Mr Khan
Newbie
- Oct 15, 2016
- 47
- 10
After AI era, I thought let's start a tech blog and see what is working and what is not. First we build hand made content later we shifted with Claude AI paid plan and created similar content but long form. Also we first made feature images and other needful images using Canva later we shifted to ChatGPT and made same also used Gemini and build some Python tool to add our copyright on each images.
What we did in starting
In starting we published articles per week with hand written content only. Quality was good but speed was very slow, one writer was taking 3-4 hours for one 1200 word article. Cost was also high and output was not enough to cover our keyword database. So after 2 months we decided to test AI.
Shifting to Claude
We took the paid plan and made our own prompt system, not simple "write me article" type prompt. We gave it our own outline, our own tone, and told it to avoid AI patterns like "in today's digital world", "moreover", "delve" and all. Then our editor was doing final pass, adding real screenshots, real steps and fixing anything which sounds robotic. Output went from 5 articles per month to 25 articles per month with same team size. Word count also increased, earlier 1200-1500, after shift we started publishing 2000-3000 word guides.
Images part
Canva was fine but it was manual work, one feature image was taking 10-15 minutes. Then we moved to ChatGPT and Gemini for image generation. Gemini was giving better results for real looking product and device images, ChatGPT was better for illustration and concept type images. But problem was our images were getting stolen and used on other blogs. So we build one small Python tool which takes the full folder, adds our logo and copyright text in corner, compress it also and export in WebP. Now one batch of 50 images is done in less than a minute.
What worked
- Long form content ranked much better than our short hand written articles, same keyword, big difference in position.
- Error and troubleshooting keywords ("how to fix", "not working", error codes) were getting traffic very fast, low competition and high intent.
- Internal linking between related error pages, this gave good boost after 2-3 months.
- WebP + our own hosted images, page speed stayed under seconds even with 8-10 images per post.
What did not work
- Pure AI content without editing, we tested some posts like this in starting and they were either not indexing or ranking and dropping after few weeks.
- News type content, we were late every time compared to big sites, so we stopped it completely and focused only on evergreen guides.
- Very broad keywords, we wasted articles on high competition topics where we had no chance with new domain.
- AI images in review type articles, readers can catch it, better to use real screenshots there.
Result
After 100 months we reached 3500 monthly sessions and got AdSense approval in 4.5 month. Cost per article came down from 500$ to 50$, and quality is better than before because now our writer is spending time on editing and adding real value instead of writing basic paragraphs.
Main learning
AI is not replacing the writer, it is replacing the first draft. Whoever is doing only prompt and publish will not survive, but whoever is using AI for speed and human for accuracy, that combination is working very well. Same for images, generate fast but always protect it and always keep your own branding on it.