About the idea of reserch for generating AI texts filled with facts

About the idea of reserch for generating AI texts filled with facts

So that my content generation isn't just based on rewriting the top 10 and AI knowledge - I do reserchs.

It's not about the task of «fooling the search engine» here, but rather something I'm genuinely convinced of: with this approach the texts are of higher quality and more interesting.

This is essentially how academic copywriters work: their job is to find sources and make material from them, not to do their own research.

What is the essence of deep reserch:

• organise the outline of the article into topics and subtopics
• send to reserch either the whole plan at once or in parts - depends on the reserch system
• For each topic and sub-topic, find materials, studies, case studies, etc.

After that, the results of the reserch can already be translated into an article.

Where to do the reserch

Google's Deep Research is, in my opinion, one of the best systems available on the market right now. You can use it in Gemini as well as in NotebookLM.

Promt for reserch can be about this:

Your job is to break down my article outline into topics and subtopics to find information on each.
Look for materials in sources: {language, sites}.
Types of materials: {cases, studies, and so on}.
What is required to extract: {facts, quotes, thoughts, figures}.

That's enough for 90% tasks if you spell out exactly what you want normally.

And don't fuck your brain with complex promts right now: AIs have smartened up a lot, and their level of understanding has increased a lot.

Lifehack on how to make a simple dip-research system for your AI pipelines

In general, there are ready-made frameworks for AI-research, including those of major market players, but they can be expensive to implement, use and maintain - this should be tested separately. According to our tests, in some cases it is cheaper to use Perplexity for reserch.

Perplexity reserch scheme in n8n:
https://skr.sh/saP57T7PNHG

To automate client tasks, we sometimes do roughly the same scheme, but on more stable systems written in programming languages.

How this reserch algorithm works:

• article structure
• make AI-step of splitting the article into topics for reserch and convert topics into mini-prompts for information retrieval - get n topics for reserch
• calling Perplexity for the most important reserch on the subject.
• then run a loop of calls to cheaper modes or Perplexity models to search for less relevant topics