Screen freelance copy & agency deliverables for AI
The problem
Content teams paying premium rates for human copy often receive low-effort AI drafts. The risk is not only tone: search engines demote scaled unedited AI content, and an AI-heavy piece under a named byline is a credibility event.
Screening after publication is too late. The cost has already been paid twice, in the invoice and in the correction.
How Stipple helps
Stipple screens drafts at intake, before publication or invoice sign-off. Style analysis flags AI-heavy passages and the tells trace shows exactly which sentences carry the signals.
Editors send specific revision notes instead of a bare score. Contractors get actionable feedback; managers get a consistent, documented intake gate.
The call
View Full Recipe →curl -X POST https://www.stipple.sh/v1/detect-ai-text \
-H "Content-Type: application/json" \
-d '{"text": "Draft agency submission..."}'Maintain brand authority and ensure clients get genuine human craft.
What the result looks like
A per-document probability with a confidence band, flagged passages with the specific tells, and abstention on short snippets.
A flag is a revision request with evidence attached, not a verdict about the contractor. A clean result means no AI-style signals fired in the prose analysed — it is not proof a human wrote it.
Provider watermarks need the provider’s own unreleased detection API, so for freelance screening today style analysis is the operative check. The EU AI Act’s transparency duties make it prudent to ask contractors about AI use directly rather than infer it.
How the corpus was built, what was caught, and what the method does not prove.
Questions
- How do I screen freelance copy for AI?
- Screen at intake: run the draft before acceptance, or call the API from your CMS workflow. The tells trace gives sentence-level findings you can send back to the writer. Pair the result with your contract's AI-disclosure clause, because the tool measures style and the contract governs conduct.
- Will search engines penalise AI content?
- They reward helpful content and demote scaled, unedited, unhelpful output. So the practical screening question is not ’was AI involved’ but ’is this specific, accurate and edited’ — a detector flag plus human editorial review answers that combination better than either alone.
- Can this prove a writer used AI?
- No. Style analysis produces a probability with evidence, and paraphrasing or humanising lowers any style score. Provider watermarks, where they exist, are verifiable only through the provider. Treat results as an intake signal and a conversation starter, with the contract as the leverage point.
- What if the writer admits partial AI use?
- That is usually fine. The question is whether the deliverable is accurate, original in voice, and edited to your standard. Screen for quality signals, treat AI use as a contractual matter, and keep the report with the invoice as an audit trail.