Managing content for multiple sites and the verification question comes up constantly. not just “is this ai” but “how confident are we and what do we do with that information”
currently my workflow is:
- read it myself first (gut check - does it sound human)
- run through 2 different detection tools and compare
- if results conflict, check with the writer directly
- look for specific tells: overly balanced paragraphs, perfect structure, lack of personal voice
but this takes forever and its not scalable when youre reviewing 20+ pieces a week. and honestly step 1 is still the most reliable part which is kind of pathetic
curious what others are doing. anyone built a more efficient system?
my workflow is similar but ive added a few things that help with scale:
- i ask writers to submit their google docs link (not a copy). i check revision history before anything else. real writing has messy revision patterns - deletions, rewrites, sections moved around
- for repeat writers i compare their “voice fingerprint” across articles. if someone suddenly writes completely differently its a flag
- i only run detection tools as a tiebreaker, never as primary evidence
the revision history check alone catches like 90% of issues and takes 30 seconds
revision history is key. as a teacher ive been requiring students to write in google docs with sharing enabled since last year. not because i dont trust them but because it protects everyone. students who write honestly have proof, and the revision patterns for ai-pasted content are obvious (copy paste, minor edits, done)
for content thats already submitted without revision history its much harder. i basically triangulate: does it match their previous voice? are there specific details only they would know? does the structure feel natural or templated?
At my agency we’ve basically abandoned automated detection as a primary tool. our workflow now:
- writer briefs include mandatory research links they must reference
- we check that references are real and correctly cited
- content must include at least 2 original insights not found in top 10 serp results
- any flagged content goes to a senior editor for manual review
its more work upfront but the content quality is also way higher. treating every piece as suspect until proven human actually raised our overall editorial standards
@SilentBean64 the voice fingerprint thing is really smart. Never thought to formalize it but I have enough writer history to do that