Thread by @henkvaness

It was inevitable. Spammers use #ChatGPT to create fake websites using natural language processing. It allows malicious actors to quickly generate bogus sites that generate revenue from ads or other malicious content. Next few tweets: how is it done? (1/4)

Just 24 hours ago, a certain Pallab G. started to sell for around $9 an automated workflow that includes #chatgpt. The unlisted video is in this tweet, at at the time of writing, had 4 views, including 3 of mine,

#chatgpt writes a main page of a website, including photo suggestions, as you can see in this example "Write the main page of a website, including photo suggestions, about how to travel in
Arizona when you are on a tight budget, have no income, and can't … twitter.com/i/web/status/1…

Here's how the #chatgpt tool generates the first part of a website. In 15 seconds. The text is written by #chatgpt, so you will not find any plagiarism if you Google it. Next step is to insert ads or spam devices. Done. (4/5)

Is it bad that #chatgpt can generate fake websites like this? Most of the text is better written than SEO spam and has more information. I'm curious if Google will be able to fight these sites. If I ask a tool to detect if #chatgpt text is machine made, it says "no" : it's 100%… twitter.com/i/web/status/1…

#chatgpt's machine text detector couldn't detect that the fake website wasn't man-made. Take a look at platform.openai.com/ai-text-classi… Welcome to the future ... (6/..)

Here's another website I made in 5 minutes with the spam tool and #chatgpt . The classifier of #chatgpt considers the text to be unclear if it is AI-generated, but it certainly is (7/..)

The instruction I gave #chatgpt was "write a homepage with 4 headers about how to pluck oranges in windy environments" I know, it's a silly topic. #chatgpt came up with "Welcome to the Orange Plucking Guide for Windy Environments!" and four tips. When I included "plucking in a… twitter.com/i/web/status/1…

The scripts that autogenerate fake websites with #chatgpt are still very elementary. They do not use the paid version of #chatgpt ($20) yet. But soon they will, is my best guess. (end)

Wordpress add-ons can also generate text with #chatgpt. As a human, you often can't tell that the text was written by a machine when the topic is broad and in popular domain.

But if the text is outrageous, like the one I asked #chatgpt to write (see pic), it will often be detected. Here are some tools to detect a machine written text I use:

1. AI Text Classifier
platform.openai.com/ai-text-classi…

2. GTP Zero
gptzero.me

3 copyleaks.com/features/ai-co…… twitter.com/i/web/status/1…

Fake websites are not new. "Link injection" is a popular technique that makes brand-new pages appear on existing sites. The sole purpose is to sell stuff, earn money from ads, or rank higher on Google. Humans still write most of the added pages. The text docs.google.com/spreadsheets/d…… twitter.com/i/web/status/1…

Listed below are some examples of sites that use both autogenerated and human-written text. They mostly copy and paste from the web instead of using #chatgpt:

meidilight.com
zoomlocalnews.com
tetracycline-abc.com

As a proof of concept, I will show you two… twitter.com/i/web/status/1…

Will Google be able to fight the fake #chatgpt sites? The sites that use auto-generated texts, for now, still in a rather crude way (see this thread with 2000 examples), were not detected by Google as spam. And Google will not forbid #chatgpt by default developers.google.com/search/blog/20…… twitter.com/i/web/status/1…