~6 min
37
Author
Share
People are moving from search to AI assistants because of how the mind works, not because the technology is fashionable. Five mechanisms are at play: we save effort, hand complex thinking over, respond to a machine as a conversation partner, open up more easily to something that does not judge, and hold on to a sense of control. In an experiment with 239 participants, people were more candid when they believed they were talking only to a computer. For a business this means the first and most honest questions are no longer put to you.
In short. A customer who asks ChatGPT instead of Google is not being lazy or chasing a novelty. They are doing what the human mind has always done: choosing the path with less effort, less risk of looking incompetent and more control. Below are the five mechanisms behind this, the research they rest on, and the limits of these explanations. At the end you will find what it means for your website.
Because every link is a small decision, and decisions cost energy. Researchers at Nielsen Norman Group describe choosing a page through two estimates: how likely it is that the answer is there, and how long it will take to get it. A person runs this calculation before every click, mostly without noticing.
A search results page demands that calculation ten times in a row. Then you still have to read what you found, compare it and decide whom to believe.
An assistant removes this work. There is one answer and nothing to hunt for. According to a Bain and Dynata survey (1,117 people, December 2024), about 80% of consumers already settle for an answer without visiting a website in at least 40% of their searches.
Sometimes, but less often than is claimed. The best-known evidence is the 2000 experiment by Sheena Iyengar and Mark Lepper. At a table with 24 kinds of jam, 60% of passers-by stopped but 3% bought. At a table with six kinds, 40% stopped and about 30% bought.
However, a later meta-analysis of 50 experiments (Scheibehenne and colleagues, 2010, about 5,000 participants) found an average effect close to zero. Choice overload does not always occur.
It occurs under certain conditions: when a person knows the subject poorly, the options are hard to compare and a mistake is expensive. That is exactly what choosing a lawyer, a clinic or a renovation contractor looks like. Here a single ready answer really does make the decision easier.
Because the brain always does this when there is something to hand it to. Psychologists call it cognitive offloading: we write shopping lists, use a calculator, ask a navigation app for directions. In 2016 Evan Risko and Sam Gilbert reviewed the research on this phenomenon for the journal Trends in Cognitive Sciences.
Only one thing is new. We used to offload memory and arithmetic. Now we offload comparison and the conclusion. Daniel Kahneman described two modes of thinking: fast and intuitive, slow and analytical. Comparing three contractors on price, timing and guarantees means switching on the slow mode. The assistant does it for the person, who only has to agree.
For a business the consequence is direct: the comparison happens without the customer and without your website.
Because social responses switch on automatically. In 1994 Clifford Nass and colleagues showed that people unconsciously apply the same rules to computers as they do to people. In a 1997 experiment, participants were more willing to “help” the computer that had helped them earlier. They knew it was a machine and still returned the favor.
With assistants that write in fluent language the effect is stronger. In 2023 Nielsen Norman Group described four degrees of this attitude: courtesy (“please”, “thank you”), praise for a good answer, asking it to play a role, and treating the assistant as a companion.
A search box triggers no such response. Nobody talks to it. People type words into it.
Because the fear of being judged disappears. In a study by Gale Lucas and colleagues (2014, 239 participants), people went through a health interview with a virtual interviewer. Half were told a human was operating it, half that it was only a program. Those who believed they were talking to a program were less afraid to open up, managed their impression less and showed sadness more openly.
The same works in everyday matters. It feels awkward to tell a sales manager “I don’t understand how your service differs from the cheaper one” or “is it normal that this costs so much”. With an assistant it is fine.
So the first, simplest and most honest questions about your service are no longer put to you. The customer arrives with a picture already formed.
Because people return more readily to places where they run the situation themselves. In Edward Deci and Richard Ryan’s self-determination theory, autonomy is one of three basic psychological needs, alongside competence and relatedness.
A search engine decides for itself what to show and in what order. With an assistant the customer sets the terms: shorter, simpler, only in London, only under €1,000, compare it in a table. The answer adapts to them.
That is also why queries got longer. The customer writes the budget, the city, the deadline, the team size. A page without those details does not fit such a query.
Your website has to meet the same needs the assistant meets. Then the assistant has something to take from it.
| Mechanism | What the customer feels | What to do on the site |
| Saving effort | “I don’t want to read ten pages” | say at the top of the page what you do, for whom and for how much |
| Offloading thinking | “compare it for me” | give your own honest comparison with the alternatives, as a table |
| Machine as a conversation partner | “I ask the way I speak” | write headings the way customers phrase their questions |
| Safety without judgment | “I’ll ask what I’m embarrassed to ask” | answer basic questions in plain words |
| Control | “I need it on my terms” | name specific terms: prices, cities, timelines, who it suits |
They explain behavior, but they do not measure it in your niche.
A lab is not a market. The Lucas study was about a conversation on health, not about choosing a contractor. Nass’s experiments were run on 1990s computers. Carrying the findings over to modern assistants is logical, but it is a carry-over, not direct proof.
The choice effect is disputed. The 2010 meta-analysis showed that on average it barely exists. It appears only in complex and expensive decisions.
Nobody has left Google. 95% of ChatGPT users also use regular search (Similarweb, data through May 2026). The assistant was added to search. It did not replace it.
Bain’s figures are a survey, not your analytics. Bain sells consulting, so treat them as a reference point.
You will need one evening and any assistant.
If nobody named you after step one, start with our step-by-step checklist for getting into ChatGPT answers. Where this work sits in the overall plan is shown in the Haivoronskyi Framework.
Request a free AI visibility mini-audit from F-47. We will run your customers’ main queries through ChatGPT, Perplexity, Gemini, Claude and Google AI and show you who gets recommended instead of you.
The mechanisms involved are older than any technology. Saving effort, fear of judgment and the need for control will not disappear with the next model version. The tool will change, but the reasons people choose it will remain.
No. They hand over the comparison and a draft conclusion, the way they once handed arithmetic to a calculator. The final purchase decision is still theirs. But the shortlist they choose from is no longer put together by them.
Because in front of a person they manage the impression they make. In the 2014 experiment, participants were more candid when they believed they were talking only to a program. Basic questions like “do I even need this” are easier to put to an assistant.
No. 95% of ChatGPT users keep using search, and assistants mostly take their answers from pages that rank well in Google. Work for assistants is built on top of SEO.
The AI visibility mini-audit is free. A full audit costs €200-500 one time. After that comes retainer work: €500-700 or €900-1500 per month. The scope is described in our article on ordering GEO/AEO optimization.
Share
We will check the technical readiness of your code and page structure to attract super-converting GEO traffic.
~7 min
18
~5 хв
173
~8 min
187
