ChatGPT launched in Autumn 22, and has been adopted faster than any other recent digital innovation, so how do you know when a piece of content written by AI and copy and pasted from ChatGPT versus when it’s being used well?
AI can be a great kick-starter, and productivity boost, especially now that we have tools for creating images, video and audio as well as text. However, it is no surprise that not everyone trusts these technologies. So, how can you tell if a piece of content has been written by AI?
5 hints that a piece of content has been written by ChatGPT
A reader will interpret a piece of content based on the context it is set in. Your site will usually talk to a specific audience, so a reader can infer purpose, culture etc. AI generated content may not always consider these factors, resulting in copy which is “off-tone” or even in some cases, offensive to the reader.
Tone and pattern
In generative AI models, you can say what tone you want your article to be written in, “friendly”, “formal” or “funny” for example. ChatGPT will adopt the instructed tone across the whole article, whereas when a human writes, they tend to add light and shade to their copy. The lack of transition words and varying tones can make the output read as a bit robotic.
Also, watch out for technical jargon – if you spot a pattern of using industry “buzzwords” this can also be a flag.
The data used to train ChatGPT was from 2021, as this was when it was being “taught”. There is therefore always a risk that it hasn’t calibrated for information updated more recently than this, so always double check your facts are current and accurate before publishing.
This is probably the simplest way to identify AI created content. When people write copy for digital use, they are often sharing an emotion (e.g. “I am excited”, “I was disappointed”) or an opinion on the topic. AI generated content tends to be more matter of fact, so you usually won’t find emotions or beliefs in the text.
Repetition or cliches
AI technology relies on pre-programmed templates and formulae, so often generates copy based on the predictable language it is trained on. This can result in use of over-used generic cliches such as “revolutionary new product” that don’t explain what is actually revolutionary about the product at all.
Also, when people use multiple questions or keywords as part of their prompts, the output can result in repetition of that phrase. Where this happens, Google sees it as “keyword stuffing” and marks you down for your SEO.
Use an AI detector
If content is suspiciously similar to other content you have read, run a plagiarism check.
AI detection tools are being developed in response to all of these concerns. They scan content and give you a probability that the piece has been written by AI, rather than a definitive yes or no. Some examples are GPTZero, Content at Scale and Copyleaks. Even Open.ai (who run ChatGPT) are getting on the bandwagon with their own AI Classifier.
Google will not mark you down for SEO for using AI to generate your content. However, it will mark you down for bad content, so look out for the flags listed in this blog when approving your copy as they are all seen as indicators of poor content – whether written by AI or not.
By all means, use ChatGPT to get started, research, rewrite, create your base content – it is brilliant and a huge time saver for these. However, never post the output without reading it. Edit and personalise the copy with your reader in mind – remember, if you are not talking to someone, you are talking to no-one.