Key Takeaways
- If you want to make ChatGPT's tone more human-like, use ChatGPT Plus subscription to switch to GPT-4 rather than GPT-3.5 Turbo. You'll also have access to the "Humanize" plugin which will also make the text more natural.
- If you don't have ChatGPT Plus, you can simply ask for more humanized text, or prompt ChatGPT to imitate a specific author's voice.
By default, ChatGPT has a voice and writing style that might not come across as completely natural or could even be flagged by AI text "detectors," but you don't have to settle for the default tone and style ChatGPT uses.
We're using a popular AI text detector here as a way to demonstrate that these methods make a real difference to the text, but it's important to note that AI text-detection tools are not a reliable way to tell if the text is AI-generated or not . These tools have fundamental problems that make them unsuitable for determining whether text is human-written. On top of this, while text may pass a detector, it might still obviously read as nonsense to a human reader. So use your own judgment on whether the humanized output is good enough for your needs or not.
Method 1: Use ChatGPT-4
If you're a ChatGPT Plus subscriber, you'll have access to GPT-4, which is much more capable and sophisticated than the standard GPT 3.5 Turbo model that's on by default. Just left to its own devices, GPT-4 produces much more natural text, and in my testing using the popular GPT Zero service is much less likely to trigger a result that's confident of AI text. Here I've asked GPT-4 to write an opinion piece on why manual transmissions are better than automatic (because they are).
In this case, on the second re-roll of the prompt, we get a good result and manually editing the few low-perplexity sentences is no big deal.
Not sure what that means? With AI text detectors, you'll often run into the terms "perplexity" and "burstiness." These are two specific measures used to estimate whether a piece of text was generated by an AI. Perplexity is a measurement of the randomness in the text, and burstiness is a measurement of how much perplexity itself varies.
The detector uses the results of these two measurements to estimate how likely a text is to be AI-generated. Different AI detectors might use different algorithms or measurements, but the principle is broadly the same. If we put the text from this article into GPT Zero we get the result "Your text is most likely human written" (phew!) and this is what the perplexity and burstiness scores look like.
Method 2: Use The "Humanize" Plugin
This is another method only available to ChatGPT Plus subscribers. Using the third-party "Humanize" plugin, you can process text generated by ChatGPT so that its vocabulary, cadence, and other text attributes are more human-like. We just ask ChatGPT to use the plugin on our text.
However, plugins are still a beta feature, and at the time of writing, the Humanize plugin seemed to have issues, although I've had good results with it in the past. There are also many other plugins in the store (with more added frequently) that promise to rephrase or otherwise humanize text for you, so have a browse of what's on offer.
