You Paid for Content. Did You Get AI Output?
If you commission writing from an agency, a freelancer, or a contractor, the risk is not the invoice. It is publishing thin generated copy under your own brand. Here is how to check what you received.
The economics of commissioned content changed quietly and completely. A writer who used to spend six hours on a 1,500-word article can now produce something structurally similar in twenty minutes. Most of them have not lowered their rates, and there is no particular reason they would tell you.
Sometimes that is fine. A writer who uses a model to draft an outline and then does the actual thinking is doing the job. What you are trying to catch is the other case: a deliverable that was generated, lightly reformatted, and invoiced as original work.
Why this is your problem and not the contractor's
The instinct is to frame this as getting your money's worth. That is the smaller issue.
It publishes under your name. Commissioned content goes on your site, with your brand attached. Whatever is wrong with it becomes a reputational cost you carry, long after the contractor has cashed the payment and moved on.
Search engines already discount it. Google's guidance targets unhelpful, mass-produced content regardless of who or what produced it. Thin generated copy is a bad investment even in the case where no reader ever notices. You are paying for an asset that does not appreciate.
Sources may not be yours to use. This is the one that escalates. If a deliverable contains substantial passages lifted from a source you have no license to, the exposure is legal rather than editorial, and it attaches to the publisher.
What to actually check
Three checks cover the realistic failure modes, and they run on the same pasted text.
- AI detection, for whether the prose reads machine-generated. Read the sentence-level report rather than the overall number. Light AI assistance and a wholly generated draft look quite different at the passage level.
- Plagiarism, for overlap with existing published sources. This returns the matched sources and highlights the overlapping text, which is the part that matters if you ever need to show someone.
- Fact-checking, for claims that cannot be supported. Each claim comes back labeled with a verdict and source links, rather than folded into a single score.
Running all three on something you already published is also worthwhile, particularly the plagiarism check. Anything already indexed is the highest-exposure content you own.
Put it in the contract first
A report is far more useful as evidence against a clause than as an argument in its absence. If originality matters to you, say so in writing before the work starts, and say what you mean: no wholly generated drafts, disclosure of AI assistance, no fabricated sources and citations, whatever your actual standard is.
Without that, a flagged deliverable puts you in a debate about whether a probabilistic score proves anything, which is a debate you will lose, because it does not. With a clear clause, the same report becomes documentation of a specific breach.
Reading the result fairly
A high score is a signal, not proof. Detection misfires in known ways: on short text, on heavily-templated formats, and on writing by non-native English speakers. A contractor who writes clean, plain, efficient prose can get flagged honestly, and treating that as caught-in-the-act will cost you a good writer.
The better move is to share the report itself. Every saved report has a read-only share link, so you can send the specific flagged passages and ask about those, rather than paraphrasing a number at someone. A writer who did the work can talk through it. That conversation resolves most cases faster than any score will.
To check something you have been sent, the content verification page runs a free check with no account. If you are on the other side of this and need to show a client that your own work is yours, proving you wrote it covers that instead.