When looking for information on the internet, people are typing in individual keywords less and less frequently - and instead asking complete questions. AI assistants like ChatGPT, Gemini, Perplexity, Copilot, or Claude are fundamentally changing research on the web: Instead of searching for websites, users directly ask for an answer and receive it summarized and formulated.
The fact that this shift is no longer a niche phenomenon is supported by data. According to a survey by the IT industry association Bitkom, 67 percent of Germans now use generative AI. Among 18- to 29-year-olds, a third (30 percent) already prefer AI tools over traditional search engines, as determined by the Trendmonitor Deutschland.
Complete Questions Instead of Individual Keywords
The core of the change: It's no longer humans adapting to the machine, but the other way around. "Until now, we've oriented ourselves to the functions of search engines, now the machine is learning how we communicate", says communications expert Michael Frohoff from the agency Kruger Media in an interview with the news agency spot on news. Instead of isolated search terms, you can describe an entire situation to the AI - including contradictions.
Frohoff illustrates this with summer vacation: In the past, if you were looking for a travel destination, the search engine gave you many individual results and you had to compare them yourself. An AI, on the other hand, processes a query like "My last vacation with the family was a disaster. Some want mountains, others want the beach. I'm at a loss. It can cost 3,500 euros, what do you suggest?" and recommends specific destinations including individual preferences.
What Happens in the Background with LLMs
Technically, this is based on a probability model of so-called Large Language Models (LLMs) that the AI tools use. Frohoff illustrates it with an incomplete sentence: "My grandma rides in the ..." - with high probability "motorcycle" or "chicken coop" would follow. A generative AI searches through large data sets, analyzes the texts, and calculates the most probable word sequence.
The difference from traditional search can be summed up simply: Google shows where an answer might be hidden in links; ChatGPT or Gemini formulate it right away, "just as a conversation partner would", according to Frohoff. This means the AI takes over googling from the user, but also draws on other sources.
Which Tool Does What Best?
By now, each system has its own strengths. ChatGPT from OpenAI is considered an all-rounder for explanations, creative work, and everyday life. Gemini is part of the Google ecosystem and can also draw on YouTube as a source. Perplexity cites the source for almost every answer and is thus suitable for research and fact-checking. Microsoft Copilot demonstrates its strengths primarily in Microsoft 365, i.e., in connection with Word, Excel, or Teams. Claude, in turn, is considered by many to be a strong partner for analysis, strategy, and texts.
Frohoff sums it up like this: "ChatGPT for versatile everyday use. Claude for thinking and writing. Perplexity for research. Gemini for Google users. Copilot for everyday office work with Microsoft." For normal everyday use, the differences hardly matter anyway - more important is to decide on one system after some experimentation, because it can then learn along with you and the results increasingly fit the person better.
However, the quality of the answer depends not only on the tool used, but also on the quality of the question. According to Frohoff, a vivid comparison is buying a bicycle: "When you go into a store, you don't just say: 'Which bicycle should I buy?'. Context is important: As much information as possible, as you would explain it in the store or to a friend: 'I commute 15 kilometers through the city to work every day, have a budget of 2,500 euros, want a sporty seating position, on weekends I use it for tours. It should be robust, but also kind of cool. Which models fit me and why?" Generally speaking: The better the AI knows the initial situation, the more individual and therefore better the recommendation turns out.
When AI Hallucinates
Despite everything: As convincing as the answers sound, they are not error-free. Experts speak of hallucinations: A language model only calculates the probability of a statement, but doesn't check whether it's actually true - and formulates even falsehoods with confidence.
How consequential this can be is shown according to "Tagesschau" by recent cases, in which, for example, the consulting firm KPMG cited invented case examples that were based on AI hallucinations. Sources in academic papers are also repeatedly "invented" by AI and end up in other papers through unchecked adoption. And as recently as the end of last year, according to a study by the European Broadcasting Union, almost half (45 percent) of news-related chatbot responses contained errors. In addition to hallucinations, the reason is also an outdated "knowledge base" of the LLMs.
Consider What You Disclose - and Check What Comes Back
You should be especially skeptical when no sources are cited, when surprisingly precise numbers appear, when cited studies or experts cannot be verified, or when it concerns medical, legal, or financial questions. Because in the end, the user is still personally responsible for their decisions - and under almost every chat are sentences like: "AI can make mistakes. Verify important information."
In addition, there is questionable data privacy with the consistently American providers. Data privacy expert Ayten Öksüz from the Consumer Advice Center NRW advises users in a "Focus" article not to enter any personal data such as names, date of birth, or place of residence of themselves or third parties. Health and account data as well as passwords are particularly sensitive, since user inputs are reused for training in many applications. Just asking about a strange skin rash or a low-cost loan can be stored and processed as health or financial information.
How We Will Soon Search
Despite all the problems, much suggests that development is accelerating. Frohoff is convinced: "I believe that the word 'search' or 'google' will disappear." AI tools will become personal assistants that not only research tasks but complete them entirely - from travel planning to booking, or from the initial purchase consideration to ordering and delivery of the product. At the same time, according to his prediction, the thirst for human expertise and genuine exchange will grow in parallel.
