How Can We Help?
< All Topics
Print

AI Detection 1.0

For a more advanced version of AI Detection, including your BEST TARGETS (based on competition and other tested trends), check out Editor!

Ensure your content isn’t AI-watermarked with our precise AI checker. Trained on billions of words, it forecasts if your content is human or machine-generated. Be aware of AI detection scores for your material. Quickly discern if the text was written by humans or from an AI like Chat GPT, GPT-3, etc.

To start, input your content and find out in seconds if it’s AI-generated. Our Chat GPT detector goes deeper than a generic classifier to detect AI writing. 

Your content will be represented with a background green highlight of each word if it is in the top 10 predictable, Yellow for top 100, Light blue for 1000, and Violet if 1000+. Words in the Top 10 are highly predictable and should be changed to a word greater than the top 10. Use the dropdown menu to find other words or use synonyms that are not in the top 10 to improve your content.

First we offer an overall Predictable percentage.  The higher this percentage is, the more likely the content was AI generated.  

We follow the GLTR model.  Your content will be represented with a green highlight of words if they’re in the top 10 predicted, yellow for top 100, light blue for 1000, and violet otherwise. 

Test your texts to compare machine-generated vs human-generated writing. Hint: click on words for more info such as the words top k position as well as 20 other words in the top 20, it’s NLP score and Neural Matching Score (if a Keyword report has completed).

Patent Pending analysis of text are 2 more metrics unique and only on KeywordSpy.

Average Predictability Length – This counts the 2 or more occurrences of top_k 10 predictable words in a row and averages out over the entire content how many top_k 10 words you have.  The lower this is the better, however, 

Worst Predictable Length – This metric is a very clear indicator if your content is AI generated because for a human to use 15 or more top_k 10 predictable words in a row is highly improbable and only an AI could or would use more than 15 top predictable words in a row.

Note: When you run a Keyword Report in SEO Co-Pilot, you will also get your top 20 SERP competitor metrics to ensure you’re in the optimal zone for predictable text.

Let’s break down the metrics in more detail:

  1. Avg Predictability Length of tokens in a row of the topk 1st – 10th most likely replacements:
    • This metric refers to how many tokens, on average, in a row are among the top-k most likely tokens predicted by the model. For example, if the metric is looking at the top 10 (topk) likely tokens, and you get an average predictability length of 5, it means that, on average, there are sequences of 5 tokens in a row that the model would have predicted as being among the top 10 most likely tokens. Longer sequences might indicate that the text is machine-generated. Our data shows you need to be under 4.1 tokens in a row on average or SpamBrain might nab you.
  2. Worst Predictability Length of tokens:
    • This is the longest sequence of tokens in the text that are among the top-k likely predictions of the model. A very long sequence might be an indicator that the text is machine-generated since it follows the model’s predictions very closely. The data for this is all over the place thus far, everyone has a very predictable “contact us for…” sentence or two.
  3. Predictable:
    • This is percentage based indication of whether the text is believed to be machine-generated or not based on the GLTR’s analysis. If the text has many sequences of tokens that align closely with the model’s top predictions, it might be labeled as “predictable” or machine-generated as per how many tokens in the whole document were in the topk 1st – 10th likely replacements. Our data shows to be 71% or under for this as well.
  5. People who hit those numbers are much more likely to rank better. There are other numbers to hit as well including Content Score being 1-10 pts under avg comp (in the green), and number of EMQs in url (usually none), h1-6 and body respectively (check your competition and match them (in the SEO tab on the top left). You will rank better if you also watch the “spread” RED missing single keywords, especially those that are scored highly (check the “Content” tab on the right).
Table of Contents