Chat GPT knowledge base VS Your Own

[Denis021]

Hey guys I am new here so I wanted to start with an idea I am currently testing.

Recently I got an email from Authority Hacker regarding sites that went up while a massive amount of sites crashed.
Since all 3 sites from their email have pretty unique content I decided to test if they would show better results, ranking-wise.

I figured out Chat GPT is scraping its data from search engines like Google so mind also asked GPT to try to get some answers.

I asked it questions like Analyze this article and come back with a percentage of unique information and a percentage of information you already possess in your knowledge base.

After a few tweaks of the prompt, I got something like this:
This article has 40-45% unique information while it has 55-60% information already known.

These are widely known information that they put in GPT's knowledge base and then he uses them to spin content and make hundreds of blogs with basically 0% of unique information, information from "your" knowledge base.

I put this to the test and started working on some articles that I made using chat gpt + human editing.

I plan to rewrite manually these articles using unique information, now that's a catch it is hard to get to those. But simple review, comparison, or some user experience could be unique information.

Here, I shared the idea, let me know if it makes sence.

By the way I will post updates here.

 

[bernard]

Good strategy to think about AI content.

I don't think you can trust those percentage numbers it gives you, it doesn't really have any kind of insight into its own processes so to speak. At least to my understanding. It doesn't do database lookups in the way we're used to.

I definitely agree that when using AI, you should always provide it with unique information and data, otherwise it simply won't be good enough long term. I think Google has figured out raw ChatGPT content already. How I don't know, but ChatGPT does write in a very distinct manner. It's easy to spot for humans.

Still not so much a problem if you feed it unique information and have a human read it and correct. One thing I would suggest is that intros, both in article and subheaders, are human written or at least not written by ChatGPT. You need those hooks. Most people will read the intro and decide. With increasing AI content, people will learn to bounce when they recognise the telltale mark of ChatGPT.

You also have to consider user metrics, which is also what the Authority Hacker guys talk about. User metrics allows google to infer a lot of stuff with their own machine learning algos. They don't need to understand as much content anymore vs looking at user metrics. It's perfectly probable that they also have correlations between what content looks like and reads like and what expected user metrics it would give. So for all intents and purposes, you should write for users.

 

[Denis021]

Good strategy to think about AI content.

I don't think you can trust those percentage numbers it gives you, it doesn't really have any kind of insight into its own processes so to speak. At least to my understanding. It doesn't do database lookups in the way we're used to.

I definitely agree that when using AI, you should always provide it with unique information and data, otherwise it simply won't be good enough long term. I think Google has figured out raw ChatGPT content already. How I don't know, but ChatGPT does write in a very distinct manner. It's easy to spot for humans.

Still not so much a problem if you feed it unique information and have a human read it and correct. One thing I would suggest is that intros, both in article and subheaders, are human written or at least not written by ChatGPT. You need those hooks. Most people will read the intro and decide. With increasing AI content, people will learn to bounce when they recognise the telltale mark of ChatGPT.

You also have to consider user metrics, which is also what the Authority Hacker guys talk about. User metrics allows google to infer a lot of stuff with their own machine learning algos. They don't need to understand as much content anymore vs looking at user metrics. It's perfectly probable that they also have correlations between what content looks like and reads like and what expected user metrics it would give. So for all intents and purposes, you should write for users.

Yes I agree, this is just a starting point to try figuring out what in the article is considered widely know facts and what falls under my own unique knowledge base without even optimizing the content.

What it helps is that I own site in the same niche as the job I 've been working for for the past decade but I am trying to develop SOP so I can apply this aproach to sites in different niches or with writers who are not in this niche.

You mentioned user metric, Eric Lancheres mentioned time on page as a good starting point.

What you think?

 

[bernard]

You mentioned user metric, Eric Lancheres mentioned time on page as a good starting point.

What you think?

The best is just having random people test your site which they will typically do on mobile and they'll spot stuff that tools won't do.

 

[Denis021]

Yeah not a bad idea.

I see you are mentioning api integration with chat gpt in your thread.

I am eager to learn more on that. I stopped writing with GPT, it doesn't work that well on this site or in this niche I think but if I can make the research faster and easier it would help with unique info in the article.