The Articled AI Detection Disclaimer
Detector scores provide an estimate and not sufficient evidence. The purpose of this page is to explain Articled’s perspective on detector scores, how we think they are misapplied, and the rationale for why a flagged draft initiates a conversation with the writer and is not an indictment.
Why the AI Detection Disclaimer Matters
A delivery containing a report handed over without context could be perceived as settling a question with a percentage. This is not the case. Each of the tools included in a delivery runs a statistical model based on samples of both machine and human generated writing. The statistical model then attempts to classify the writing samples as either machine or human generated writing. A guess containing a number is still a guess.
This is important because these scores are used. A content lead sends a draft to a legal department. A department sends a submission to the academic integrity panel. An editor makes a decision on a freelancer. If we allow any of the scores to stand for evidence in any of these situations, we are supporting a decision that we believe the tool is unfit to support.
This page is referenced in our delivery emails and acts as much of a binding contract for us as it is an informative page for you. No page on this site sells a detector result. No writer on this site fails an assignment because of a tool’s response. When we say that we cannot tell by the score, we say that, rather is say something that would be of more confidence.
- Show who wrote the document, when, with what assistance.
- Differentiate careful ordinary writing from generated writing.
- Remain consistent across tools, settings, or a vendor’s next model update.
- Withstand being applied to a short passage rather than a complete draft.
- Replace the notes, sources, and drafts that document true provenance.
The strongest claim we will make
A named author under a no-AI agreement produced this draft, and we can show the notes and sources. The tools we used agree. We will never say a report proves it.
What the AI Detection Disclaimer Says About False Positives
Detectors mainly analyze predictability. Text comprising appropriately selected commonly occurring words in standard word order brings out the machine in text for them, as does a text with a meandering structure. For a variety of good human-centered writings, predictability is valued, including: a procedure, a compliance summary, a set of FAQs, a specification page. The tool has no way to distinguish between writing that was intentionally kept simple and writing that was automatically drafted.
One conclusion must be voiced directly as it would be to a friend. As most non-native speakers of English use more simple language and less variation in sentence structure, they tend to be detected by automated systems writing assistance software more than native speakers do. This phenomenon is considered one of the better documented problems with machine assistance systems. If we claim it does not exist, this discussion would lose all significance.
A flag on an Articled draft opens a review, not a case. An editor reads the draft side-by-side with the author’s notes and sources and their previous drafts. Rather than sending an accusation, we call the writer and ask how the passage was created. Then we check on a different tool. A score has not ended an engagement here, and it won’t in the future.
- Short drafts in which the model has too little text to evaluate.
- Writing by non-native English speakers.
- Procedures, specifications, policy summaries, and FAQs that use formulaic formats.
- Drafts that have undergone heavy editing for the sake of clarity, which removes the variation that the system looks for in sentence structure.
- Segments of text that contain standard and repetitive language, including standard terms, disclaimers, and definitions.
A flag is not an accusation
If a check is run on a draft of your order and comes back with a high score, you hear it from us: the score, the draft, the feedback and notes, and the author’s overview of how the piece was put together.
How the AI Detection Disclaimer Fits Our Human-Written Guarantee
The guarantee is about provenance, not about scores. Someone we commissioned wrote your piece. It involved no generative AI in the outlining, drafting, editing, research, translation and paraphrasing. That claim is verifiable by looking at the working file. A percentage can neither confirm it nor refute it.
When your detector disagrees with the report we sent you, we do not argue about whose score we think is more correct. Just send along the tool and score, and we will re-open the order. We will guide you through the brief, sources, notes, and drafts as well as the check history. If you have concerns after seeing our work, know that the two revision rounds you can request are valid for fourteen days beyond the order fulfillment. Also, the refund policy is applicable after the two revision rounds have been completed.
The reverse is the most overlooked portion of a document. A clean score does not show that you can do the work. Using a text machine and running that output through a ‘humanizer’ will fool our tools, and this is why humanizers are forbidden and why their use results in the removal of a writer’s access. Detection provides corroboration. The paper trail is what gives us evidence.
- A named writer working under a signed no-AI agreement.
- Research notes and a source list provided with the order.
- Draft histories available upon request.
- Detector and plagiarism reports are included in the email sent with your order.
- An indefinite ban on humanizers, by terminating the engagement.
AI Detection Disclaimer FAQs
Yes, often. Every tool on our list generates a probability, and, naturally, probabilities can be incorrect. The most drawing false positives are short samples, procedural writing, and text from non-native English speakers. For this reason, that one score never decides anything in our case.
No. Detection is the final step, not the foundation, of the claim we make. We know who wrote this and for what reason. We commissioned the work, and have their notes, sources and even drafts. The reports corroborate the record we have, rather than substituting it.
Tell us the tool, the score, and the excerpt your client worries about. We will send the order to be reviewed, and in the meantime, we will provide the writer’s notes and source materials. We will also conduct our own assessments. Your revision rounds and our refund policy will come into effect if the content is still deemed unusable.
No, and a supplier who does is describing something that they can’t control. Scores change when a vendor retrains their models, when settings are changed, or when the same text is pasted into different tools. We can guarantee who wrote the text every time, and we will show the steps we followed to arrive at the final text.
We have to ask this every time, and we will conclude. This time we will talk about how the piece was researched and drafted. We will also note the aspects of the piece that look unusual to the software. Reap what you sow, as we judge the writers based on their work ethics and not a single unfortunate situation.
No. Copyright transfers to you upon delivery with no acknowledgment required, regardless of what the reports state. The disclaimer governs how we interpret detection reports, and not what you own. Publish, edit, or resell the work how you would with other commissioned work.
This disclaimer explains how we employ detection tools, this does not constitute legal advice. If you require it for a contract, a procurement review or an academic misconduct process, have someone with legal training or a committee review it in accordance with your own rules. Email your request to support@articled.org and you will receive a response.
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.