What is the impact of generative A.I. on search engines? 🎯
🤖 Revolution or mere evolution?
Behind the promises, more nuanced realities emerge. In this first article, we explore how search engines are (really) integrating A.I. and what it truly changes – or not – for your digital strategy.
La place de l’I.A. générative dans les moteurs de recherche
The arrival of Large Language Models (LLM) has revolutionized access to knowledge and information retrieval. The weaknesses of a technology based on generating the most probable output are obvious, as this approach does not guarantee the most relevant or true output. However, in many cases, the result effectively meets users’ needs. The quality of results also improves week by week with constant evolutions in algorithms.
It is therefore not surprising that search engines, especially Google, which holds the largest market share, quickly followed OpenAI‘s lead to offer this innovation by integrating it into their solutions.
However, we must give Google its due credit. The beginnings of generative A.I. were, in part, made possible by some of their work on the attention model and transformer technology (BERT, MUM).
Examples of A.I. in Search
These examples illustrate how search engines implement AI-generated results in their search engine.
The fundamental difference, if we take, for example, Google (which, let’s remember, holds more than 90% of the market share in Search) is that, where traditional results gave a first position, the A.I. overviews are a condensed version of texts, links, and images generated from multiple results. There is no longer really a direct “winner.” The competition can therefore be more complex.
Why generative A.I. in search engines is not a revolution?
1
Not every search generates an A.I. result
During the beta that was only available in the U.S., we had the opportunity (thanks to VPNs) to extensively test Google’s A.I. responses. While at that time, a large number of searches ended up generating A.I. texts, this number seems to have decreased significantly. Recently (March 2025), Sistrix & Ziptie, two SEO tools, analyzed the evolution of A.I. overviews in the United Kingdom. These appear to be growing strongly, even if they remain fewer than in the U.S. Moreover, these recent studies confirm that they remain strongly linked to informational queries.
Additionally, certain domains seem highly impermeable to this type of generation (and fortunately so). For example, all Your Money Your Life (insurance, health, investment) topics generate few A.I. results. It seems that the backlash Google received due to some rather unreliable results (the glue cheese, thanks to Reddit) has forced them to take precautions.
2
A.I. generated results remain heavily based on the first results
While during the early stages of the beta, some results used in text generation didn’t even seem to be ranked in a traditional search, current A.I. results seem to be extremely linked to traditional results. In other words, a good position in Google’s basic algorithm allows you to appear in A.I. overviews.
3
We’re still waiting for a release date in French/Dutch
While SearchGPT (which we’ll talk about soon) has already been released, Google still hasn’t communicated a clear release date for its solution in our national languages. If we assume this delay is caused by legal issues (GDPR? Copyright?), it doesn’t prevent it from being difficult to prepare a clear plan for a solution whose direct impact or even scope is complicated to calculate. It is nevertheless interesting to note that Google has recently launched these in Belgium in English.
4
Does Google have the desire and need to provide A.I. results everywhere?
In the current discussion, we often hear that Google is lagging behind other solutions and that its integration of generative A.I. is clunky compared to other solutions. Those who have tested the pro version of Gemini 2 in API know that, although OpenAI remains at the head of the pack, Google doesn’t necessarily have to be ashamed. It’s also always interesting to remember that they master the technologies behind generative A.I. models and that certain elements at the basis of models like ChatGPT have been used since BERT (Transformers / attention) in their tool.
So it’s interesting to ask “why”, in this case, A.I. overviews are restricted to informational content and why Google isn’t working harder and faster to improve them.
First, it remains important to remember that Google’s model for Search is still, unsurprisingly, based on advertising. Paid is the lifeblood and organic is, ultimately, just a way to build user loyalty (and this is an SEO telling you that). While Google has already tested paid results in A.I. overviews, we wonder if the problem is with the cost of an A.I. result.
For Google, the process involves extracting X relevant pages from its index and then using them to generate a result. So there is an input and output cost. If a traditional search costs 0.0001 cents and has the power to generate, on average, 0.05 cents, Google is able to generate a significant margin. If, on the other hand, an A.I. search costs 0.001 cents and has the power to generate only 0.005 cents, because the answer “limits” the power of advertising, it’s clear that Google has every interest in slowing down as long as possible before fully embracing this model. Even if the given figures are completely made up, it remains interesting to keep in mind that generating a convincing result through generative A.I. is not difficult, but it does have a cost that not everyone may be ready to pay. Not to mention the environmental impact of technologies related to generative AI.
✨ Despite all these elements, this type of technology remains capable of disrupting certain things, particularly the way consumers find and consume information. This offers opportunities for brands that will find the right approaches to position themselves throughout the customer journey.
How generative A.I. could disrupt Search?
1
Transactional searches could be completely transformed
Although Google search pages already tend, in some cases, to look like e-commerce category pages with filters (a quick look at US pages only reinforces this statement), the ability of generative AI to analyze an input (the user’s search) and propose the most statistically probable output (a list of products extracted from a web search) allows it to generate convincing results even for extremely complex product searches (512GB smartphone with 1GB processor and RAM).
During the beta, AI overviews generated relevant product feeds. Even if it seems that, in the current version, this type of output has decreased somewhat (or at least is less relevant), it doesn’t prevent Google from being very interested in moving towards this kind of output to act as a general e-commerce platform and convert users within its environment (where the user is also exposed to Ads that generate revenue). They have an opportunity to provide results that actively compete with Amazon, for example.
Furthermore, for comparison searches (which tools like SearchGPT or Perplexity already do extremely well), Google could outmaneuver many comparison sites (car insurance, energy contracts, etc.) by directly comparing brands in its search engine (e.g.: “what is the best insurer for an 18-year-old new driver”) thanks to their enormous amount of content and their 25 years of experience in content indexing.
2
Branded searches can be impacted
One of the potential impacts in its current operation is found in branded searches. Many users search for a brand with a navigational intention, meaning to go to the website. However, many others type the brand + a product, a need, a question to know certain elements (specific insurance coverage, phone specifications…).
In this case, the responses generated by AI overviews often contain what users need to know, and it’s possible that users may no longer always need to go further directly. This can therefore impact the number of visits coming from SEO, even though the user will have found the information they wanted.
It is however evident that users will always have questions and that these generative AI responses will not transform their needs. However, for brands, it will be important to monitor the responses to ensure that the output responds correctly and with the necessary information. In terms of analytics, impressions could become more important if clicks decrease.
What should you do to prepare?
It is difficult to give a clear answer as the lack of information about the deployment of AI overviews in Europe makes preparing a strategy complex. However, certain best practices remain critical.
✅
Store your Search Console data
Search Console offers 16 months of data by default. However, there is the possibility to store this data in BigQuery to keep it indefinitely. This is a good practice in a situation where the market can change significantly. Having a vision of the past will allow you, in case of change, to make informed decisions and relevant ones and to act where it’s critical.
✅
Continue to invest in quality and personalized content
The possibility of asking long questions means the need for increasingly personalized answers. Writing or generating very precise content allows you to better reach each target audience.
✅
Follow what’s happening in the U.S. or in countries where AI overviews have been present for a long time
Nothing better than an inexpensive VPN to test how search pages appear in certain countries. AT BEST, you will be reassured to see that the changes are not drastic, AT WORST you will have ideas on how to adapt your content for the future.
