Back in 2022, when there still wasn't really an open, general-purpose AI writing tool, I was at the start of a six-month, large-scale content production project, working with a tool that was then called Jarvis (later renamed Jasper, before it shifted its whole service direction) and a team of around seven editors. In those days, content was only ever checked for copying, as in whether passages had been lifted from other sources.
The very next year, in 2023, things started to change. ChatGPT had recently gone mainstream, but it was still early days for producing long, high-quality content. We were using the same tool again for another project, and it was during this period, when AI content production was just taking off, that we first ran into AI content detectors. Our client had set a requirement that a certain share of the content (at least 80%) had to pass these detectors. So we tested many of the tools that appear on this very list. They all claimed to run different detection mechanisms, but in practice, especially the ones that came out in 2023, they were mostly checking surface-level structural patterns like punctuation and spacing regularities, and to this day most of them still run on roughly the same logic. That made it very easy for editors, or for other tools, to game them.
By July 2026, LLMs like Claude 4.8 and Fable 5 have become almost flawless in this area. You can now tell your assistant to write in your content like a native speaker, at a C1 level, to be excellent but leave small imperfections, and it will produce content accordingly. Google and others now assess whether content is AI-written using systematic, data-driven signals. And yet, just as in the early days, these tools are still very easy to manipulate, and their reliability remains an open question.
One distinction is worth drawing out here. Tools built for long-form work such as scientific papers, legal documents, and theses don't evaluate a single 1,000 to 2,000 word article. They assess pages of text against factors like semantic coherence and overall structure, which is exactly what sets this class of tools apart from the rest.
