At Topcontent, we are closely following and evaluating the latest language models. We have done this the last few years and we are very impressed by their abilities. We are also very aware of their limitations. At the moment, AI is helping a lot of people create text they would otherwise not be able to create. But a good professional writer will still outshine AI. Especially when it comes to facts and new information.
We do understand that a lot of our clients are worried about what AI will mean for their SEO businesses and whether we can guarantee that text from us is still written by people.
To address the second question: Yes, all text ordered from us is still written by people.
Having said that, we believe the new text-generating AI tools will completely change how SEO is done and we are soon going to be releasing a suite of products that will combine the great parts of AI with the efforts of people. – We are living in exciting times.
You will still be able to order human-made content. Which brings us to how we ensure our content is made by people.
How we ensure human-made content
The big question on everybody’s mind is: How can we detect if a text was written by a person or generated by a machine. Unfortunately, the AI detectors available are not good enough alone. They result in lot of false positives (read our article on the topic here).
We have a massive amount of text in our archives and we have used it to run extensive tests. We used this text archive together with similar texts we generated using OpenAI. When looking at the results of the test, we realized that detectors can be used as an indicator but that we needed more.
What we came up with was a process to determine if a writer was using AI. The first step is the AI-detection tools.
1. Machine Or Human Test Algorithms
There is a wide variety of these algorithms circulating on the internet. We have tried several of them and picked the best ones. These algorithms work by analyzing probabilities of words (or more accurately: Tokens) and the tone of voice. These algorithms catch AI-generated content but also result in a lot of false positives. We use the results from these checks as indicators for further investigation.
2. User Behaviour Data
To fully determine if a writer has used AI to generate content, we look at the user behaviour data. In our mind, this is the only way to make sure the text is written by a human.
We have a huge amount of writers creating wonderful text in our system. They use our built-in editor which enables us to track and store how they do it. We call this the user behavior data. We screen this data to find abnormalities and are able to detect when writers add machine text to the system instead of writing it themselves.
Again, this is the best way to spot AI generation.
3. Strong deterrence
When we have determined that a writer has uploaded machine-generated content, we close their account and freeze their pending earnings. They will no longer be able to work for us. All writers who sign up are made aware of this. This has a strong deterrent effect.
Our end goal is to only have writers that write, not writers who use AI to generate text.
We want to be open and transparent with how we work to ensure that the text we delivered are written by people. Since the AI-check tools are all brand new and not yet perfect there is a lot of false positive checks at the moment. That’s why we have developed our three step process.