had something wild happen last week and need to know if this is becoming a trend
delivered a website redesign project - full copywriting included as part of the scope. wrote everything myself, researched the clients industry, multiple revision rounds, they approved everything. project done.
then yesterday i get an email: “we ran your deliverables through an AI detector and several pages flagged as potentially AI generated. can you explain?”
i was stunned. the copy wasnt AI generated. i write in a clean professional tone because thats literally what they hired me for. but apparently clean professional tone = “suspicious” now
so i started looking into how ai detectors work and honestly? its not reassuring. these ai detector tools basically analyze statistical patterns in text - word predictability, sentence structure consistency, vocabulary distribution. the problem is that good professional writing ALSO has these patterns. because good writing is supposed to be clear, structured, and predictable
i tested my own copy through a few detectors. results were all over the place:
- one said 85% human
- one said 62% AI
- one said “mixed”
which raises the question: can ai detectors be wrong? absolutely yes. they disagree with EACH OTHER constantly. if you want to check if something was written by ai, which detector do you even trust? theres no reliable ai detector that gets it right 100% of the time
what is the best ai detector if you actually need one? honestly i dont think any of them are reliable enough to accuse someone of using AI. the technology just isnt there
anyone else had clients pull this? how did you handle it?
This happened to a designer friend of mine last month. client ran their deliverables through originality.ai and it flagged like 30% of the copy. she showed them her google docs history with every keystroke visible. they apologized but the damage to the relationship was done
how to check for ai writing reliably? you cant. thats the honest answer. these tools are probabilistic, not definitive
From a technical standpoint: how do ai detectors work is actually pretty simple. they analyze perplexity (how predictable each word is given the previous words) and burstiness (variation in sentence complexity). AI text tends to have LOW perplexity and LOW burstiness. but so does well-edited professional copy
the fundamental flaw is that these ai detector tools are trying to reverse-engineer the writing process from the output. thats statistically possible in aggregate but unreliable for individual texts. can ai detectors be wrong? they are wrong roughly 10-20% of the time depending on the tool
Had a similar experience with packaging copy i wrote for a client. they ran it through some detector (dont remember which) and questioned me about it. i told them straight: if you dont trust my work, theres no point in our working relationship. they backed off
you need to set boundaries. include a clause in your contract about this - something like ‘all deliverables are original human-written work. AI detection tools are known to produce false positives and their results are not grounds for dispute’
This is terrifying as a junior freelancer. im already anxious about client relationships and now i have to worry about false AI accusations on top of everything else??
if someone asked me what is the best ai detector id say none of them are reliable enough to accuse someone. ive tested my own handwritten college essays and they flag as partially AI. the whole space is broken
The irony is that clients who use AI detectors on freelancer work are often the same clients who ask you to “make it sound more professional” which is exactly the kind of writing that triggers these tools
my approach: i now include process documentation (drafts, notes, revision history) as part of my standard deliverables. not because i should have to prove anything but because it preemptively shuts down this conversation
update: had a call with the client. walked them through my writing process, showed my research notes and drafts. they were actually pretty receptive once i explained how ai detectors work and why false positives happen
the real issue wasnt distrust of me specifically, it was their legal team requiring AI checks on all outsourced content. so now theres a corporate policy driving this. expect to see more of this honestly
@MaxFlare83 the contract clause idea is brilliant, adding that to my template immediately