AsstModerator
FPA Forums and Reviews Admin
- Messages
- 5,748
How can company representatives avoid false positives for AI usage in rep replies?
How was the problem noticed?
In 2023, I started noticing text generated from Chat GPT and other AI bots in many posts to the forums and also in some newly submitted reviews. AI-written text is often used for SEO purposes and can be produced in unlimited quantities. Penalties from Google and other search engines were inevitable. Because of this, the FPA began seeking good quality AI checking software. Google has recently announced the beginning of penalties for sites with large amounts of AI generated text.
While testing out the two best AI checkers we've found so far, I entered what looked like human text from a company representative's reply. It came back positive for AI. The majority of representative comments I checked came back positive. One company admitted to using an AI generator to help. The others said they wrote the responses themselves. What was going on?
Why are so many human-written replies from reps being detected as positive?
Chat GPT and other AI bots learn from human examples. The wording of representative comments on reviews at the FPA and elsewhere is surprisingly similar between companies. For a good review, the review is thanked and appreciated. Often, some feature offered by the company is mentioned and then the reviewer is invited to contact the company if any issues arise later. Negative reviews are of several types...
1. The client can't be identified by the information left. Please make contact.
2. Expressions of regret at inconvenience, followed by a contact request or a statement that the issue is resolved.
3. Explaining that the company followed procedure correctly.
4. Accusing the client of being a cheater or a fake reviewer.
Originality.ai calls this type of text formulaic and lists it as one of the most common reasons for false positives. I could take some of the responses given by forex companies and swap them to reviews for different companies and people reading the reviews wouldn't notice in most cases. Don't worry, I won't actually do this. It's just scary that a good review comment could easily be applied to good reviews for a hundred other companies. Bad review comments are a little more varied, but still could be swapped to many other reviews.
How to avoid false positives?
If you get a note from me or from some FPA system that a comment is showing up as AI, don't just rephrase it. As an experiment, I reworked an AI written review over and over again. I managed to get it from 100% AI down to about 97 or 98% AI. Instead, you need to break away from the formula. You need to write something new and different.
Currently, the FPA relies on the AI detectors from CopyLeaks.com and from Originality.ai. At minimum, you can get the free version of the CopyLeaks app for your browser. This will let you do an initial check before submitting your comment. To be more certainm you should also give it the latest scan from Originality, the 3.0 Turbo scan.
Why should I bother?
Most of you reading this want your comments easily found by search engines. You also want your review pages found too. Minimizing AI content and formulaic content that detects as AI will make both of these easier.
For anyone who gets the bright idea of using AI generated bot text to harm the SEO of a review page with bad scores, you are too late. Every new review has been manually checked for several months and the automatic system is already online. Rep comments are still being manually checked, but those will start being scanned automatically soon.
What about existing comments?
I've been contacting companies one at a time over new comments that show up as AI. Once the process is a little more refined, I'll be asking for replacement text for existing comments that show up as positive for AI. For now, there is no way for a representative to submit a replacement, so this will be me or one of my team members contacting the representative via email or PM and manually replacing the text. You can also use the form below to submit a replacement, but please don't do this unless the FPA has notified you that it needs to be replaced.
A Temporary Solution that I don't recommend.
If you REALLY want me to approve a comment or replacement comment that is showing up as AI, it can be approved, but...
A. The FPA is working to make those comments require a second click to open. This will make them less likely to damage SEO, but they will be less likely to be seen by readers, including the person who left the review.
B. Your review page may take some SEO damage in the meantime. If the damage gets too bad before we have a way to shield those comments from search engines, it may be necessary to temporarily remove them from public display.
Related Questions
Other company representative questions? CLICK HERE
Other review questions? CLICK HERE
This form is only for company representatives replacing comment text that detects positive for AI with text that does not detect as AI positive. Please make sure to check the new text before submitting it. Please also include the name and date on the review so that the replacement text will be matched with the correct review.
Please note that the form above is only for company representatives.
How was the problem noticed?
In 2023, I started noticing text generated from Chat GPT and other AI bots in many posts to the forums and also in some newly submitted reviews. AI-written text is often used for SEO purposes and can be produced in unlimited quantities. Penalties from Google and other search engines were inevitable. Because of this, the FPA began seeking good quality AI checking software. Google has recently announced the beginning of penalties for sites with large amounts of AI generated text.
While testing out the two best AI checkers we've found so far, I entered what looked like human text from a company representative's reply. It came back positive for AI. The majority of representative comments I checked came back positive. One company admitted to using an AI generator to help. The others said they wrote the responses themselves. What was going on?
Why are so many human-written replies from reps being detected as positive?
Chat GPT and other AI bots learn from human examples. The wording of representative comments on reviews at the FPA and elsewhere is surprisingly similar between companies. For a good review, the review is thanked and appreciated. Often, some feature offered by the company is mentioned and then the reviewer is invited to contact the company if any issues arise later. Negative reviews are of several types...
1. The client can't be identified by the information left. Please make contact.
2. Expressions of regret at inconvenience, followed by a contact request or a statement that the issue is resolved.
3. Explaining that the company followed procedure correctly.
4. Accusing the client of being a cheater or a fake reviewer.
Originality.ai calls this type of text formulaic and lists it as one of the most common reasons for false positives. I could take some of the responses given by forex companies and swap them to reviews for different companies and people reading the reviews wouldn't notice in most cases. Don't worry, I won't actually do this. It's just scary that a good review comment could easily be applied to good reviews for a hundred other companies. Bad review comments are a little more varied, but still could be swapped to many other reviews.
How to avoid false positives?
If you get a note from me or from some FPA system that a comment is showing up as AI, don't just rephrase it. As an experiment, I reworked an AI written review over and over again. I managed to get it from 100% AI down to about 97 or 98% AI. Instead, you need to break away from the formula. You need to write something new and different.
Currently, the FPA relies on the AI detectors from CopyLeaks.com and from Originality.ai. At minimum, you can get the free version of the CopyLeaks app for your browser. This will let you do an initial check before submitting your comment. To be more certainm you should also give it the latest scan from Originality, the 3.0 Turbo scan.
Why should I bother?
Most of you reading this want your comments easily found by search engines. You also want your review pages found too. Minimizing AI content and formulaic content that detects as AI will make both of these easier.
For anyone who gets the bright idea of using AI generated bot text to harm the SEO of a review page with bad scores, you are too late. Every new review has been manually checked for several months and the automatic system is already online. Rep comments are still being manually checked, but those will start being scanned automatically soon.
What about existing comments?
I've been contacting companies one at a time over new comments that show up as AI. Once the process is a little more refined, I'll be asking for replacement text for existing comments that show up as positive for AI. For now, there is no way for a representative to submit a replacement, so this will be me or one of my team members contacting the representative via email or PM and manually replacing the text. You can also use the form below to submit a replacement, but please don't do this unless the FPA has notified you that it needs to be replaced.
A Temporary Solution that I don't recommend.
If you REALLY want me to approve a comment or replacement comment that is showing up as AI, it can be approved, but...
A. The FPA is working to make those comments require a second click to open. This will make them less likely to damage SEO, but they will be less likely to be seen by readers, including the person who left the review.
B. Your review page may take some SEO damage in the meantime. If the damage gets too bad before we have a way to shield those comments from search engines, it may be necessary to temporarily remove them from public display.
Related Questions
Other company representative questions? CLICK HERE
Other review questions? CLICK HERE
This form is only for company representatives replacing comment text that detects positive for AI with text that does not detect as AI positive. Please make sure to check the new text before submitting it. Please also include the name and date on the review so that the replacement text will be matched with the correct review.
Please note that the form above is only for company representatives.
