In this presentation, Michał explores the struggles with generative AI tools and the reasons why AI content fails in the long run and provides a case study of an AI content blog recovery.
The main struggle of generative AI tools is that they can create something nice but it will be nice only on the surface. With AI, it is difficult to create something exactly as you want it. If you want to feed it with so many facts, with so many entities, with so many points that you really want to get in the order, you will have trouble.
To get great results, you need three areas covered:
There are three things that may stop you:
It will be all about keyword clustering and the proper data-driven research for topical authority, not research for the search volume. It looks good on the reports when we target only the bold keywords and so on. But if you want to build topical authority, if you want to create a really comprehensive source of information about your project and niche, it is really important to do keyword research not based on the search volume but based on the topic and covering as many pages as you can.
It’s really critical to avoid cannibalization and write a lot of articles that will target the proper group of keywords. These keywords should be targeted together. If there is an overlap in search results, Google displays the same articles answering these two keywords – just join them. There are tools for it or you can do it on your own as well.
The first phase is the analysis and the second is obviously the execution. When it comes to analysis, you need to figure out the money pages while doing the research and gather all the keywords around. All the information content you can think of should be gathered, grouped, and planned as supportive pages. We also have to leverage the internal linking.
When it comes to the execution, just make sure that one keyword is only in one cluster and one cluster is only one article, just to avoid cannibalization and other issues.
Another tip would be about the internal linking anchors. Keywords that are in the cluster should be anchors pointing to that page. If you do this properly, you will get Google to understand what’s on the other side of the link, and which page should rank for that keyword. Cover all the topics no matter what their search volume is.
Great clustering tools:
Compare all the keywords. Find neighbors. Analyze semantics. And after that, semantic analysis kicks in. This is how connections between keywords look like:
It’s about data for optimization, not fresh concepts and news. You need entities and topics that are covered in pages that are currently ranking in Google. You need facts and statements, which are basically the bits of information that these websites are sharing, so you can replicate or inspire the AI. And then you need the user intent, which is obviously essential, and AI is not going to take you far without it.
So, you just need to crawl the SERP because AI can read and it can segment the gathered information. It also can get inspired by what it learned, just like when writers are doing research before writing an article.
ChatGPT + Web Pilot is extremely helpful for gathering information from the web. You can ask Web Pilot to read a website or a piece of content, and it will create a summary. It can create a list of entities, a list of questions, a list of definitions, or everything that could be found in a brief for a real writer.
If you are using plugins for web browsing in ChatGPT, make sure to use them the right way. Create a knowledge base for the article and then use that knowledge base in the generation process. Avoid a situation of generating the article with a single prompt. It will be limited and the longer the prompt, the trickier it gets to have all the entities, questions, and answers found in the article, to have a coherent structure.
Lastly, don’t fall for cheap AIs. You will spend much more money to fix repetitions, hallucinations, facts, and so on. It is OK only for PBN content. It’s cheaper only if you don’t value the time.
You can check the cost comparison and see it for yourself:
