Today’s digital world demands content integrity and originality. Copyleaks and KeywordSpy, two AI-powered detection tools, lead the market for plagiarism detection and keyword analysis. This article compares their features, capabilities, and pricing to help you decide which offers superior content analysis and protection. Dive into the strengths and weaknesses of Copyleaks vs KeywordSpy to find out who has the edge.
GPT-3 Generated Text
Our first test was using Copyleaks GPT-3 example text and you can see below the results. Copyleaks with 99% certainty labels the content as AI generated. KeywordSpy has an overall Predictable score of .79, but also provides two additional metrics that really drill down as to why the text is highly AI predictable and thus AI generated.
- Avg. Predictability Length of 6.33
- Worst Predictability Length of 22
What do these metrics mean? Average Predictability length is the average top_k 10 most probably words/tokens throughout the entire body of text. The higher this number, the more predictable and thus AI generated it is. Additionally, the worst predictability length is a measurement of how many top_k 10 tokens in a row are in the body of the text. A number greater than 12 to 14 is on the cusp of being predictable and therefore more likely AI generated.
ChatGPT Generated Text
Again, Copyleaks labels the ChatGPT text as 99% AI generated and KeywordSpy shows this text is even more predictable / AI generated than GPT-3. What stands apart visually, in addition to the metrics provided on the right, is the background color of the words. The green highlighted words are the top_k 10 predictable words, yellow top 100 and so on. So visually one can see where the text is actually a run of very predictable words whereas the Copyleaks only shows red that the entire text is AI generated. In the next analysis we will show you how KeywordSpy allows you to not only identify AI / Predictable text but also how to improve it / make it less AI / predictable.
AI text + Human editing
Again, we used sample text from the copyleaks website and our findings are as follows. Copyleaks detected some AI which is in red, but then only 32.9% chance that the light red highlighted text was AI generated, and text with no background it reports that it’s a 99% chance it’s Human generated text.
The dropdown menu of words with percentage numbers represents the percent predictability of that word being used next based on the words before it. Additionally, the tool allows you to see that
- Where that word ranks, e.g. is the the 3rd most probably word to be in that position based on the words before it (note: the counting is from 0-9, but it’s still the top 10)
- The probability percentage
- NLP / Score are placeholders for right now as KeywordSpy will be providing users, based on the query they are trying to rank for, an NLP and Proprietary score for each word to help them optimize their content to the nth degree unlike any other tool
As shown below, KeywordSpy also is showing less predictable text with the mix of yellow, violet and blue backgrounds mixed in a lot more than the previous examples.
So as you can see, the two tools both identified AI generated text pretty easily and where the human + AI was introduced, Copyleaks does not provide much context to really drill down where exactly it’s finding AI generated text. This is common with all the AI detection tools, provide an overall meter/score without really drilling down word by word like KeywordSpy.
Lastly, although not in the free version of our Advanced AI detection tool, premium subscribers get access to a competitor analysis of the three AI detection metrics: Overall AI predictable, Avg. predictability length and worst predictability length. We compare the top 20 competitors in your SERP so you can see exactly where your content needs to be to rank on Google.
As a fun experiment, I analyzed these last two paragraphs and indeed my own writing (yes i’m human) passes with flying colors. 🙂
