AI Content Pipeline: My Experience

Many AI systems have emerged. You got curious too. You wanted to create content and generate a blog post with AI. But your wish isn’t something overly artificial. If a human wrote it, it would be full and rich, perhaps supported with images and infographics. The n8n I mentioned before is tailor-made for this! Why? Let me explain:

Data collecting

n8n can collect data as you wish with a webscraper or RSS. Let’s say you chose the easier route: RSS. You got the URL data via RSS. n8n can make an HTML request to the relevant page with its web request component. In fact, with HTML Extract, it can take the page source, detect the content holder from within, and instantly get the content.

You can either clean this data and create training data for yourself, or you can have it rewritten according to rules you define. Rewrite? How will this happen? Yes, your path has led to an LLM’s door. Now you can easily get a token from any LLM you use and continue. Moreover, n8n even documents how to do this for you. You got the API Token and got started. Along with a nice User-System prompt… Great? No, it’s not. Because the API pricing of the LLM you’re using is different from the pricing you’ve done on web and client. Your monthly payment doesn’t cover the API. -Bad news :(- APIs generally work with a pay-as-you-go logic. So what should you do? You should use a cheaper but definitely stable LLM, but how? This is where great solutions like GROQ come into play. It has the same stability as your LLM but hosts many models much cheaper and doesn’t send you a bill that makes you regret it at the end of the month.

Data processsing

Now the content is ready. But what about the image/images - if the content you’re going to make is going to be so boring and monotonous that it’s just plain text, I have nothing to say, but the internet user decides how something is by looking at its cover first. Isn’t there a solution like GROQ for this? Actually, there is. There’s a great site called REPLICATE. Here, all the image generation models so far and their pricing are written. Moreover, using its API is child’s play. But one thing is very important: price/performance. Because the data you give will extract an image prompt from your content, and from there you’ll go to REPLICATE with the prompt you have. But to which model? This is where pricing comes in. For a quality eye-catching, realistic, even creative image solution, you need to spend some money. Otherwise, the images don’t turn out very attractive. (You can choose flex/schnell to be cheap, but there’s not much difference between creating an image with flex/schnell and having a kindergarten student draw it :D)

It’s not really possible to enter without calculating costs, is it? So will you publish this data as internet content? Where? On your own site. What about its social media marketing? X, LinkedIn, Reddit, Hackernews, Bluesky… which one? With what parameters? When? Yes, now you’re also working on social media marketing. Frankly, this is the most challenging step. (Like an end-of-chapter boss) Because it’s not important that you send your content to these social media platforms. It’s also important that you comply with their publishing principles. Otherwise, you’ll get banned. In fact, you often need to do this manually. Or you can train a real AI to do this. I haven’t seen such a project yet. Social media marketing editor AI :) Actually a good idea.

Final word… For people who think AI is zero cost or very low cost, this experience ends with disappointment. Like the bosses who misunderstand artificial intelligence because they chase zero or very low costs.