Manual editing.
Unless it's for robots, not people.
Why:
1. Paraphrasing tools make your text a mess. Synonyms have different connotations and meanings. Paraphrasing either goes for the closest one (EXCITING results turn into FASCINATING results) but it's the same thing and AI can do that just as well. It's the next most probable word. It definitely lowers the AI score in cheap or free detection services but it does not alter the structure of the sentence. It's still really obvious. However, if you go mess with the "temperature" of it, you get awkward results (think COMPELLING results or INVIGORATING results). Totally out of tone for most.
2. AI detection is based on sentence structure just as much as it is on language. ChatGPT is an absolute sucker for complex-compound sentences with multiple subjects and an adjective for each subject. And for passive voice. God I hate passive voice. And since syntax is an irrational-ass mess of linguistic rules, it's pretty much impossible to automate. Unless you have a really nicely crafted prompt running in several iterations, but it's a different story.
So, manual editing is the ONLY way to go so far. At least if you really want it to be UNDETECTABLE, even for a trained eye.
Source: making AI content look less like AI content was (and partly is) a major part of my work, I've trained people to do it and crafted several guides precisely on this topic lol.
So. Set your goals. Like, really
is absolutely right. WHY do you want it to be harder to detect? (not saying undetectable because someone with an eye for it, like, say, me, will detect it anyways, especially in long-form posts).
Is it for Google, because you're afraid of being deindexed and deranked once they find out your articles are AI-generated? Or you don't want to be called out for using AI content by your audience?
Unless it's for robots, not people.
Why:
1. Paraphrasing tools make your text a mess. Synonyms have different connotations and meanings. Paraphrasing either goes for the closest one (EXCITING results turn into FASCINATING results) but it's the same thing and AI can do that just as well. It's the next most probable word. It definitely lowers the AI score in cheap or free detection services but it does not alter the structure of the sentence. It's still really obvious. However, if you go mess with the "temperature" of it, you get awkward results (think COMPELLING results or INVIGORATING results). Totally out of tone for most.
2. AI detection is based on sentence structure just as much as it is on language. ChatGPT is an absolute sucker for complex-compound sentences with multiple subjects and an adjective for each subject. And for passive voice. God I hate passive voice. And since syntax is an irrational-ass mess of linguistic rules, it's pretty much impossible to automate. Unless you have a really nicely crafted prompt running in several iterations, but it's a different story.
So, manual editing is the ONLY way to go so far. At least if you really want it to be UNDETECTABLE, even for a trained eye.
Source: making AI content look less like AI content was (and partly is) a major part of my work, I've trained people to do it and crafted several guides precisely on this topic lol.
So. Set your goals. Like, really
is absolutely right. WHY do you want it to be harder to detect? (not saying undetectable because someone with an eye for it, like, say, me, will detect it anyways, especially in long-form posts).
Is it for Google, because you're afraid of being deindexed and deranked once they find out your articles are AI-generated? Or you don't want to be called out for using AI content by your audience?
