What's your take on AI generated content?

Microsoft is already planning to add ChatGPT to Bing. If Bing using ChatGPT instantly provides a great answer to your question, better and more specific than any blog post could do, why would any user ever click through to any of your sites?
All it takes is 1 wrong answer from the AI for people to go back to reading human-written content. Imagine trying to change the oil in your car, and the AI answer tells you to use the wrong type of oil, and now it has to go to the mechanic.

Regarding the quality of the AI content provided, aside from giving wrong answers, it’ll probably be fine for snippets. But long form content? No way. The same things get repeated like 90 times in a 400 word section of text all the time and you (the human) end up having to rewrite things.

The correct answer to all of the AI nonsense is simple: don’t produce garbage 1 cent per. word content, use images, videos, etc. Market your content. The brainless will be eaten but the semi-competent will stay alive (for now, of course).
 
All it takes is 1 wrong answer from the AI for people to go back to reading human-written content. Imagine trying to change the oil in your car, and the AI answer tells you to use the wrong type of oil, and now it has to go to the mechanic.

Regarding the quality of the AI content provided, aside from giving wrong answers, it’ll probably be fine for snippets. But long form content? No way. The same things get repeated like 90 times in a 400 word section of text all the time and you (the human) end up having to rewrite things.

The correct answer to all of the AI nonsense is simple: don’t produce garbage 1 cent per. word content, use images, videos, etc. Market your content. The brainless will be eaten but the semi-competent will stay alive (for now, of course).
The thing is that you think about the current chatbot. The new improved ones will be connected to the internet, it can fact check itself and check though databases faster than anyone else. Read trough science, manifacture manuals, calculate things to see what is best for this and that.

This is probably not even far away.
 
The term "AI" is going through its biggest transformation and revaluation in the public mind ever. It's going to get more controversial.
 
Bing and ChatGPT - I don't care what Bing does. All I know is I am still seeing Duck Duck Go advertise on TV, so I'm going with them.
 
I have lots of thoughts about this. There are a few key points I come back too:

1. AI content has not (yet) surpassed native-language human written content in terms of quality. Even if you feed it a perfect template, headings, and outline, a good writer will come out on top. At some point this may change but we aren't there yet. Also, ChatGPT gets a lot wrong. If "content is king", AI does not win.

2. At the point something like ChatGPT content beats handwritten content, it still doesn't do the other stuff like tables, original research, infographics, etc. or know how to properly integrate them with the output.

3. Even in a world where all content online was AI generated, the content itself is only one part of SEO. Yes, it is usually the most time-consuming part, but site structure, page structure, interlinking, backlinks, headings, building a brand, etc. etc. are all important skills involved with success.

4. I think the idea of a ChatGPT chatbot replacing Google is not going to happen. In the same way voice search hasn't taken off, people like to skim read. Interacting with ChatGPT doesn't give you that. Featured snippets might ultimately become AI generated but even then I would expect sources to still be linked to which is still a traffic opportunity.
 
I don't know if it was mentioned here before, but i found another tool for detecting GPT generated content: GPTZero.

I just checked a few samples, and it was pretty accurate when the text has enough sentences.
 
This guy fooled originality.ai simply by using the re-phrase tool in Jasper. Doesn't get simpler than that. Whether that would fool Google is however another question.

 
To me, AI content is as good as your prompts! Also, AI ought to be treated more like an assistant than leaving it to do the heavy lifting.

When Jarvis (Jasper?) came out, I used the tool to write 3-4 articles. 3 of those are still ranking in the top 3, and earning me decent affiliate revenue todate.

Here are the steps I take:
  1. Do proper keyword research
  2. Formulate an outline or ask AI to
  3. Write prompts for each section and ask AI to generate
  4. Edit the output to suit your blogging style, and add relevant stats or links to useful resources/backups
  5. Run the output through grammar editing software like Grammarly (helps reword AI's wordiness into more concise text)
  6. Publish
Lather, Rinse, Repeat!

Google only cares (when they do) that your content is valuable. How do they know your content is valuable... Well, among other things, by assessing how well your audience stays engaged on your pages.

I don't know if it was mentioned here before, but i found another tool for detecting GPT generated content: GPTZero.

I just checked a few samples, and it was pretty accurate when the text has enough sentences.
I've run 5 samples. Got only 1 right. So, this catch and mouse game will continue for a long time
 
Any thoughts on this?

"Google now intends to unveil more than 20 new products and demonstrate a version of its search engine with chatbot features this year, according to a slide presentation reviewed by The New York Times and two people with knowledge of the plans who were not authorized to discuss them."

"We’re excited by the transformational advances discussed above, many of which we’re applying to make Google products more helpful to billions of users — including Search, Assistant, Ads, Cloud, Gmail, Maps, YouTube, Workspace, Android, Pixel, Nest, and Translate. These latest advances are making their way into real user experiences that will dramatically change how we interact with computers."
Source: https://searchengineland.com/google-search-chatbot-features-this-year-391977
 
I'm very excited to see how shit this will be and how easy to exploit it is. Cannot wait to include information about how my competitors businesses are horrible and how mine is accepted as the best in the area, according to local experts of course.

I can get garbage AI content indexed and ranking. They are then going to use that same content in the corpus for training the algorithm. This can surely only deliver quality results to users

Snark aside, an enormous amount of searches each day have never been done before, because the world evolves very quickly. I see no way currently that even Google can afford to re-train the AI with up-to-date information. So, they either have to use new information without training deeply on it, making it very easy to exploit, or they can use this with out-of-date information. That limitation makes it only useful for a certain subset of queries.

I think it's more likely that this ends up like a voice search type product, where it returns you a snippet answer and a single link to the recommended source. I don't think it will look like ChatGPT where you get a precise answer, with no additional information or expectation that the user reads more.

But, maybe I'm wrong and if you get enough smart people together with unlimited resources, they might figure it out. I'm doubtful, given that search is littered with parasite spam, the .it network of malware sites still ranks in top 100 on nearly every imaginable keyword and they can't even re-write titles and descriptions correctly.
 
New AI classifier for indicating AI-written text

4425rph.gif

That just burst a ton of SEO's bubbles. I mean, ALL the other versions of A.I. were able to be detected... Hopefully nobody went balls to the wall and started firing staff and hiring ChatGPT experts to load up a ton of content...

Always, always, always assume when something is launched, there is an even more powerful version behind closed doors they aren't releasing into the public. I mean if I have crazy stuff I coded at my SAAS, others with more phDs and more resources have them as well.
 
OpenAI's tester has the same false positive and false negative problems as the rest, which is telling, though they fail at a lower rate than other testers.

The real death knell is that OpenAI has said that they're "going to" (meaning it's done) add cryptographic signatures to the hashed versions of the content, meaning the word choices and orders and syntax and grammar will be selected in such a way that the hashed versions have secret codes in them.

Good luck finding the codes, and even if you have them, you won't have the key to unlock them. OpenAI has suggested that they will share the key with the right entities, such as Google, Universities (for students trying to cheat), etc.

Finding the code already indicates that it's AI content. I'm not sure what would be encoded within that needs to be locked behind a key.

This begs the question if a simple rewrite or even just changing some words here and there will matter. First of all that's a giant waste of time. But second of all... I'm trying to come up with an analogy here.

Let's say that a river is flowing to an ocean. So you dig a new trench to flow the river in a different direction to a different ocean. Congrats, you beat the "river-to-ocean" detector, right? No, because they're not measuring which river to which ocean. They're simply measuring "volume of water flowing with kinetic energy and thermodynamic blah blah over time with 1 parts per million of undetectable radioactive dye" or whatever.

My point is you can look at the hashes like a meta-reality, an order of existence where what you're doing to the content in your word editor isn't going to effect enough to matter. You may get rid of one crypto-signature but there may be 5 others in that 1,000 word article you couldn't wipe out without a complete rewrite.

Ultimately, trying to hide the fact that you're using AI content is fruitless. What you need to assume is that Google is not going to fight this in the way we think they are. You don't fight the future, you adapt to it. Everyone will be producing AI content at a very non-negligible rate. Google, instead of deeming all AI content bad, will have to assess if it's worth the user's time, which is what they already do as it is with non-AI content. So if your plan is to use AI content, you better punch it way up and enhance the hell out of it, not just generate it and paste it in. Fact checking is one thing, but enhancing it is another.

It's very much like their stupid "above the fold" algorithm which rightly got rid of ads above the fold. But people may not recall they were also trying to punish for big splash images used as featured images, which pushed the text below the fold or too far down above it.

People basically ignored Google and kept doing their thing and Google had to adapt and get rid of that part of the algorithm. You can't fight and police the people when everyone ignores you. You only hurt the quality of your results by otherwise punishing high quality content and surfacing crap SEO's produced.

That is very much how this AI content thing is going to go, except we won't have to go through the steps of them punishing for it first. They already know how to assess quality and user satisfaction. It's business as usual except SEO's will waste their time pumping out volumes of garbage (which is no different than spam, which will be punished, and it won't be because it's AI content).
 
@Ryuzaki I believe in the guidelines it doesn't specifically say AI content is bad as long as its "helpful to the user". You've pretty much summed up everything.

I feel like history likes to go full circle.
  • All foods used to be organic and then became "conventional"... now organic came back pretending like its the new kid on the block.
  • All clothing used to be hand made and then everything became mass produced. Now hand made clothes are considered luxurious.
Perhaps non-AI influenced content will make a resurgence later on in the future and everyone's going to think its a novel thing!
 
The real death knell is that OpenAI has said that they're "going to" (meaning it's done) add cryptographic signatures to the hashed versions of the content, meaning the word choices and orders and syntax and grammar will be selected in such a way that the hashed versions have secret codes in them.

Good luck finding the codes, and even if you have them, you won't have the key to unlock them. OpenAI has suggested that they will share the key with the right entities, such as Google, Universities (for students trying to cheat), etc.

I don't buy this as a solution for Google.

OpenAI get 3 factors; hard to detect signature, many signatures, good content. They get to pick 2.

Anything that is hard to detect and numerous will significantly alter the quality of the content, particularly when the API users are tweaking all the settings to get the content they really want.

Many signatures, but good content, makes it easy to detect the signatures and so easy to remove.

And if you have a hard to detect signature but good content, you can only add a single or very few signatures into the content. Meaning that even small changes to the content will have a good chance at removing the signal.

OpenAI is going to succeed because of the quality of the output, so I don't see how they can significantly alter the algorithm. They would need a 1000x larger LLM to get enough variation to achieve all 3 at once. But, there is tons of competition out there and open source AI models are getting better. So, we'll all just move over to that instead, which Google won't get the keys too.

It's absolutely clear that this technology is not going away, it's getting better every year, and Google cannot detect AI content (done well) at scale. It's too resource intensive.

So, they are just going to have to catch the absolute garbage, and accept that most of the time I'd rather read an AI response than the ESL content that most sites are publishing right now. Ultimately, they'll just PR themselves out of this, like I generally believe OpenAI is doing and most SEOs will believe them, because they are morons.

I went ahead and tested this detector a lot since it was launched.

Here are their classifiers:

  • "Very unlikely to be AI-generated" corresponds to a classifier threshold of <0.1. About 5% of human-written text and 2% of AI-generated text from our challenge set has this label.
  • "Unlikely to be AI-generated" corresponds to a classifier threshold between 0.1 and 0.45. About 15% of human-written and 10% of AI-generated text from our challenge set has this label.
  • "Unclear if it is AI written" corresponds to a classifier threshold between 0.45 and 0.9. About 50% of human-written text and 34% of AI-generated text from our challenge set has this label.
  • "Possibly AI-generated" corresponds to a classifier threshold between 0.9 and 0.98. About 21% of human-written text and 28% of AI-generated text from our challenge set has this label.
  • "Likely AI-generated" corresponds to a classifier threshold >0.98. About 9% of human-written text and 26% of AI-generated text from our challenge set has this label.

I've run at least 100 AI responses through this and almost every single one was classified as at least unlikely to be AI, if not better. Meaning, it thinks it's more likely to be human than AI generated.

A bunch were scored as very unlikely to be AI.

On the flip side, I inputted a bunch of different articles that I've written over the years, on a range of topics. Most were scored as "unclear if it is AI written".

The most consistent way that I can get this detector to classify content as "very unlikely" to be AI is to maximize the randomness of the content, which means high frequency and presence penalties in the OpenAI API.

You know why? Because the content starts to become worse as it gets longer, since it's penalized against using the words that it wants to because of the frequency and presence penalties.

The first 1/3 is great, the second 1/3 is okay, the last 1/3 is complete gibberish like

"With these five tips under our belt hopefully now armed enough knowledge begin journey towards raising personal Credit Limits successfully Hopefully provided information here helps kickstart process necessary obtaining improved financing options overall while simultaneously providing ability protect finances best interests moving forward Best luck out there"

People are treating this like it's an easy problem to classify content. It's extremely hard and it's also very computationally heavy. I can make this classifier tell me content is "very unlikely" to be AI generated by just picking an obscure topic that it's not trained heavily on.

If I give it the EXACT same prompt, asking about Harry Potter, and then about Alipogene tiparvovec, with Harry Potter the detector says it's "likely AI", but for Alipogene tiparvovec its "very unlikely".

That means on the exact same prompt, the content for Harry Potter "corresponds to a classifier threshold >0.98. About 9% of human-written text and 26% of AI-generated text from our challenge set"

But for Alipogene tiparvovec it "corresponds to a classifier threshold of <0.1. About 5% of human-written text and 2% of AI-generated text from our challenge set".

The results could literally not be worse, it's picking the two most extreme classifications for the same prompt. But if I read the content of both, nobody would guess that one is AI generated and the other not.

I did the same for Harry Styles and Chlorophytum comosum. The former was "unclear" the latter is "very unlikely" to be AI.

Then for the Harry Styles content I just slowly turned down the temperature, frequency penalty and presence penalty. At every increment, the detector becomes far more confident that the content is AI generated.

So, I repeated the same test using Alipogene tiparvovec. Guess what? It's never confident that the content is AI even when I use the absolute lowest randomness, at which point it's basically just repeating the exact same sentence structure with different topics.

The same is also true the opposite way. You can over dial the randomness where it becomes trivial to classify it as AI, because it's so random that no human would write that way. When the content is in the middle, or dialed to be very human, it's hard to detect.

I wonder why... because it's basically just reading word by word and making a prediction as to what word could come next. Then, it's extracting factors from the data and compares it to the training data to classify and group. So, as you dial up the randomness, the content is "more human". This is also why it doesn't work well on complex topics, because when it reads "it is the first", it does not expect the next word to be "gene".

If Google dedicates a $50b per year to running these models and improving them, they can probably reliably classify content about extremely popular topics. Anything obscure, and especially something that's new, is going to be exponentially harder to detect. This completely ignores the problem of doing this on an exponentially growing data set. Oh wait, they already found a solve for that, by just refusing to index content.

I haven't even talked about countless ways that you can edit the content from the AI. Don't just copy paste it onto your sites you apes. At the very least, you can spin the content if you're really lazy.

Here's a tip, Google logit bias and think how you might use that to improve the initial output.
 
Last edited:
First, John Mueller suggests AI content is considered spam. So, short term, I think there will be attempts to de-index sites using AI content.

As far as how Google would detect? OpenAI uses a cryptographic pseudorandom function to generate text. Essentially each letter, space, quotation, etc all Tokens - are not completely random, but are pseudorandom (via cryptographic pseudorandom function). Only OpenAI has access to this key, but I could see them giving Google and Microsoft, etc, access.

Could rephrasing eliminate this (hidden to humans) cryptographic watermark? Definitely. But to what level would you have to rephrase? A whole rewrite? I'm not sure. And what about most rephrasing tools? If they are based on OpenAI API, would that defeat the rephrasing purpose, by adding more cryptographic watermarks? Again, not sure.

But Google has many other ways of detecting AI content besides this cryptographic watermark. They have and will train models by looking at large datasets of human vs ai content. There are distinct patterns machines can pick up on that humans miss. However, it does get more difficult to detect the more advanced these systems get and the bigger seed datasets.

The real question: what happens when there are other models, open-source models, where people custom train these models, advance, etc? It becomes harder and harder for google to detect. It's not like the days of content spinners, where Google easily leaped that hurdle. It is more of a cat-and-mouse game. At the same token, don't underestimate Google's abilities here.

But if everyone can create general AI text, then it is saturated. SO the only real way to rank would be to have something better than all the other "AI" content. Ultimately, the goal is the same: put in human effort to create better content. It may be that AI is just used as a tool to make this good content, building brick by brick, paragraph by paragraph, a good piece of content.

The other real question is: what happens when a general AI takes market share from traditional search engines? Fewer clicks for SEOs. What happens when a general AI is literally placed on top of traditional search results (in place of featured snippet content)? Fewer clicks for SEOs. I think this is a real trend we will continue to see even more than what we already have been seeing.

I've run at least 100 AI responses through this
I'm Curious, what detector are you using?

I haven't found any reliable ones yet. They are all different, and some seem better, but give false positives.
 
First, John Mueller suggests AI content is considered spam. So, short term, I think there will be attempts to de-index sites using AI content.

As far as how Google would detect? OpenAI uses a cryptographic pseudorandom function to generate text. Essentially each letter, space, quotation, etc all Tokens - are not completely random, but are pseudorandom (via cryptographic pseudorandom function). Only OpenAI has access to this key, but I could see them giving Google and Microsoft, etc, access.

Could rephrasing eliminate this (hidden to humans) cryptographic watermark? Definitely. But to what level would you have to rephrase? A whole rewrite? I'm not sure. And what about most rephrasing tools? If they are based on OpenAI API, would that defeat the rephrasing purpose, by adding more cryptographic watermarks? Again, not sure.

But Google has many other ways of detecting AI content besides this cryptographic watermark. They have and will train models by looking at large datasets of human vs ai content. There are distinct patterns machines can pick up on that humans miss. However, it does get more difficult to detect the more advanced these systems get and the bigger seed datasets.

The real question: what happens when there are other models, open-source models, where people custom train these models, advance, etc? It becomes harder and harder for google to detect. It's not like the days of content spinners, where Google easily leaped that hurdle. It is more of a cat-and-mouse game. At the same token, don't underestimate Google's abilities here.

But if everyone can create general AI text, then it is saturated. SO the only real way to rank would be to have something better than all the other "AI" content. Ultimately, the goal is the same: put in human effort to create better content. It may be that AI is just used as a tool to make this good content, building brick by brick, paragraph by paragraph, a good piece of content.

The other real question is: what happens when a general AI takes market share from traditional search engines? Fewer clicks for SEOs. What happens when a general AI is literally placed on top of traditional search results (in place of featured snippet content)? Fewer clicks for SEOs. I think this is a real trend we will continue to see even more than what we already have been seeing.


I'm Curious, what detector are you using?

I haven't found any reliable ones yet. They are all different, and some seem better, but give false positives.
The one right here: https://www.buildersociety.com/thre...n-ai-generated-content.6484/page-4#post-72311

Should also note, you'd think the classifier would work better the more text you give it. But that actually doesn't appear to be the case. The more text you give it, the less confident it gets. The caveat here is that you want to combine text from different prompts. But now we're starting to get into the territory of giving away tips that are actually useful.

Generally, my thinking here is that this is just an IQ test. If you are idiot, you'll get caught. If you put even the minimal amount of work in you can likely get away with a lot, for quite a long time. If you really put a ton of time into doing this properly, the output can be truly amazing and the only feasible way of detection would be if OpenAI can somehow include a cryptographic signature that isn't easily removed, yet retains strong output quality. We'll get a pretty good idea of if they do, because I don't see how they can retain logit bias and other controls over the output, while including cryptographic signatures.

Also, for everyone thinking this is a real possibility, think how many unique signatures would have to be generated. At least hundreds of billions. How is Google going to reference that database of signatures and iterate over it when it crawls and indexes every single page?

OpenAI only has value when the output is nearly equal to a human. Once you deviate below that, the value drops exponentially. Until somehow can give a good argument for how they can balance all the tradeoffs, while simultaneously making a scalable technical solution at Google, I'm going to remain extremely skeptical. It's much easier to just use the PR team.. I mean, web spam team, to reduce AI usage and then tackle only the most obvious cases.
 
Last edited:

If you read the limitations of this AI tool:
  • The classifier is very unreliable on short texts (below 1,000 characters). Even longer texts are sometimes incorrectly labeled by the classifier. That is the usual length of a social media post: between 142 words and 250 words.
  • Sometimes human-written text will be incorrectly but confidently labeled as AI-written by our classifier.
  • AI-written text can be edited to evade the classifier.
Just to clarify - I personally would not take a chance and get any AI content on my websites. Anything that would jeopardize their success (even with a small risk margin) is not acceptable to me so I am staying far away from any AI content on my websites.

That being said, I don't see why content writers wouldn't use it to edit their content and send it back to me. How am I going to know if it's AI or them? If it passes copyscape, quality checks and accuracy. I don't see how I would know if it's AI content or not? Sure, I can run it in some tools but they are unreliable at this point.

Obviously, if the writer suddenly starts finishing articles quickly or I notice repetitive phrases, then I would be alarmed and probably switch writers or bluntly ask them if that is what they are doing.

All that being said, good luck detecting AI content on social media sites.

UWQJvpG.png
 
That being said, I don't see why content writers wouldn't use it to edit their content and send it back to me. How am I going to know if it's AI or them? If it passes copyscape, quality checks and accuracy. I don't see how I would know if it's AI content or not? Sure, I can run it in some tools but they are unreliable at this point.

Obviously, if the writer suddenly starts finishing articles quickly or I notice repetitive phrases, then I would be alarmed and probably switch writers or bluntly ask them if that is what they are doing.
Yet another reason why just hiring a writer is not sufficient. Unless you're writing everything yourself or can somehow be incredibly confident that it's human written, you might as well just use the AI. I can guarantee that a huge % of hired writers are now just using AI, editing it and to hide by not being seen as abnormally fast, are just taking 10 different jobs instead of 1.

I would say that it's a skill to use the AI well. If you're a dunce, you're going to output crap unless you manually edit it. But if someone is really skilled at it, they should take 100 - 1000 Upwork jobs, and pump out content. Could cross $1m / year pretty quick.
 
That being said, I don't see why content writers wouldn't use it to edit their content and send it back to me. How am I going to know if it's AI or them? If it passes copyscape, quality checks and accuracy. I don't see how I would know if it's AI content or not? Sure, I can run it in some tools but they are unreliable at this point.
This is already happening. I've already caught some writers using AI. This is going to be a HUGE problem for companies hiring writers, especially if Google does get really good at detecting AI. There will definitely be an arms race of quality AI detectors, but the thing is, I think this tech will outpace those detectors. Only a company like Google has a chance, but they would keep their detectors internal.

(detectors that rely on patterns of AI content, not cryptographic signatures)

Yet another reason why just hiring a writer is not sufficient. Unless you're writing everything yourself or can somehow be incredibly confident that it's human written
You have some good points in your posts. Makes you think...Maybe people will be resorting to hiring writers in-house, or utilizing screen capture software to police the freelancers usage of AI tools.

At this point I think its still a major risk using AI content on your main sites. I am still not certain whether google will be able to detect AI content (via common patterns in AI text, or via a cryptographic signature).

Also, about the AI detectors (classifiers), the better these get, the easier it will be for people to evade AI detection. It's a catch 22.

Regardless, AI is here to stay, and its probably going to saturate many markets. We will see how well Google will do at classifying AI content, at scale. And more than that, I know many people will disagree, but I think AI will disrupt search, in a major way. I feel like people have their head in the sand in terms of how disruptive general AI will be, to not only web, but to the job market, schools and to how people learn, how gov. uses it to control and social engineering, etc.

Post 2016, like after 2017-2018, Google has gotten worse at finding very specific, granular search results. Want to find specific answers? Need to append "reddit" to your search term, etc. Looking for very specific programming questions? Much harder now than it used to be. There was a very distinct change, and it was a their new machine learning, that made the common searches better, but harmed long tail searches. This is one place that ChatGPT is much better currently than Google, in many cases. It is also good at giving very succinct, quick answers. Wait until it gets even better with more reliable results.

I've been using ChatGPT in place of google for some specific queries, even advanced topics, and its been very impressive, and better than Google in many cases.

Being SEOs and content marketers, I think we are all biased. Just like some graphic designers I've talked to. They have a natural tendency to not want to see the writing on the wall, in regards to all the Art AI tools. Search engines will have its place, but I think AI searching will disrupt many informational search queries, going into the future. At the minimum, AI models and traditional search engines will complement each other.

Remember this is day 1 of this tech... It's 2005, and YouTube launches... Except we are on the upper hockey stick curve of technological growth rate. 5 Years in AI will be like 20 years in internet growth.

However if ChatGPT / Microsoft partnership evolves to the point where they do charge $42/mo for ChatGPT, then that will not displace search. Unless Google comes out with their own AI assistant that answers questions in place of Featured snippets. Maybe once they are confident with the factual reliability of the output, they will.
 
Last edited:
  • Like
Reactions: Boy
How did you catch them?
That's the thing, there is no fool proof way to detect it. So I didn't really have any solid evidence. I used a combination of AI detectors, along with manually analyzing the content in question, and it had certain things that, to me, appeared to have signs of being AI content. But AI content detectors can also give false positives (content I've written so I know for a fact not AI), as high as 10-20% in my testing, depending on the tools.

You have to ask yourself: "If I was a content writer, would I be using these to speed up writing?" Time as an input = money as an output for content writers. So anything that will speed up process to free time will be used. If AI content isn't easily detectible, and especially in the beginning when its the wild wild west, there is too much temptation as a content writer to use AI tools to speed up writing. People act based on incentive, so content writers will try to utilize these tools. Maybe not for all the writing, but maybe for 80% of the output, for instance.

I could almost guarantee, that if not now, but eventually, any remotely savvy content writer is attempting to integrate AI tools. I would not be surprised if 50%+ of writers are or will be using AI tools.

Assuming Google classifies human written content is white hat, and AI content black hat, and assuming Google gets really good at classifying AI content, I foresee the issue of content creators using AI content for companies who want to stay white hat with their marketing, being a major issue in the future. Unless we get more reliable detectors (None of them are very good, even OpenAI's new one has only a 26% true positive, and a 9% false positive rate).
 
Last edited:
Bing detects AI content and is blocking it from being indexed.

V1AOxBT.png


There was someone on BS that was making content with OpenAI so I wanted to play a little bit with it. Will try different methods to see if I manage to index it.
 
Bing detects AI content and is blocking it from being indexed.
That's probably not because it's AI.

My AI site indexes fine on Bing.
But my non-AI site was deindexed by Bing for over a year.
 
Back