A little warning about chatGPT

[Sgt_Kickaxe]

Martinibuster wrote a nice article about chatGPT, watermarking, SEO uses etc on SEJ - [searchenginejournal.com...]

I think it needs to mention one more thing. You can't even log into chatGPT without google knowing, it employs Google captcha.

- The captcha is part of the chatGPT login
- Your IP and other details are captured
- Your browser is cookied
- etc

How hard would it be to cross reference that your Google account also uses adsense or analytics for such and such websites... and to flag them for monitoring to spot AI generated text? Not very... they already do something similar for signs of illegal activity if I remember correctly.

Regardless, chatGPT follows a bunch of patterns. I cant remember the last time I had a conversation with someone that began a sentence with "additionally", a topic with "there are many" or finished the conversation with "finally". ChatGPT does all 3, a lot.

Use sparingly. Perhaps get help with page titles and descriptions, or come up with ideas, but if you're copying and pasting Google's likely already flagging your sites the moment you load that captcha. If you log in to try it out occasionally that's probably fine but if you're tripping that captcha daily, multiple times a day... it looks sus.

Google may not use all information they know, it might be working wonders for you even, but you put yourself in a position they could, if they choose to down the road.

One more thing - one day, probably not too long from now, some queries are going to have no results at all. A chatGPT-esque bot is going to answer your query right on Google. Maybe even with voice. If AI content is too prolific online they may as well use their own version to serve queries right?

The last hurdle for that is monetizing it imo.

[christianz]

It doesn't matter too much. There will be bots like this and better in few years readily and anonymously available to spammers.

Google already often ranks copy-paste and AI spun/generated content above human content. So nothing will change. Spam and more spam.

[Sgt_Kickaxe]

Correction "It does not matter...", chatGPT doesn't use apostrophes. It has a lot of quirks like that making it easy to spot.

Regardless, I was pointing out that you're already telling Google, "hey, my sites are probably full of AI generated content contrary to your terms" when you complete that Google captcha on the chatGPT login screen too often. If you also use adsense or analytics, they know who you are and which sites you've claimed.

Google knows AI content is coming, I bet they have an AI bot ready to answer all informational queries directly on the google page, eventually. No more serps when AI text becomes too rampant. Notice chatGPT can already answer a metric ton of informational queries? Serps will change to voice/text answers directly from Google the moment they figure out how to monetize it(and keep people flowing to Youtube).

Plan accordingly, it's going to matter to your bottom line if you need traffic from Google and they switch to a chatbot instead of ranking sites.

[robzilla]

You can't even log into chatGPT without google knowing

You can, via the API.

I'm a little reassured about the future of the Web as we know it after reading that article, thanks @martinibuster.

[engine]

If anyone uses the text from there it could easily end up a duplicate. Caveat: You're not supposed to do that according the T&Cs.

[Sgt_Kickaxe]

It's also often wrong, or right but for the wrong reasons. When it is, it's confidently wrong.

[mhansen]

Noted. I have been using it lightly to help me discover formatting of a single paragraph in a Q&A type section. I drop the Q into ChatGPT and review the answer for validity, then using nearly the exact AI format, I rewrite it in bits and pieces for use on the website. In total, maybe 25-40 words or 2-3 short sentences of a long-content page. I tend to use separate browsers for this, one which never logs in to G services through my same GA/webmaster account.

I wear a pretty thick tinfoil hat myself after reading in 2005 or 2006 that Google, when they became a Domain Registrar, stated they would use every tool in their power to protect the quality of their search results. Then went ahead and got a patent on how they use whois info to assign a reputation score to domain owners, that could be spread across all domains owned by the same entity. Bill had a nice writeup on the patents and WHOIS relationship back in 2006.

[seobythesea.com ]

[martinibuster]

I updated the article last night to reflect that Google quietly changed their guidance around auto-generated content. The documentation was published in October 2022. The clarification of the documentation was made a month later in November.

Usually they announce these changes in a changelog but they didn't do that for this change. That's why I was unaware of it until last night when someone on Facebook told me about it (and I was sick and not working at the beginning and end of November).

It's strange that they didn't notate it in their changelog because they do it for other changes that are relatively trivial.

The changes reflect a clarification about what makes autogenerated content spam.

It initially said this:

“Automatically generated (or “auto-generated”) content is content that’s been generated programmatically without producing anything original or adding sufficient value;”

Google updated that sentence to include the word “spammy”:

“Spammy automatically generated (or “auto-generated”) content is content that’s been generated programmatically without producing anything original or adding sufficient value;”

That change appears to clarify that simply being automatically generated content doesn’t make it spammy. It’s the lack of all the value-adds and general “spammy” qualities that makes that content problematic.

[phranque]

from the same team who developed ChatGPT, there is a GPT-2 Output Detector [huggingface.co] available.
i would assume that a GPT-3(.5) will also be available soon.
while it can't refer to the source of plagiarism like a typical plagiarism detecting software (TurnItIn/Copyscape/etc) it should be a useful BS detector.

[Martin Potter]

ChatGPT enters, stage right.
ChatGPT, in hushed tones : "Note to self. Sound less confident. Learn to use apostrophes. Avoid starting sentences with 'additionally' and 'finally'. Then Master Google should be happy."
ChatGPT exits, stage left.

[christianz]

from the same team who developed ChatGPT, there is a GPT-2 Output Detector [huggingface.co] available.
i would assume that a GPT-3(.5) will also be available soon.
while it can't refer to the source of plagiarism like a typical plagiarism detecting software (TurnItIn/Copyscape/etc) it should be a useful BS detector.

It's accuracy must be dependent on length of the text being examined. For very short text I don't think it can detect AI reliably.

[Sgt_Kickaxe]

Thanks for the update, Martinibuster. If "spammy" is the threshold for using AI content, to each his own tolerance I guess.

It will be interesting to see if Google eventually tires or AI generated content, or if it decides outsourced articles are more problematic(which would often include AI content). Apparently, given enough content, Google can determine, with a high probability, if the listed author is actually writing the content.

[martinibuster]

For very short text I don't think it can detect AI reliably.

Detection with that tool becomes reliable with at least 50 words. I couldn't get it to show me a false positive with articles I knew were not autogenerated.

But it was able to detect ChatGPT content.

[No5needinput]

What ad-laden hell site is that "supposedly" SEJ SEO site you linked to? I counted no fewer than 42 ads plus 4 popups - before I lost count - and the site still hadn't finished loading minutes later (desktop). I have a powerful computer with 64 gigs of ram, yet the fans were working at full speed, trying to keep from frying my computer while loading that site from hell. SMFH

[Pjman]

Yeah Boys, not trying to sound any alarm or anything, but if you have an Info driven site that AI could compete against, I would think of selling ASAP. I've seen this game before and it is not time to leave cash on the table, especially with the awesome interest rates now. I sold (all of my sites) 4 months ago and earned 6 figures on interest already, not bragging by any means, just trying to help the good people of this community.

[phranque]

What ad-laden hell site is that "supposedly" SEJ SEO site you linked to?

horrible UX but often good content so i hold my nose and duck the ads when necessary...

[engine]

The only consumers of AI generated content are humans. Clearly, creating volumes of material is easy for AI, but is it creative, is it interesting, or entertaining?
Will the content be unique?
Will it pass Google's E-E-A-T?

[Sgt_Kickaxe]

The Bigger Picture

There is no point in debating the merrits of it. Your only concern should be your business/profesional exposure to it.

A reminder

- Free Google traffic is not yours, Google owns that source.
- Free Facebook traffic is not yours, Facebook owns that source.
- All sources of traffic you don't control ARE NOT YOURS.

So if AI content is really about to rock the boat, and you might see your traffic impacted from sources you don't own, take stock of traffic that IS yours.

Traffic you get from your email list, for example, is yours. Have you been building one? Other sources include sites you own, groups you manage, associations you have, networks you control and old-fashioned return visitors using a bookmark.

Work on those!

[ROLAND_F]

I wouldn't be too woried about what Google says. They will end up being just an identity provider for ChatGPT if they don't do something big quickly. I use ChatGPT a lot since I got access to the public beta. And Google a lot less.

If I had to use ChatGPT for content creation, I would specify requirements for the generated text to not sound too generated and I would rewrite a few wordings so the text use my expressions and sound more personnal.

[Sgt_Kickaxe]

chatGPT is seeing explosive growth in users, it won't be free much longer. Their costs are increasing too.

[edited by: Sgt_Kickaxe at 4:17 am (utc) on Jan 2, 2023]

[dolcevita]

For me, chatGPT is fun to use and interesting as a helper for essays or content creation. But in no case did it impose itself on me as a search tool instead of a search engine. It is too monotonous and monochromatic and therefore uninteresting even compared to websites.

[Sgt_Kickaxe]

Think voice, they'll merge soon enough.

[martinibuster]

Google's AI for voice is indeed progressing. I suspect in a few months when Google holds their yearly search presentation they'll announce some new voice related advances.

Something really cool that Google, Amazon, Microsoft and Apple are working on together is creating an AI that can understand the voices of people with impaired speech, like people who've had strokes or have some other issue that impacts their voices.

Something like that, in my opinion, is a useful advance in technology. So it's good to see these companies come together to share their research to create these advances.

[Sgt_Kickaxe]

What to look for next: Upcoming events where plans and changes may get discussed.

- The World Economic Forum is holding it's meeting of elites in Davos the 16-20 of this month. This was Google CEO Subdar Pinchai's presentation in 2020. It was about quantum computing and data protection, but look for signs of global search or information related search change in the 2023 presentations - [weforum.org...]

- He's presenting at IO 2023 in May. Upcoming tech, wearables and other new products typically get presented but since a code-red has been called, and chatGPT has moved to the top of Google's importance list, it's another date to know because things may change around that time.

- Alphabet recently altered Sundai's pay to more closely tie it to Google performance. Not long after, rumors began swirling that Google may cut up to 10,000 jobs this year. Pay attention to who gets cut, from which departments/products, and who doesn't, as these changes may signal Google's immediate direction.

Ultimately, if your site offers informational content, guides or recommendations, you best make sure that your site offers it in a way not easily matched by AI. While AI can tell you some things, it can't do it in a predictive way, yet. It's not going to know what the upcoming football score is, or give intuitive fantasy ball lineup suggestions, so AI is not going to replace everything.

Small businesses will need to manage their reputations more closely else AI may ignore them completely. It's expected AI will need to get very local if it replaces traditional search, so focus on exposure locally.

DO NOT PANIC - If you wake up one morning and a Google chatbot replaces the search box, and you get no more free traffic from Google, it's not the end of the world. Change brings new opportunities. When information went from books to online the books didn't cease to exist, and neither will the websites.

ChatGPT is likely not Google's biggest threat. An actual browser/search engine/AI chat system like Neeva, which was created by ex-Googlers, probably is. 2023 looks to be packing more change.

WORRY ABOUT FINDING AND HELPING/PLEASING PEOPLE - Google's problems are theirs.

[Nutterum]

OK, I have used Chat GPT extensively since some time now, so let me tell you what it is good for and what it is bad for .

The Good :

1) Perfect for first draft on SEO articles on certain topic, that one can present to the copywriters
2) Perfect to get a second take on content snippet . For example I use the bot if I want to write a description for "summer blue dresses that are light and stylish, perfect for people of all sizes " When I want something decent written within say 100 words, the bot performs very very well with some tinkering.
3) Good for creating Meta Desctiptions that are not too salesy
4) Good for creating first draft answers for FAQ sections , especially if you have detailed FAQ for various categories of products, product reviews and such . Needs a human touch to finalize though!

What it is _NOT_ good at

1) Fact checking
2) Opinion content
3) Comparison content
4) Long form text (say 2k word product review etc.)
5) List content ( 5 things that xyz)
6) Sensitive content - politics, gambling, history, war, crime, etc. etc.

It really helps me scaled up my content initiatives.

But it's not the internet/ content killer most people hype it to be.

I does use very natural language and I do understand why Google are all-hands-on-deck over this new way of "search" though.

[superclown2]

Thank you Sgt Kickaxe for letting me know about ChatGPT and Neeva. Like most of us I'm heartily sick of Google destroying months or years of work whilst they enrich themselves on the back of that work.

I'm hardly a youngster - I was working online 25 years ago and watched the rise and fall of Altavista and Yahoo - yet I spent most of yesterday on ChatGPT and thought it was wonderful. There is no doubt it will take massive amounts of business from Google.

I tried Neeva this morning - it was clean, fast and gave excellent results. More importantly they seem to have massive resources behind them - every page of a new website I launched less than a month ago, with no backlinks, was in their cache whilst Google only has the index page listed. I checked numerous other small sites I created last year that Google hasn't listed at all but there they are, fully listed on Neeva. These people clearly mean business.

Incidentally: whilst typing this I was distracted by an interview on Sky News that praised ChatGPT; no publicity is bad publicity (and DDG is advertising on the TV here in the UK now) so I am confident that the interest in it will be huge in 2023.

My predictions in 2021 about the future downfall of Google met with an unfavourable response from certain quarters which is why I stopped posting here; but to paraphrase Churchill, this is not the end, but it could well be the beginning of the end for that company, despite the hundreds of millions of dollars they spend on 'lobbying'. Perhaps there is a need for ChatGPT and Neeva categories here on WebmasterWorld soon ?

[phranque]

another ChatGPT detector has been getting some attention - this one is called GPTZero [gptzero.me]

GPTZero turns the very technologies used to build ChatGPT around — to detect AI. It uses variables like perplexity [and "burstiness"] to fingerprint AI involvement.

[martinibuster]

I'm heartily sick of Google destroying months or years of work whilst they enrich themselves on the back of that work. ...I spent most of yesterday on ChatGPT and thought it was wonderful.

Something to be mindful of is that large language models like ChatGPT are trained on web content.

It's learning off the back of your work and delivering answers without citing where it got that knowledge from.

[superclown2]

Something to be mindful of is that large language models like ChatGPT are trained on web content. It's learning off the back of your work and delivering answers without citing where it got that knowledge from.

Oh I absolutely agree. However I was looking at it from a searcher's point of view and not a website owner's. I liked the way I could look for non commercial information, even some quite technical stuff, and get detailed answers without having to plough through ads to get at it.

[NickMNS]

The biggest caveat with ChatGPT is that it is often wrong but given the verbosity of the response and the nature of the question it can be difficult for the user to determine that the response is in fact wrong. The answer my be correct in some manner but still wrong as to specific of your problem. Coding questions are an example of this, if you search for answers on stack-overflow you get multiple answers, they may all be wrong, but what you also get is more context around your problem. There may also be several possible solutions. ChatGPT, may provide them you can to this, or that or the other. But with stack-overflow you have comments, and varying response that may lead you to better select the optimal answer, or reject all of them.

I have to say @Nutterum has the right approach.

I use it to check my spelling, it will find all my mistakes, grammar, spelling, typos whatever. I also give it my text as a block and see what it responds. Is the response what I expected? Did it identify the entities (references to real-world things/places/people) correctly, can the NLP interpret the text as I expected it, if NLP does then real world user will. There is the old adage, "write for the user and then the bot will understand". So I turn that around if the bot understands then the user should too.

I'm also,surprised that no one has brought-up using chat-gpt for writing/debugging code. I know of people doing it, typically less skilled people. From experience, it may provide some code to do one specific thing, but the net result of pasting a bunch of it together is likely going to end in one big scary mess.

Something to be mindful of is that large language models like ChatGPT are trained on web content.

Given the above, its success will lead to its downfall. If it is truly able to disrupt the "information" web as it appears it likely will, then publishers of the "information" will no longer be needed. Without new, diverse and timely content, the training of the AI model will suffer.