At We Write Blog Posts, we’re dedicated to providing a high level of content quality — no matter the challenge.
The recent release of ChatGPT represented the tallest hurdle yet in that pursuit.
We’re committed to producing 100% human-written content, and we’ve been discussing how to ensure that since ChatGPT was released.
While we have a very high level of trust in our writers — many of whom have been with us for a few years now — we appreciate the need for a process to verify that content isn’t being created with the use of AI.
Fortunately, we’ve come up with a solution that involves a really useful tool and careful analysis of how writers add content to their documents.
But before I get to that, I want to briefly explain why simply using one of the AI detection tools that have cropped up recently isn’t the magic bullet it seems to be.
Our initial thought when deciding how to deal with AI content was to use one of the AI detection tools that have cropped up recently.
However, our internal tests with tools that supposedly detect AI content have been discouraging, to say the least.
We found that each tool we tried — including Originality.ai, CopyLeaks, Writer.com, and ChatGPT’s AI detector — was woefully inadequate.
They all throw a ton of false positives, and they’re pretty easy to fool by changing the wording of the AI output.
Even OpenAI — the creator of ChatGPT — says their AI detector is only right 26% of the time.
They simply aren’t good enough to be trusted.
Therefore, simply telling you, “We’re running all of our articles through an AI detector tool,” and calling it a day would have been a terrible way to approach this.
Because the AI detection tools suck, our focus shifted from using those tools to evaluating how articles were created by the writers.
Because we use Google Docs to create our content, we started experimenting with looking through Google Docs version histories to check for large sections of text added within an abnormally small timeframe.
If a writer is writing articles normally — by performing proper research and typing the article word-by-word — the Google Docs version history shouldn’t show whole paragraphs or sections added in a single moment (which is what it would show if they pasted in AI-generated text).
Makes sense, right?
Well, it’s good in theory, but it doesn’t quite work.
The problem with this approach is that the Google Docs version history doesn’t provide enough detail. It only shows changes in batches that update every minute or so.
If someone is a fast typist, it could easily seem like they are pasting in entire paragraphs when they’re just typing efficiently. So simply viewing the version history wasn’t good enough.
The recent release of a new tool handed us a gift-wrapped solution.
Originality created a Chrome plugin that allows you to view exactly how content written in a Google Doc was added.
It’s like the Google Docs version history tool on steroids.
While Google Docs merely shows a rough outline of edits made within a given period, this tool shows every single character that was added to and removed from the document — all the way from document creation to article completion.
It’s pretty incredible. Here’s a gif that shows how it works:
That’s not all it does, though.
It also provides a graph that displays the character addition trends over time.
This makes it extremely easy to identify when large sections of text are added in a short period.
For example, the graph for an article written normally will look something like this:
While the graph for an article created with AI might look something like this:
Notice the vertical jumps in character count when large sections of text were pasted into the document.
While there are certainly instances where pasting in text is warranted — such as when you’re directly quoting someone — there is no situation in which pasting 2,000+ characters at the same time is okay.
We believe this tool offers the best way to detect and defend against AI usage when outsourcing, and we’ll be incorporating it into our workflow immediately.
To make sure we use this tool to its full potential and continue to provide a high level of quality, here’s what we’ll be doing:
Now that I’ve covered how we’ll protect you from AI content, I want to briefly explain why it’s important to avoid AI content in your articles.
Many people have been concerned about Google detecting and penalizing content created with the help of AI.
Thankfully, Google recently clarified their stance on AI usage in content.
You can read their full statement here, but I can easily summarize it in one sentence:
As long as the content is good, Google doesn’t care how it was made.
If the Google algorithm detects traces of AI usage in your content, it will not automatically penalize your website.
However, despite this clarification, I want to make something clear:
Using AI to generate content is still very risky.
The problem lies in the abundance of incorrect information AI-generated content can contain.
I’ve been researching AI since ChatGPT was released — both in testing it myself and reading about others’ experiences with it.
By and large, the consensus is that it presents information very confidently, but that information is often wrong.
Here are a couple examples of incorrect content:
This is why we’ve implemented such stringent anti-AI measures.
The worst thing you can do on your blog is publish inaccurate information, and using AI makes it very difficult to verify that what you’re publishing is true.
Despite the surface-level impressiveness of ChatGPT, it has deep-seated flaws and can’t be trusted to provide accurate information.
A carefully designed content creation process manned by professional writers and editors remains the ideal way to create original, high-quality blog posts.
We’ve taken great care to develop such a process here at We Write Blog Posts. Here are a few measures we take to maintain a high level of quality control.
Here’s the bottom line: we are fully committed to providing the same original, high-quality content we always have.
The proliferation of AI tools like ChatGPT does not change our mission, and we are confident that the processes we’ve implemented will prevent AI usage in our content.
Before I finish this post, I want to add a quick addendum about the documents we submit to you.
We submit all content to you on a We Write Blog Posts company Google account rather than submitting the writer’s version of the document.
This is to ensure your content remains accessible even if the writer decides to close their Google account or delete their version of the document.
Because of this, if you use the Originality plugin to analyze one of our articles, you’ll see that all of the content was pasted in at once.
This does not mean AI was used to create the article — it’s merely evidence of our editing team transferring the writer’s content to a company-hosted document.