Boost your skills with the weekly Growth Memo expert information. Subscribe for free!
Last week, Google officially launched Ai Mode, an AI preview on steroids, in beta.
Almost a year after the launch of AI responses in research results, we have enough data to show that the net impact on the open web is as positive as Trump’s prices on the American economy. Spoiler: not very positive.
AI glimps eliminate more clicks from the classic search results that they do not give back to sources mentioned. The AI mode has the potential to tear the gaping trade deficit of IA’s overviews already causes a large open traffic.
I argue that the appearance of the IA mode is the natural response to deep research and the potential perspective of the future of AI research.
Image credit: Lyna ™Deep research threat Google
Deep Search is an agent of AI by Open IA who creates long reports on a subject of your choice: “An agent who uses reasoning to synthesize large amounts of online information and perform research tasks in several steps for you.”
The most obvious application is market studies, but the agent can also provide rich information on consumption subjects such as buying a car, reservation of a trip or obtaining a credit.
In -depth research is built for people who work intensive in fields such as finance, science, politicians and engineering and need in -depth, precise and reliable research. It can be just as useful for demanding buyers looking for hyper personalized recommendations on purchases that generally require careful research, such as cars, household appliances and furniture.
Image credit: Kevin IndigDeep Search performs dozens of hundreds of research to compile a report. I tried prompts to purchase decisions.
When I asked for “the best hybrid family car with 7 seats in the price range from $ 50,000 to $ 80,000”, Deep Research has traveled 41 search results and reasoned its way through the content.
Some examples of research reasoning:
I found a Kelley Blue Book article on 7 -seater hybrids. It is a good starting point, but all the details are not always included.
I dig in freight capabilities for the Toyota Highlander and the Grand Highlander. The Grand Highlander is more spacious, making it a solid option for larger families.
I look at the electrical and combined ranges of the XC90 recharge, and the deviations from the MPGE figures of different sources.
The report took 10 minutes to assemble but probably saved hours of human research and at least 41 clicks – clicks that could have gone to Google announcements.
The case of deep research
Are deep research officers a threat to Google? I think yes.
Here is why:
- The results are impressive and the time savings are massive. At the beginning, Google boasted of the speed with which it gathered the search results. But it was the speed of the results, not the speed to answer. Today, deep research officers take a few minutes to get an answer, but that’s all you need.
- There is a massive personalization potential, from sources to research criteria.
- Conversation in both directions, just like with a seller in a store. Deep replay agents provide a concise summary that users can develop and explore at their own pace.
- It turns out that each search engine or chatbot AI already has a deep search agent or works on one. It could really be the future of finding complex queries.
Image credit: Kevin IndigBing had a “deep research” feature since December 2023! And that does exactly what the name promises, just faster and not as deep as the chatgpt agent.
Today’s search engines are powerful tools that help us find information on the web, but sometimes they are not of our expectations. When we have complex, nuanced or specific questions, we often find it difficult to find the answers we need. We know ourselves what we are looking for, but the search engine simply does not seem to understand.
This is why we have created Deep Search, a new Microsoft Bing functionality which provides even more relevant and complete responses to the most complex research requests. Deep Search does not replace existing web search from Bing, but an improvement that offers the option for an increasingly rich exploration of the web.1
I didn’t think I would live long enough to see the day when Google Copy Bing … but they are not alone.
Grok has a “deep research” and Gemini and perplexity have “deeply Concerningresearch. “Everyone copies each other, and they don’t even make the effort to choose a different name. What a strong sign of commodification.
Google’s AI mode (source))My theory: Google modeled the AI mode after the deep research of Bing after seeing what the deep research of Chatppt can do.
Using a personalized version of Gemini 2.0, AI mode is particularly useful for questions that require more in -depth exploration, comparisons and reasoning. You can ask nuanced questions that could have already carried out several research – such as exploring a new concept or comparing detailed options – and obtaining a useful response fueled by AI with links to find out more.2
Interestingly, the AI mode opposite the IA glimps: in the announcement of the profits from the quarter of Google, Sundar Pichai said that Google sees an “increase in the use of research among people who use new IA glimpses”.3
Thus, the AI glimps conduct more research, but the AI mode saves time and user requests:
You can ask nuanced questions that could have already carried out several research – such as exploring a new concept or comparing detailed options – and obtaining a useful response fueled by AI with links to find out more.4
I don’t think we will never go back to pre-Ai research. The universal key challenge of AI responses, whatever their form, is confidence. The obvious problem is hallucination.
It is ironic that Chatgpt Deep Research tells me that he has traveled 29 sources, but when I counted, I found 41.
However, the reasoning models improve to solve this problem with raw computer science, that is to say by “reflecting more” to their responses.
The biggest resolution problem for deep research agents is the selection of sources.
The unworthy sources of trust are the microplastics of AI responses. There is a good reason why all models of reasoning openly show their reasoning.
Even if we could pay as much attention to the details of the reasoning as to all the conditions of service, they make us feel that many things happen in the background.
Perception is important for confidence. However, selection of sources is a very resolble problem: users can simply tell the model to ignore the sources they do not want, and the model memorizes this behavior over time.
There are still two less resolution problems:
- Bias: in my IA chatbot research analysisI pointed out that LLMs have a bias towards the world brands, luxury brands, business sources and rapid feeling.
- Access: Information must be on the internet so that the deep research agents find it (this is where Google and Bing have a great competitive advantage).
The biggest question, of course, is whether deep research agents will find a large adoption or will remain in the bubble of knowledge workers.
AI mode could bring it to the masses and lead the stake deeper into the heart of information clicks.
The impact on SEO
IA previews increased in November 2000 and February ’25 (Image Credit: Kevin Indig)The impact of AI previews on SEO traffic is negative.
In my meta-analysis of 19 Studies on IA previewsI found that the AIOS reduces click rates to all levels. Will AI mode worse? Most likely. But there is hope.
First of all, deep research agents are very transparent with their sources and sometimes requests.
The in -depth research of Chatgpt literally calls what she is looking for, so we hope to follow and optimize these requests. Until now, LLMs are still based on research results.
Second, only because researchers get answers before clicking on websites, their purchase intention does not disappear.
What disappears for marketing specialists is the possibility of influencing buyers on their website before buying – as long as IA chatbots offer direct payment.
We must find other ways to influence buyers: brand marketing, Reddit, Youtube, social media, advertising.
Third, there is a chance that AI mode appears mainly for key information words, just like the AI previews. In this case, a lot of weight will come across keywords for high intention, such as “buy x” or “order y”.
Fourth, Bing does not separate the deep research response, but the station in the middle of biological and paid results, lined with links to sources. Hopefully users will always click outside the deep answer.
I wonder how Google plans to monetize the AI mode, which must be more expensive and at high intensity of resources.
To be fair, Google reduced the cost of an AI of 90%, which tells me that they understood the economy of the unit. So it is possible.
But could this be an opportunity to bring back the idea of monetizing research partially with subscriptions on the table?
Based on an information report, OPENAI plans to invoice “up to $ 20,000 per month for specialized AI agents” who could do a doctoral research, $ 10,000 for a software developer agent and $ 2,000 for a knowledge worker agent.5
Another long way to go, but that raises a beautiful theory in IA mode: what happens if Google users could pay for better models that give better answers or have better skills?
1 Presentation of deep research
2 Expand the previews of the AI and introduce the IA mode
3 Q3 winning call: CEO remarks
4 Expand the previews of the AI and introduce the IA mode
5 Openai plots invoicing $ 20,000 per month for doctoral agents
Star image: Paulo Bobita / Search Engine Journal
