The AI Algorithm: A 2025 Google Algorithm Update Retrospective

2025 really was the year when “Organic Search” stopped being about keywords and search terms, and moved towards understanding & learning – machine learning, that is.

With the rise of Large Language Models (or LLMs) for short, the way in which people search the internet has changed drastically in just a few months! It was only last year when the SEO industry had to accept that TikTok was catching up to Google & YouTube in terms of “market share” of search engines. Yet here we are, not even 12 months later, and search behaviour has shifted significantly towards using the likes of ChatGPT instead.

The phrase “let me google that for you”, once so popular there was even a meme website made using it, has been replaced with “let me ask ChatGPT”.

Because of this, 2025 was the year where Google also made strides towards this new way of searching, in a bid to not get left in the dust like they did to Yahoo and AskJeeves.

Setting the Stage: From Helpful Content to Hyper-Intelligent Context

Almost as if it were in preparation for this science-fiction like turn of events, Google has been updating and changing it’s algorithm behaviours drastically as far back as 2022. With the advent of the “Helpful Content” update, the increased focus on Core Web Vitals and the ongoing evolution of E-E-A-T (expertise, experience, authoritativeness and trust), it was as if they predicted just how much search behavior was going to shift.

Then, in 2025, Google clearly realised just how much of an impact LLMs like ChatGPT were having on the industry, releasing Gemini and incorporating it heavily into Search Results.

With the advent of the “AI Summary” in the standard search result page, we saw a shift in the Google algorithms away from the typical “ranking signals” that we have all gotten used to. Instead, “behavioural prediction” and “semantic understanding” became huge talking points in the performance of a website in Organic Search.

But what do these terms actually mean?

How Has AI Changed Search?

The terms “behavioural prediction” and “semantic understanding”, along with a tonne of other phrases coined across the SEO industry, are just umbrella terms used to describe three core changes to the way we optimise websites.

Search Becomes Conversational

Back in 2013, Google released the Hummingbird update to their core algorithm, which at the time was a game changer. It meant that their algorithm could, to an extent, understand the context and meaning of a search term, rather than just look at each keyword in it.

Then, with the introduction of “Quick Answers” and Rich Snippets around a year later, the way we optimised content for SEO took a shift towards a more conversational tone. No longer was it best to write an article specifically targeting the keyword “seo agency in London”, for example. Instead, the “focus keyword” became the question “where can I find an SEO agency in London?”

It’s funny to think that, 12 years later, this at least appears to still hold true as a key part of how Search continues to grow. If anything, it’s actually become more important.

It was the birth of Alexa, Cortana and Siri that shifted focus towards more “conversational” search terms, since people would actually ask questions to these devices. That led to a wave of people, myself included, talking about how “voice search” was the next big thing and that we needed to prepare our SEO strategies for it.

Little did we know that we were actually half right.

Whilst “voice search” as it was thought of back then never really took off in the way we expected, the influx of people now searching in LLMs mirrors the behaviour seen in voice search.

We don’t go to ChatGPT and type in “Indian takeaway near me” like we do for Google. Instead, we ask ChatGTP “can you tell me the best Indian takeaway near Nottingham central station”. We don’t “search” with keywords when it comes to LLMs; we converse with them just like we did with Alexa, Cortana and Siri.

And that’s where the phrase “semantic understanding” comes into play. In essence, it means that we need to think of the different ways in which terms and phrases can be understood, because that’s exactly what LLMs are learning. Focusing our attention on individual keywords will likely be a recipe for disaster in the long term as the LLMs are learning how people actually speak and ask questions.

We need to shift our thought processes back to when we were looking at voice search, and how people converse with each other, as that is the future of search when it comes to using LLMs like ChatGPT.

Real-Time Changes To Search Results

Moving on to the phrase “behavioural prediction”, we need to understand how LLMs are constantly learning and reacting to human behaviour.

Because LLMs like ChatGPT learn from the questions we ask them as much as the information they find online, they are able to analyse and predict trends in the way we search for information. From there, the LLMs adjust how they respond and what answers they provide based on their predictions of what we want.

Historically, these changes and updates happened with new algorithm updates, be it every few months or a couple of times per year.

Now, LLMs are able to change and update their response choices in real time, at the exact moment they make a new prediction based on their learnings. This creates the potential for an even faster changing industry than SEO already was – and SEO was already considered one of the fastest changing industries out there.

Add into this the fact that LLMs also “index” (to use a potentially outdated term now) content of every type, including video, audio and images, and we have a whole new spectrum of ways to optimise our content.

Therefore, when it comes to an SEO strategy for any business, now more than ever, we need to be agile and not afraid to completely change strategy part way through. Sticking to a pre-determined 12 month plan like many did before could actually hinder your performance because the way LLMs answer questions could have changed countless times in that 12 month period.

For example, if we were to estimate a change in LLM responses happening once every hour, that would equate to 8,736 “updates” in 12 months. Compare that to the average of 8 significant algorithm updates per year that Google has made previously. And keep in mind that this number would be a very conservative estimate too, because the LLMs can change everything in real time.

How Can We Prepare For AI Optimisation?

Now, that was a lot to take in, I know. You’re probably also now wondering how you can actually prepare for the growth of AI Optimisation and the potentially rapid changes in SEO that we will face in the future.

Tracking LLM Traffic

Well, the first thing to look into is how to actually track traffic levels and performance for people coming to your site from LLMs. To do this, you need to create a custom-defined channel group in Google Analytics. This can be done by using the following Regex expression to define the new channel group type:

There are already a number of tutorials out there on how to implement this, such as this one from SearchEngineLand.

Once that’s in place, you’ll be able to see all of the data for LLM traffic that you normally would for different channels, such as event data, conversion rates, purchases and more.

This will, for now at least, form the basis of your ongoing performance tracking. Since the developers of LLMs have already stated that even they aren’t sure exactly why the various LLMs learn and respond the way they do, there’s no truly accurate tracking tool out there at the time of writing this article. So instead, utilising Google Analytics to understand what pages users are being referred to can be used to estimate the kind of questions they are asking to LLMs.

Focus On Intent

Many SEOs out there have been pushing the idea of focusing on “User Intent” over just promoting your business or services for a while now. However, it is now more important than it ever has been.

As people converse with LLMs rather than “searching” with them, the LLMs are able to fully understand what the user wants to do. They understand the intent behind the questions that the user is asking. Because of this, you need to figure out what users want to do on your website, and make it as easy as possible for them to do that.

For example, if you run an ecommerce website selling t-shirts, when a user comes to your website they are intending to buy a t-shirt. So you need to make sure that the product pages are clear and understandable, the site is easy to navigate and the checkout process is smooth, quick and easy. On top of that, you should try and get reviews and user-generated content on other websites to mention how easy it is to buy from you.

All of this will be found by the LLMs, and they will learn that your website allows people to complete their intended action (in this example, the purchase of a t-shirt) easily. From there, the LLMs are more likely to recommend your website in their responses.

Prepare For Agility

This one is more of a mindset change than an actual strategy change. LLMs are still in their infancy, and as mentioned earlier, they are able to update their response methods in real time. So we all need to be prepared to throw away a planned strategy and create a new one at almost any moment.

Moving Beyond The Algorithm

That was a lot of information to absorb, I know. It might be a good idea to bookmark this article so you can come back to it later and reference it if you need to. But that’s because the growth in LLMs and the change in user search behaviour has been so significant that there’s a lot to decompress.

If it all seems a bit too daunting, or if you feel like you’re going to need help with running agile SEO strategies around AI Optimisation, feel free to get in touch with us today and we can discuss what would be most suitable for you and your business.