vintage / antique item identified with help from AI

Allowed, disclosure recommended. AI is a research tool here, like a reference book. Sellers who voluntarily disclose tend to face fewer disputes.

Numbers we plan to publish

Once the moderation system is processing volume, we plan to publish quarterly transparency numbers — how many reports we received, how many were upheld, how many were dismissed, average response times, appeals overturn rates. These will appear here on this page.

Transparency reporting — coming after launch

Once Hobbsy is live and the moderation system is processing real reports, we will publish quarterly transparency numbers on this page: total reports received, reports upheld vs dismissed, average response time, appeals filed and their outcomes, DMCA notices received, and repeat-infringer actions taken. We believe moderation systems that operate in secret lose trust — publishing our numbers keeps us honest. The first report will appear after we have enough volume to be statistically meaningful (we won't publish a report based on 3 reports total — that's not transparency, that's noise).

Other places to look

Questions about how moderation works? Email policy@hobbsy.shop