Insights and highlights from SEOloger naživo 2024

For me, SEOloger combines business with pleasure: swapping funny work stories, inspiration, meeting new people, catching up with familiar faces, talks, walks in nature and wellness. So what was on the programme this year?

Something dense and clever about ChatGPT

The opening talk was by David Grudl, Something dense and clever about ChatGPT. I use ChatGPT every day, but mostly as an assistant to broaden my perspective or for research. I don’t get into any bigger projects or integrations. I also have a few favourite prompts, and that’s it.

The key part of this talk for me was the explanation of neural networks and the genesis of ChatGPT. The talk connected some of the dots for me really well, but above all it gave me the enthusiasm to dive in a little deeper again and explore large language models from a different angle.

And for me, that is also one of the main points of events like SEOloger: gaining new enthusiasm for work and new impulses.

ChatGPT-3.5 has gone from a text completer to a model that can follow instructions. That is the biggest change compared with previous models, which relied only on probability and tried to predict the next most likely word.

David Grudl

David also talked about the following (I’m copying the decipherable and presentable notes from Google Keep):

• All words are tokenised. What tokenisation looks like for English compared with Czech.
• Czech is tokenised letter by letter, which makes it twice as expensive and twice as slow.
• Output is generated token by token and every step is equally fast; it’s only down to the internet. So the answer does not depend on whether the query is complex or trivial.
• Setting the level of randomness is important so that the model does not produce repetitive, cyclical text (ideally around 0.75). Because of this randomness, however, there is a chance that you will occasionally get some nonsense. For example, when it writes the word “powdered”, there is about a 0.8% chance that it will choose “salt” as the next word. If that happens, it is almost certain to make the whole answer quite misleading. So randomness is important, but it can produce nonsense.
• Ask about things that are mainstream in the training data, or provide plenty of context.
• Everything about the context window and how it developed, which is information I have been working with for some time.

The death of SEO texts

I spend a lot of time creating the best category content and researching search intent, and this talk gave me yet another perspective, even though I was already familiar with most of the points.

You can watch the whole talk on the SEOloger website.

How not to mess up SEO implementation

For me, the key point of the talk was how to set up and structure a collaboration. Simply put, have a collaboration process and keep refining it so that the client moves forward and feels good about working together.

The second point was formal errors. All SEO consultants flag formal errors as problems, but it is important to recognise and estimate the effort and benefit involved. These formal errors usually have no impact on SEO performance and only cost the client money.

SEO audits are a narrow view held by SEO specialists

At the start of a collaboration, the first topic is decisive. You need to solve it and use it to kick off the whole collaboration. Doing SEO audits and SEO analyses is pointless if, even after two months of working together, the client doesn’t feel that anything is really happening on their website.

It’s good to tell the client that if an SEO specialist works 20 hours, it might create, say, 100 hours of work on the client’s side (development, implementation, management).

Probably every SEO consultant has run into something similar. The client doesn’t have the capacity or the people to implement changes on the website.

BigQuery for working with GSC data

I use Google Search Console very often and it generally covers my needs. This talk was for more advanced users and showed how to pull data out of GSC.

If you want to get started with this topic, I recommend the materials from the September talk. The first step isn’t difficult, and afterwards you’ll see whether it’s for you. I’ve taken the first step, but at the level of detail I need, GSC is still enough for me. Of course, there are drawbacks, which Martin also describes in that article.

Google Discover optimisation

Google Discover is a black box that works wonders for some and not at all for others. There are now dozens of recommendations that are supposed to lead to the coveted traffic from this “source”. It’s interesting how much traffic Google Discovery distributes and how important it is for large magazines.

The talk covered those tips on how to get into Google Discovery more often, mainly from the media’s perspective. Some news websites get up to 90% of their traffic from it compared with traditional search. The big problem is that this traffic is completely unreliable.

---

And finally, just a panel discussion and some fortune-telling about SGE. And that’s it.

What was SEOloger 2024 like?

Photo report and source of the photos for this article by Petr Krauz

Report from last year