Requirements for Articled Approved Organizations
Six clauses, created to be verifiable rather than impressive. This is the rulebook: what a member signs, how we test it, what we do when a check raises something, and where the standard stops.
Core Articled Approved Organization Requirements
Every clause here was formed backward from a question we could actually answer. A rule that cannot be tested stands as decoration. Hence, the commitment avoids sentiments about craft and states actions instead. The word doing the work is generative. A spell-checker that underlines a typo, a style guide that flags a passive verb, a search engine that returns links: none of those compose text, and none are limited.
What is limited is any tool that generates or reproduces sentences. Drafting an outline, expanding notes into prose, rewriting a paragraph for tone, translating an article, paraphrasing a source, generating a headline or a meta description, and summarizing an interview into publishable text are all covered. The reverse operation is covered too. Running finished machine text through a humanizer is the clearest breach in the standard, because that tool exists for no purpose except circumventing a check.
The record-keeping clause does the quiet heavy lifting. Members keep, for twelve months, a record of who wrote each published piece. This can be a spreadsheet. When a detector flags something eleven months later, that spreadsheet is the difference between a five-minute conversation and an unanswerable one, and organizations that skip it tend to discover the cost at the worst possible moment.
- Generative models must not draft, outline, expand, rewrite, translate or paraphrase any published text.
- No AI humanizer, paraphraser or detector-evasion tool may be used on any text, ever.
- Records of authorship for every published piece are kept twelve months.
- Unannounced sampling of publicly available work is accepted, without notice or negotiation.
- Inquiries about a named piece are answered by a person within ten working days.
- Any change to the policy reaches us within thirty days of taking effect.
How We Review Articled Approved Organization Requirements
Review starts at the desk. We read the public policy page against the signed commitment and look for the gaps that matter: a policy that bans AI writing but is silent on AI editing, one with no effective date, one naming a contact who has left. Most first applications need at least one revision here. That saves time compared with discovering the gap during a check a year later.
Sampling follows, and the unpredictability is deliberate. We pick the pieces ourselves, prioritize the most recent, and do not disclose to the member which URLs we used. Each sample runs through the twelve detection and originality platforms listed on our verification pages, the same tests every Articled draft must pass before delivery. Scores are kept with the date, so a pattern across years stays visible even when no single check was conclusive.
Then the limits, which are real. We do not review systems, read internal drafts we were not shown, interview employees, or hold any power other than removing a listing. A member determined to be dishonest can keep a listing for a while. What sampling catches is the ordinary case: a team that quietly started using a tool for headlines, or a vendor nobody was monitoring. This is as far as a free standard reaches, and pretending otherwise would be counterproductive.
A flag is not a finding
Detector scores are probabilistic, and human writing is flagged daily. A high score means a question about one named piece. Authorship records, drafts and version history usually answer it within a week.
Ongoing Articled Approved Organization Requirements
Approval is dated, not permanent. Once a year we ask the named contact to confirm, in a personal response rather than an autoreply, that the commitment still holds and that the signatory is still in place. A new sample is collected at the same time. Members that have grown, hired or restructured usually find this is when they notice a gap in their own processes.
Between annual checks, the clock restarts for three reasons. First, a reader report about a specific piece, acted on without naming the reporter. Second, a change to the member’s editorial policy, which must reach us within thirty days. Third, an ownership change, in which case the new owner signs rather than assuming a commitment made by people who have left. Any of the three can warrant an off-cycle sample.
Failure has a ladder rather than a cliff. An unanswered check suspends a listing at thirty days and removes it at sixty, and the entry returns as soon as someone replies. A confirmed breach ends the listing that week, and the organization is free to apply again in six months, explaining what changed. Neither action is announced publicly, since we would rather correct a list than run a public shaming.
- An email to the named contact requesting re-confirmation, answered by a person.
- A fresh sample of recent published work, selected by us.
- An inquiry to establish that the badge still links to the live directory entry.
- A review of the public policy page for any edits made since last year.
- The entry’s date updated so readers can see when it was last verified.
Articled Approved Organization Requirements FAQs
Completely, with no exception for minor use. Humanizers serve a single purpose, which is helping machine text slip past a detector, so their use is evidence both of a breach and of an intent to conceal it. It is the only provision where a first confirmed instance ends a listing without a middle step.
The clauses themselves are assessed annually, and members are given thirty days of notice before a change comes into effect. If a clause is amended to introduce a stricter provision, members unable to comply may leave with their record intact rather than be removed. Compliance is assessed annually too, on a separate cycle.
Yes. Machine translation produces sentences, which means a translated article is a written article. A person must perform the translation, or a translator must rework the machine text to the point of genuinely being its author. Publishing a raw model translation as your own text is a breach.
Reading a generated summary is not an act of writing, so we do not restrict what members read. Two lines get crossed: pasting generated text into a draft, and leaning on a summary in place of a source you never opened. The latter is a sourcing problem this standard does not fix, though your audience will notice it.
No. This is a text standard. It does not address generated images, generated audio, generated video, or generated software. For those claims, members require another mechanism. We opted not to expand, because we cannot verify what we cannot sample, and we would rather run a narrow standard well than a broad one poorly.
Yes, as transcripts that have been labeled as such. Automatic transcription captures speech rather than composing it. What is not permissible is feeding the transcript to a model and publishing the generated text. If a person writes up the interview using the transcript as notes, that is ordinary journalism and it is fine.
Read the clauses before you sign, not after. Most organizations that have trouble with the standard have trouble with one specific clause, usually the authorship record or the thirty-day notice. Both are much easier to build now than to retrofit halfway through a check.
Content a person actually wrote
$10 per 100 words, and the writer keeps all of it. Our 1% sits on top, 0.5% goes to trees, and no generated text appears anywhere in the process.