I added some CHATGPT content to some sites just after Christmas. All the sites tanked. I removed that content and the sites recovered slightly last week.

Yes Google can detect it. It's easy.

@superclown2 Did you announced that the content of your articles was partially / completely written by an IA ? G said that using an IA to write is not against his policies, but it should be announced somewhere.

Whether announced or not, g will likely deprecate the content as machine made.

Use the tool, but make sure a HUMAN edits the output before posting to the web!

(which kind of seems like "more work" to me)

@tangor Source for your claim "g will likely deprecate the content as machine made."?

Did some tests with pure Chat GPT texts and all of these articles and pages tanked within days. My next experiment was to use the content as a basis for copy-writing doing like 30-40% alterations of the text and the pages still tanked. In early March I did an experiment with just building content with ideas from Chat GPT but I wrote my own texts - worked like a charm.

Conclusion : Google can see chat GPT text and it values it punitively 100%. But the text AI is still good in creating some semantic keywords and idea basis for content on the website. But it needs a human to write it properly.

@Alex 1729 ... there's two sources in this very thread! The other sources are all over the web. G has not yet codified a response greater than "AI generated text will be discouraged" (which is a paraphrase and NOT ANY EXACT QUOTE from G. The handwriting is on the wall.

Meanwhile, the built in bias toward one ideology over another is something else to consider when using ChatGPT (any version).

Everyone including their grandmother are using GPT3 or Chat. Google is going to have a field day.

Google just rolled out their "BARD" version. Might be fun for all!

Time will tell how all this works out... and what we, as site owners, might "enjoy".

Anyone have any idea how Google is detecting ChatGPT texts?

Anyone have any idea how Google is detecting ChatGPT texts?

More than likely an automated version of their own detection tool which works similarly to the publicly available AI detection tools.

If one creates AI you can bet they can detect AI. :)

Enough of AI is open source that quite a few are already ahead of the game.

I can understand Google penalising ChatGPT generated content, as it sees ChatGPT as a competitor, but presumably someone creating content using BARD as an assistant wouldn't expect to be penalised by Google at least!

I would not bet money on that. Generated AI content is generated AI content. It has characteristics and Google can recognize those no matter the tool. Bard was offered so you don't need to go to another service, not because it avoids detection.

@Mark_A it's automated content they, and anyone can detect. Its doesn't matter if it's AI-generated, it's the value of the content that counts.

Let me give you an example of a small test I've tried: I created a web page with pure AI generated content, and I created a page with hand-created content. In addition, I created a page which used AI-generated content but I re-wrote it so that the online detectors couldn't indicate it was greater well less than 50% AI.

All three pages ranked in Google very quickly (primarily because it was new). However, within three weeks, the AI-generated content tanked, and the other pages settled somewhere on page 1.
This only proves to me that there is a detector (Google's SpamBrain [webmasterworld.com]) and it downgraded the page. It was probably a rubbish page, imho, so deserved to be ditched.

Note AI generated content is not against Google's policies.

I threw AdSense on the pages, and it made no difference to the ranking.

I doubt it's just OpenAI's ChatGPT content that could be penalised. Let's face it, there are now so many generative tools around, some of which have been around for quite a while prior to ChatGPT's profile appearing on the radar. In fact, for many years there's been automatically-generated content.
I suspect for Google it's all about EEAT and SpamBrain.

Google does indicate about automated content [developers.google.com...]

And this page is updated about using automated content

Using automation—including AI—to generate content with the primary purpose of manipulating ranking in search results is a violation of our spam policies.

[developers.google.com...]

HTH

Hi engine, thanks for that.

And interesting the experiment you did certainly.

One thing is:

Using automation—including AI—to generate content with the primary purpose of manipulating ranking in search results is a violation of our spam policies.

My purpose in trying ChatGPT to generate text, isn't because I want to manipulate ranking, I just find it a very convenient way to generate content. However while I don't want to manipulate ranking, that isn't to say that I find it acceptable for my content produced in this way to be penalised, I don't.

My purpose in trying ChatGPT to generate text, isn't because I want to manipulate ranking, I just find it a very convenient way to generate content.

That's great, and you should be fine, and it'd be even better with human-edited text. However, one caveat is that the "facts" it may generate need checking. There are so many examples online of "facts" being plain wrong.

I wiil do fact checking and rewriting. Still should be easier than composing from scratch.

One caveat when using ai generated text, even if human edited, it will remove the spontaneity and insight that a fully engaged human brain can add. These suggested texts, even if edited by a human, will have an overall sameness about them.

More importantly, some of these generators have rule sets you might not be able to change and could ignore important information for your mission, or USE wrong information (all too often seen in recent weeks!).

In some there is a bias that might not include a full audience spectrum, if that is suitable, then by all means use the tool.

Sometimes "more" is just "more" of the same, particularly in ecomm environments and the possibility of thin or duplicate content can be compounded at the press of a button.

Yes, I have found a slight blandness in texts, also I found that I get the best stuff by ongoing questioning and then knitting it together and editing afterwards.

@Mark_A ... Please keep us informed on your efforts and results! AI has the potential to be a good thing. Real world results, however, are the final thing!

>detecting

The basic gist seems to be the difference between how humans write vs an AI probabilistically guessing the next word. Human idiosyncrasies in language seem to be a big enough signal. AI output seems very readable, on-topic and maybe hard to detect to a human reader, but sufficiently different.

I would bet there are some 'content creators' who can say differently in that they're receiving more traffic rather than less, but they're sufficiently obfuscating the AI output.

Also to add to the pot, tested with a 4000 page site, fed product descriptions through chatGPT and the descriptions are clearly unique vs being aff feed descriptions, site CTR's have clearly went down.

@Mark_A ... Please keep us informed on your efforts and results! ...

Will try, feedback is likely to be anecdotal though ..

It seemed logical to ask ChatGPT: "How can / does Google detect content generated by AI / ChatGPT"?

The output was mildly interesting and, perhaps, predictable - as to content and . . writing/speaking style.(?)

Lesson learned, from speaking with ChatGPT: There may be some value to being / writing that is . . slightly daft? Variant? Human?

At least for a few more years.

Alrighty. Back to the hedonistic treadmill. Thinking . . predicting . . all too soon doing . . will be left to the more capable . . durable . . clever . . AI driven machines. :-/

The following short video is for entertainment purposes only. Screenwriters and TV producers and scientists weren't really up to speed on AI and superiority was really just a bit too unfathomable / scary to put in front of the TV watching masses . . and likely a . . cough . . career killer for Captain James Tiberius Kirk and the Star Trek franchise.

[youtube.com ]

@Mark_A ... Please keep us informed on your efforts and results! AI has the potential to be a good thing. Real world results, however, are the final thing!

Hi Tangor et al, I pushed a new page built with the help of ChatGPT online, It was a long page, constructed from three or four ChatGPT interactions on one subject (and reworded a bit in my own words). It doesn't yet rank (on P1 to P3) for its, or related subjects.

G has crawled and indexed it, because searching for a string of text from it does return the page.

Perhaps it is early days, either that or I do need to rewrite it more, perhaps it is too long,
I might have another go because the subject isn't covered extensively online, and explaining it at length I think adds to our credibility.