[Journey] Ranking AI Sites For Passive Income
- Thread starter spectrejoe
- Start date
- Sep 25, 2013
- 4,698
- 2,694
Yep I don't think people figured out just how wildly different your results with AI can be by actually prompting things correctly.
Will post an update tomorrow as it's the end of the first week of the journey. Early on there won't be much to update on since it's just pumping articles every day. Things should pick up a couple months into this
- Sep 9, 2023
- 2,801
- 1,571
Looks solid. Starting general gives flexibility. Your in-depth AI content will build authority. Focusing on Amazon first, then ads, makes sense. Scaling sites and selling them once they plateau is smart while keeping quality high.
- Sep 25, 2013
- 4,698
- 2,694
Update #4
Sites are fresh, not much to happen realistically, but that doesn't mean the work stopped, in fact, I've been in a constant struggle with my workflow & with continuously improving my prompting.
- Added images to H2 headings to my AI Workflow (when it fits, AI will decide if an image fits the h2 or not, all images generated are relevant to each individual / chosen h2)
I did this because I think it adds a lot to the article, also adds effort that I think a lot of AI Spam sites simply don't bother doing other than inserting a single image for the featured image and that's it. Relevant images can help visualize the text for the readers and increase the time on page metrics which is surely important for rankings.
All images have their own title & alt text
- Banging my head against the wall to try and reduce AI text detection on Originality.ai and app.gptzero.me
So I know a lot of people seem to think that it's impossible for google to detect if text is AI generated or not because some AI articles rank, while I do think there's some merit to that I also genuinely believe that they do have some sort of detection albeit extremely rudimentary compared to the "advanced" detectors.
The reason being that some articles do get ranked, others no matter how good they are simply won't rank AT ALL even deep in google pages despite literally having 0 competition past page 2-3. This is a huge red flag for me and I don't recall it ever happening when I actually put in the effort to write things manually. It only seems to happen to me with AI Articles.
The merit for this claim is that doing some kind of really advanced text analysis and AI detection if not rudimentary to literally billions of texts/articles would have astronomical costs, time consumption and just not be a very profitable thing to do for Google + high chances of false positives could have extremely negative outcomes as a whole for their business. So it makes sense that they aren't going extremely deep into it but do have some sort of detection for the easiest AI detection parameters, what those are, I have no fucking clue.
I've been fucking around a lot with prompting and trying to change, tone, burstiness, avoiding typical AI footprints / phrasing / writing styles. I absolutely can't seem to fool originality.ai at all, and while SOME articles seem to do Okay'ish in app.gptzero.me I can't really get consistent results and MOST of my articles still get flagged at 100% there as well, despite the articles being significantly different than they were initially.
I'm at my wits end on this part and my first "real" wall.
I don't want to just write "better" articles than my competitors, I want to write "better" articles than every other AI user, I'm not competing within just my niche I'm competing with literally everyone using AI as well so I want to excel in both and I will have to find a way to be able to get past this wall.
I don't know if anyone has achieved any results in this area but I'd be grateful for some feedback. And yes I know there are "tools" and services out there that can help but I was trying to avoid using 3rd party services.
Some Stats
Site #1
Posts: 38
Site #2
Posts: 18
Earnings: 0$
Sites are fresh, not much to happen realistically, but that doesn't mean the work stopped, in fact, I've been in a constant struggle with my workflow & with continuously improving my prompting.
- Added images to H2 headings to my AI Workflow (when it fits, AI will decide if an image fits the h2 or not, all images generated are relevant to each individual / chosen h2)
I did this because I think it adds a lot to the article, also adds effort that I think a lot of AI Spam sites simply don't bother doing other than inserting a single image for the featured image and that's it. Relevant images can help visualize the text for the readers and increase the time on page metrics which is surely important for rankings.
All images have their own title & alt text
- Banging my head against the wall to try and reduce AI text detection on Originality.ai and app.gptzero.me
So I know a lot of people seem to think that it's impossible for google to detect if text is AI generated or not because some AI articles rank, while I do think there's some merit to that I also genuinely believe that they do have some sort of detection albeit extremely rudimentary compared to the "advanced" detectors.
The reason being that some articles do get ranked, others no matter how good they are simply won't rank AT ALL even deep in google pages despite literally having 0 competition past page 2-3. This is a huge red flag for me and I don't recall it ever happening when I actually put in the effort to write things manually. It only seems to happen to me with AI Articles.
The merit for this claim is that doing some kind of really advanced text analysis and AI detection if not rudimentary to literally billions of texts/articles would have astronomical costs, time consumption and just not be a very profitable thing to do for Google + high chances of false positives could have extremely negative outcomes as a whole for their business. So it makes sense that they aren't going extremely deep into it but do have some sort of detection for the easiest AI detection parameters, what those are, I have no fucking clue.
I've been fucking around a lot with prompting and trying to change, tone, burstiness, avoiding typical AI footprints / phrasing / writing styles. I absolutely can't seem to fool originality.ai at all, and while SOME articles seem to do Okay'ish in app.gptzero.me I can't really get consistent results and MOST of my articles still get flagged at 100% there as well, despite the articles being significantly different than they were initially.
I'm at my wits end on this part and my first "real" wall.
I don't want to just write "better" articles than my competitors, I want to write "better" articles than every other AI user, I'm not competing within just my niche I'm competing with literally everyone using AI as well so I want to excel in both and I will have to find a way to be able to get past this wall.
I don't know if anyone has achieved any results in this area but I'd be grateful for some feedback. And yes I know there are "tools" and services out there that can help but I was trying to avoid using 3rd party services.
Some Stats
Site #1
Posts: 38
Site #2
Posts: 18
Earnings: 0$
jamesvick
Senior Member
- Jul 26, 2010
- 998
- 701
excellent update. Keep pushing. First income seems on the way. To beat AI detector is hard. I don't know if you are familiar with the concept of finetuning but that is the only way to beat AI detection. Takes insane amount of resources though. You basically train AI along very specific domain related data to make it generate non-filler content or AI slop. But yeah it's just hard. So are you posting detectable content because that would be risky long-term wise.
krishnacode360
Supreme Member
- May 30, 2012
- 1,356
- 422
Following your journey
- Sep 25, 2013
- 4,698
- 2,694
Yea I do know about fine tuning models, the issue is that with publically available solutions I don't think I could create a model that produces better content than ChatGPT 5.4 Thinking / Claude Opus 4.6 could produce and content quality + how well they follow my prompt is a huge deal for me. I also tested and ChatGPT API is completely different from the browser solution so I also decided not to use the API as the quality is simply inferior and more expensive. The browser bypass solution I'm currently using although slow, it's "better" for my use case.
And yea the articles do get detected by Originality.ai (always says 100% AI) and GPTZero (90-100%). While some other AI Detectors returns between 0 to 20% AI like ZeroGPT (Which is different from GPTZero lol).
So it's a mixed bag. 2 of them I feel like are just unbeatable with prompting alone while the others I've already beat for the most part. Now I need to realize if what I did is enough to fool Google or not. Time will tell I guess.
Easily spent over 12 hours yesterday testing prompts, doing research and trying all kinds of things to beat these 2 websites and I'm genuinely sure that it's not possible to beat them with ChatGPT at least, unless you make an article that's completely uncoherent or filled with errors and issues of course, which then would defeat the purpose.
I also came up with a potential few solutions which I'll address next update.
PaulRD
Senior Member
- Jun 17, 2018
- 1,006
- 459
How'd you do it with API. Like letting AI read the articles before generating? Or have you tried using API to generate articles for this journey.
- Sep 25, 2013
- 4,698
- 2,694
I'm not using the API because IIRC ChatGPT and Claude don't access the internet for research through the API, only through the web browser so I built my own automated solution to get everything I need through the web browser (with selenium).
- Sep 25, 2013
- 4,698
- 2,694
I've never tried services niches because it always requires significantly more effort from my part to actually monetize it.
Selling lead calls to actual businesses requires me to reach out + have proof of traffic
Reselling the services myself would be a nightmare to create the framework & workflow for it
etc...
But props to those who do though, there's huge amounts of money to be made with them for sure
PaulRD
Senior Member
- Jun 17, 2018
- 1,006
- 459
sweet
- Sep 25, 2013
- 4,698
- 2,694
Yea I stepped down the volume to 3 articles a day for each website.
I'm scaling horizontally instead, making more sites.
10 sites x 3 articles per day = 30 articles per day
Although with how much I prompt I'm kinda scared i'll reach some limits soon, I'm guessing the limit will likely be 4-5 sites at most per day until I have to get a new account
chrischris15
Registered Member
- May 5, 2014
- 95
- 55
I could be wrong but with OpenAI I think the newer "response" instead of the "chat completion" API is able to web search and has much better thinking.
Developing with it is a little more complex so most people just use the chat completion because it easy.
- Sep 25, 2013
- 4,698
- 2,694
I know the Codex / AI Agents can browse but yea not sure about the rest. Will have a look though thanks!
- Sep 25, 2013
- 4,698
- 2,694
Quick Update / Knowledge Dump
Not including much stuff website related since again, still early wont have anything exciting to share about them for a while. I'll share their stats once a week. Instead I'm gonna focus on sharing some wisdom/knowledge i've picked up along the way. This might be a little bit long.
Splitting this in 2 Different Sections:
- Knowledge Dump (Making AI Content Undetectable)
- My Current Focus & Backlinking
Goal: Make AI Content Undetectable
Prompting Engineering
Prompt engineering is the strategic process of designing, refining, and optimizing text-based inputs (prompts) to guide generative AI models toward producing accurate, relevant, and high-quality outputs.
I put a lot of emphasis in ensuring that:
- My content was EXTREMELY high quality
- Instantly give the reader the answer they were looking for (Fulfilling user's search intent)
- In-depth Explanation of the subject
- Introduces topical clustering & semantic words into the article
- Using a well known Framework: Problem–Solution–Proof–Application
- The AI doesn't sound like the AI by making it significantly different from the AI Standard.
The above not only fulfills the objective of all articles (providing value to the reader & answering the keyword's search intent) but it also ensures that none of my articles can ever be penalized for "thin content".
All of that can and is easily achieved by Prompting Engineering.
Afterwards I wanted to ensure that the text doesn't get detected by AI detectors and therefore, no issues with Google's Anti-AI Updates. For that I focused on:
- Breaking common AI Patterns (Avoiding words that AI overuses, em-dashes, etc...)
- Humanize AI Writting Patterns (Varying sentence lengths, etc...)
- Custom tone
- Inserting custom html table pattern where it fits, inserting custom html code for making text blocks, making things stand out, etc...
- Avoid the AI Jargon / useless fluff
- Avoid heavy repetition at word, sentence & paragraph level.
- Each H2 must always be different and cannot regurgigate content that has already been addressed previously.
- Added some personality to the writing. (Punchy lines, talking directly to the reader, including fake experiences, etc)...
The above IS NOT a prompt, i'm just giving SOME examples of things I've worked on with prompt engineering.
The results is that websites like Ahrefs doesn't detect my content as AI at all, it only detects it as "Possibly Light AI Assistance". Other AI detection tools detect between 0-20% which is very very good. The only tool that detects it as AI is "Originility.ai" but after testing this website is actually so inaccurate it's not even worth thinking about. So many articles are detected as AI despite being very clearly human and even old articles I wrote before the time of AI show as AI written.
Overall I'm extremely happy with the results. It's literally everything I could ever ask for, it's the type of article I'd pay really well to have prior to AI assistance a few years ago.
My Current Focus & Backlinking
Now that prompting is out of the way, the workflow is gucci and the framework is gucci, I decided to now move my focus towards:
- Scaling horizontally (build more sites instead of pumping more articles & risking velocity penalties)
- Getting Backlinks (My weakness)
Scaling Horizontally
I decided to scale horizontally as in, building more websites instead of ramping up my article posting velocity because It's simply not realistic for small websites that aren't owned by big companies to be pumping 10-20k+ words of content every single day, the same way it isn't natural for new websites to get tons of links unless it's from big companies.
So instead of posting 6 articles a day (3 per website) and then go AFK the rest of the day, I decided to simply create more websites in different niches and begin posting to those websites aswell.
I haven't began keyword research yet but will do so tomorrow and will start working on those websites.
Backlinks
So backlinks has always been a pain in the ass for me, they're extremely important but good ones are just hard to get. It's also the one area of SEO I have the least experience in.
Since my articles are so good I am hoping that they'll get some natural links over time but I'll also try to put in some effort to grab some.
Over the next few days / weeks I'll be trying to create a program that will scrape google for specific keywords & try to find contact pages so I can start reaching out for guest posts or link exchanges.
I don't know if there's a better way to do it or not but if there is feel free to let me know as I'm always looking to learn!
Regardless, I've always been successfull without much if any link building so even if I'm not successfull at outreach I'm sure i'll still succeed with just good content and proper on-page SEO + good keyword research.
Not including much stuff website related since again, still early wont have anything exciting to share about them for a while. I'll share their stats once a week. Instead I'm gonna focus on sharing some wisdom/knowledge i've picked up along the way. This might be a little bit long.
Splitting this in 2 Different Sections:
- Knowledge Dump (Making AI Content Undetectable)
- My Current Focus & Backlinking
Goal: Make AI Content Undetectable
Prompting Engineering
Prompt engineering is the strategic process of designing, refining, and optimizing text-based inputs (prompts) to guide generative AI models toward producing accurate, relevant, and high-quality outputs.
I put a lot of emphasis in ensuring that:
- My content was EXTREMELY high quality
- Instantly give the reader the answer they were looking for (Fulfilling user's search intent)
- In-depth Explanation of the subject
- Introduces topical clustering & semantic words into the article
- Using a well known Framework: Problem–Solution–Proof–Application
- The AI doesn't sound like the AI by making it significantly different from the AI Standard.
The above not only fulfills the objective of all articles (providing value to the reader & answering the keyword's search intent) but it also ensures that none of my articles can ever be penalized for "thin content".
All of that can and is easily achieved by Prompting Engineering.
Afterwards I wanted to ensure that the text doesn't get detected by AI detectors and therefore, no issues with Google's Anti-AI Updates. For that I focused on:
- Breaking common AI Patterns (Avoiding words that AI overuses, em-dashes, etc...)
- Humanize AI Writting Patterns (Varying sentence lengths, etc...)
- Custom tone
- Inserting custom html table pattern where it fits, inserting custom html code for making text blocks, making things stand out, etc...
- Avoid the AI Jargon / useless fluff
- Avoid heavy repetition at word, sentence & paragraph level.
- Each H2 must always be different and cannot regurgigate content that has already been addressed previously.
- Added some personality to the writing. (Punchy lines, talking directly to the reader, including fake experiences, etc)...
The above IS NOT a prompt, i'm just giving SOME examples of things I've worked on with prompt engineering.
The results is that websites like Ahrefs doesn't detect my content as AI at all, it only detects it as "Possibly Light AI Assistance". Other AI detection tools detect between 0-20% which is very very good. The only tool that detects it as AI is "Originility.ai" but after testing this website is actually so inaccurate it's not even worth thinking about. So many articles are detected as AI despite being very clearly human and even old articles I wrote before the time of AI show as AI written.
Overall I'm extremely happy with the results. It's literally everything I could ever ask for, it's the type of article I'd pay really well to have prior to AI assistance a few years ago.
My Current Focus & Backlinking
Now that prompting is out of the way, the workflow is gucci and the framework is gucci, I decided to now move my focus towards:
- Scaling horizontally (build more sites instead of pumping more articles & risking velocity penalties)
- Getting Backlinks (My weakness)
Scaling Horizontally
I decided to scale horizontally as in, building more websites instead of ramping up my article posting velocity because It's simply not realistic for small websites that aren't owned by big companies to be pumping 10-20k+ words of content every single day, the same way it isn't natural for new websites to get tons of links unless it's from big companies.
So instead of posting 6 articles a day (3 per website) and then go AFK the rest of the day, I decided to simply create more websites in different niches and begin posting to those websites aswell.
I haven't began keyword research yet but will do so tomorrow and will start working on those websites.
Backlinks
So backlinks has always been a pain in the ass for me, they're extremely important but good ones are just hard to get. It's also the one area of SEO I have the least experience in.
Since my articles are so good I am hoping that they'll get some natural links over time but I'll also try to put in some effort to grab some.
Over the next few days / weeks I'll be trying to create a program that will scrape google for specific keywords & try to find contact pages so I can start reaching out for guest posts or link exchanges.
I don't know if there's a better way to do it or not but if there is feel free to let me know as I'm always looking to learn!
Regardless, I've always been successfull without much if any link building so even if I'm not successfull at outreach I'm sure i'll still succeed with just good content and proper on-page SEO + good keyword research.
chrischris15
Registered Member
- May 5, 2014
- 95
- 55
From a page speed/hosting standpoint I've found static html pages are super fast and you can host for free in some places.
For interlinking I found the best way to do that is produce all the articles at once so you can figure out pillars, silos, and interlinking in one shot.
For interlinking I found the best way to do that is produce all the articles at once so you can figure out pillars, silos, and interlinking in one shot.
- Sep 25, 2013
- 4,698
- 2,694
Yea that for sure helps with interlinking, right now the way it works is I get my AI to scan my website and check for highly relevant posts to my "new" article and then add those as interlinking opportunities in the article.
This is really good but also kinda bad because:
- Older articles get tons of interlinking = Higher rankings
- Newer articles get 0 interlinking = Less rankings + Longer to Index
Ideally I'd also update old articles to interlink to newer articles "if they fit" but it's not something I've worked on just yet. Maybe in the future I will.