What is GEO optimisation — and what really happened to search clicks?

GEO optimisation refers to editing content so that artificial intelligence will cite it in its responses. The term originates from a research paper from 2024, and it is already being marketed as a standalone service in Finland. This article covers two points: what GEO actually means, and the one measurable change that actually occurred in search clicks — when Google displays a summary generated by artificial intelligence, the number of clicks on search results drops by about half.

What does GEO optimisation mean?

GEO is an abbreviation for Generative Engine Optimisation. The term is not a marketing invention but stems from research: the paper by Pranjal Aggarwal and colleagues, ‘GEO: Generative Engine Optimisation’, introduced the term in 2023 and was published at the ACM SIGKDD conference in 2024. It was the first study to demonstrate, in a controlled experiment, that content can be modified so that AI-generated responses cite it more frequently.

In practice, GEO generally sells three things: content formatting for AI extraction, the addition of source references and quotations, and technical files such as llms.txt. The first has research evidence to back it up, whilst the third has none whatsoever.

First things first: Google states in its own guidelines that optimisation for generative search is search engine optimisation – not a separate discipline. We covered this in detail in our article ‘AI optimisation: what’s being sold and what’s true’. This article focuses on another question: what actually happened in terms of clicks.

That one proper study

There are many figures circulating on this subject, most of which are the individual observations of specific websites without a control group. One published study stands out: the Pew Research Centre tracked the browsing behaviour of 900 US adults in March 2025. Participants installed an app that recorded their actual browsing behaviour — so this is not a survey, nor is it based on analytics from a single website.

The dataset contained 68,879 Google searches, of which 12,593 produced a summary generated by artificial intelligence. This represents approximately 18 per cent of the searches. Results:

What was measured	Summary displayed	No summary
The user clicked on a search result	8 %	15 %
The user clicked on the source link in the summary	1 %	—
Your search ended here	26 %	16 %

Three figures, three different points. The first shows that clicks have become about half as frequent. The second shows that appearing in the summary does not make up for the lost click — a one per cent click-through rate is effectively zero. The third is the most interesting of the three: the summary did not direct the user elsewhere; it ended the search.

What this does not mean

The figure of 8 per cent versus 15 per cent has been circulating online in the form of ‘AI took half the traffic’. That is not the same thing. Three qualifications that are part of the figure but get lost along the way:

• The summary appeared in around 18 per cent of searches, not in all of them. The remaining 82 per cent behaved as before. The overall impact is therefore much smaller than the difference in a single search.
• The data is from the US and is in English. No comparable figures have been published for searches in Finnish. The situation in Finland may be better or worse — nobody who hasn’t measured it can know for sure.
• That was the situation in March 2025. The search results page has changed several times since then.

If anyone is selling you a service in this day and age without mentioning any of these three, they haven’t read the source material.

Which search queries lose clicks and which don’t

At this point, we move from measured data to inferences, and this is stated explicitly. Pew did not distinguish between different types of search, so the following is based on where the summary appears in the first place — not on the measurement itself.

An AI summary works well for questions that have a clear answer. ‘How long does concrete take to dry?’ is one such question. The user gets the answer and doesn’t need the website. However, a summary cannot:

• sell a product — purchases are still made in-store
• Book an appointment or request a quote
• Let us know if your company has any available slots next week
• to address the fact that customers want to see examples of previous work before they ring

Practical implication: content providing general information loses the most, whilst content leading to a purchase or enquiry loses the least. If the site’s traffic comes from guides, the change will be noticeable. If it comes from searches for the creator, the change will be smaller.

Assess your own situation — it’s free

You don’t have to believe the US figures or us. Google Search Console tells you how your own site is performing, and it’s free. Take a look at these three:

What are you looking at?	What does that tell us?
Page views vs. clicks, 16-month timeline	If page views remain steady but clicks are falling, you’re still getting views but not the clicks
The same content, categorised by search terms	You can see which topics lost out and which didn’t. The difference is usually significant
Breakdown by page	Discusses whether it is still worth writing guides or whether the work should be outsourced

One caveat regarding interpretation: Search Console does not specify whether the AI summary was displayed. The comparison is therefore before and after, not side-by-side — and many other factors may have changed in the meantime. That is why a breakdown by search term is more important than the overall figure.

What does this mean?

• Do not purchase a separate AI visibility package based on this chapter. The same work that boosts search results also boosts the sources in the summary. The reasoning is set out in a separate article.
• Shift the focus to content that a summary cannot replace. Prices, availability, work done, your local area, your way of doing things.
• Measure before you decide. If your bill is small, this whole discussion is purely theoretical in your case.
• Don’t stop writing. A summary needs a source. If you’re not mentioned anywhere, you won’t be in the summary either.

In demand

Is there a corresponding figure for searches in Finnish?

Not published. We have not found any data that includes both a Finnish-language sample and a control group. That is why we use the US figure and state it clearly each time. You can find the figure for your own website in Search Console, and that is the only figure that applies to you.

Should AI bots be blocked from the site?

Rarely. Once blocked, the site cannot appear in the search results that some users now use as their starting point — and it does not generate clicks, as users do not return to the search results. Blocking is mainly justified when the content is pay-to-view or when server load is a proven problem.

Does the AI summary appear in all search results?

No. In Pew’s data, the summary appeared in around 18 per cent of searches (12,593 out of 68,879). The proportion varies depending on the type of search and the time of year, and it has increased since the survey was conducted — but it has never been present ‘in all searches’ at any stage.

Should I stop writing guides?

Don’t give up; just choose differently. A guide that explains a general topic in general terms is exactly what a summary does best. A guide that offers its own price, its own assessment, its own example or its own opinion is what a summary cannot provide — and at the same time, it is precisely what the summary refers to as its source.