At a glance
- Company
- Workado (formerly Content at Scale AI)Software vendor: AI writing and detection tools
- Size
- Not stated in the source
- Industry
- Agencies Agencies and publishers buy tools like this to check whether text was written by AI.
- Where and when
- United States. Complaint April 2025, order final August 2025.
- Who produced the figures
- Official recordEverything on this page comes from two announcements by the US Federal Trade Commission. The figures are the regulator’s allegations, settled by agreement.
What happened
Workado sold an “AI Content Detector”. The tool was meant to tell whether a piece of text was written by a person or by generative AI. The company advertised it as 98% accurate.
It also said the detector had been built using many kinds of writing, such as blog posts and Wikipedia entries, so that it would work for ordinary users.
The numbers
- 98%: the accuracy the company advertised.
- 53%: the accuracy on general-purpose content in independent testing, according to the regulator’s complaint.
- One kind of text. The regulator said the underlying model had been trained to classify academic writing only.
The head of the regulator’s consumer protection bureau said the product “did no better than a coin toss”.
The outcome
The company agreed to an order, made final in August 2025. It must:
- make no claim about how well a detection product works unless it holds competent and reliable evidence at the time;
- keep that evidence;
- email eligible customers about the order;
- report to the regulator after one year, and then each year for three more.
What the case shows
- A headline accuracy figure may come from a narrow test. The tool was tuned on academic writing. On everyday text it was right about half the time.
- Ask what the figure was measured on. If the test material does not look like your material, the figure tells you little.
- Run your own small test. Fifty pieces of text whose origin you already know would have shown the gap.
What this says about the practices
Practices that were missing
- Ask AI vendors for proof before you buy, and test on your own material The vendor’s accuracy figure had no adequate evidence behind it. A buyer who asked for the test results would have found none that fitted its own kind of text. Do first · Effort: medium (weeks, a few people)
- Claim only what your AI has been tested to do The company advertised 98% accuracy. The regulator said independent testing found 53% on general-purpose content. Do next · Effort: low (days, one person)