ⓘ This page has been translated using artificial intelligence.
Generative text AIs and language models
Whether as a chat assistant on your smartphone, integrated into Office applications or as a research tool in your browser, generative artificial intelligence has long been part of everyday life. And the technology behind it is constantly evolving: from simple text generators to AI agents that perform complex tasks independently. From standard chatbots to reasoning models that think in multiple stages.
On this page, you will find interesting facts about text AIs: What models are available? How do they work? And above all: How can you use them profitably in everyday life, at work and at school?
Go directly to topic
Share this page
How do text AIs work?
Topic
How does generative text AI work?
We humans do not constantly rethink everything; rather, we learn and build new knowledge on what we have learned before. Our thoughts and memories remain in our minds and thus form the basis for new connections.
Generative AI also works with such links and must therefore be able to access a neural network in which information remains available. But how is such an artificial memory created, and how can AI generate answers from it?
1995: LSTM method
The LSTM (Long Short-Term Memory) method is a piece of software that has been around since 1995 and is used to recognise certain patterns in data. This enables the software to remember previous information and reuse it when necessary.
For this technology to work, LSTM uses a special neural network called a recurrent neural network (RNN), which works with repetitive processes. This means that the software is able to process information step by step.
However, LSTM also has its limitations. Since it processes information sequentially, it cannot handle large amounts of data well and tends to overlook more distant information.
2017: Transformer
In 2017, Google presented the scientific paper ‘Attention is All You Need’(opens in new tab). In it, the authors suggest that new attention mechanisms can significantly increase the quality and efficiency of previous models for neural networks:
The research team proposes a ‘simple network architecture, the Transformer, based exclusively on attention mechanisms [...]’(opens in new tab). This Transformer uses novel mechanisms to better understand context and process it more efficiently. Experiments were conducted with machine language translation tasks.
The Transformer has significantly advanced the development of generative text AI. Because it can better handle long dependencies in texts and be trained more efficiently, it is ideal for tasks such as text generation, chatbots and other applications that need to generate or understand natural language. No wonder, then, that all well-known text AIs today are based on the Transformer architecture.
Why do some text AIs have better memory than others?
Several factors interact and influence the memory of a generative AI. These include, for example:
- Context window (short-term memory): The context window determines how much text an AI can process at once. This amount of text is measured in tokens. The larger the number, the more the AI keeps in mind when responding.
- Memory feature (long-term memory): AIs such as ChatGPT can store important information in an external database so that they can access this information in later chats.
- Projects (multiple chats, one memory): Some language models, such as Claude, offer the Projects feature. When chats are assigned to a specific project, the content of the chat contributes to the project memory and can be retrieved from other chats in the same project.–Web search function: If an AI has direct access to web search, it significantly expands the information available.–Reasoning models: These are designed to solve complex thinking tasks and therefore usually have more memory capacity. Please note: They often consume more tokens, which reduces the available credit of token-limited subscriptions more quickly.
- Chatbot vs. model: It is important to distinguish between the two terms: AI chatbots are products that contain AI models. The same AI model (e.g. GPT-5) can function differently in ChatGPT and Perplexity if additional systems such as security filters or search functions differentiate the output of a response from chatbot to chatbot.
Topic
LLM vs. AI browser: real time or not?
Perhaps you have asked yourself: How does AI actually know what it knows? Of course, it is trained. But why does it (only) sometimes know the latest news? The reason lies in the technical distinction between so-called LLMs and AI browsers:
LLM stands for Large Language Model
During training, these models have read enormous amounts of text and learned how language works. They are, in a sense, highly educated language assistants who generate their responses based on their current knowledge (usually at a specific point in time in the past).
Key point: LLMs can explain how photosynthesis works, but they cannot tell you who won yesterday's football match.
AI browsers conduct live research on the internet
An AI browser is a hybrid model between an AI and a browser. Perplexity is a good example of this, as are ChatGPT models with an integrated browsing function. AI browsers provide you with linguistically polished answers in chat format, but these are generated from the latest information available on the internet.
Key point: While LLMs live in the past, AI browsers are also aware of today's world and can search Google in real time.
It's not always easy to distinguish between LLM and AI browsers. Here's a trick that helps: ask the AI about a current event – for example, something from yesterday's news. If it knows a specific and correct answer, you're working with an AI browser. If not, you're dealing with a classic LLM.
Topic
What are the most well-known
AI text generators?
The number of providers of generative text AI has been growing exponentially in recent months. Among the best known are:
myAI (Swisscom)
Multimodal AI assistant from Swisscom, developed in Switzerland and for Switzerland. myAI is based on the Claude language model and supports you in writing, translating, researching and analysing. myAI understands all Swiss dialects, answers questions via text or voice, analyses uploaded documents and generates images. Seamless integration with Swiss services such as SBB timetables, MeteoSwiss forecasts and Blue TV.
Age rating (myAI)
18 years and older (according to Claude age rating)
Access (myAI)
Web, app (requires Swisscom login, no Swisscom subscription required)
Strengths (myAI)
- Prioritises Swiss search results, understands local peculiarities and Swiss dialects.
- Voice control: chat via text or natural spoken language.
- Image generation with transparent AI watermark for responsible use.
Weaknesses (myAI)
- Less well known than international competitors such as ChatGPT or Gemini.
- Rather cautious with speculative queries (high factual accuracy).
- Swisscom login required.
Safety (myAI)
- Data processing exclusively on Swiss and EU servers.
- No profiling or use of data for AI training.
- Users can delete their conversations at any time.
Educational value (myAI)
- Multimodal model for everyday questions with a Swiss focus.
- Document analysis for school projects.
- Helps with research using Swiss sources.
Classification (myAI)
- Multimodal model for everyday questions with a Swiss focus.
- Data protection and data sovereignty are top priorities.
- Reliable answers thanks to Claude technology.
ChatGPT (OpenAI)
AI assistant from OpenAI based on the latest GPT-5.2 models (as of February 2026). ChatGPT understands natural language, processes text and images, generates images and can also create videos with Sora 2. As a multimodal model, ChatGPT has advanced functionalities: Canvas for editing documents, Agent Mode for autonomous tasks, Deep Research for comprehensive research, and Connectors for integration with Gmail, Google Drive, Dropbox, and other tools.
Age rating (ChatGPT)
Recommended for ages 13 and up
Access (ChatGPT)
Web, app, API for developers
Strengths (ChatGPT)
- Multimodal and versatile: text, image, code, video.
- Canvas feature for interactive document and code editing in a separate window.
- Advanced features such as Agent Mode, Deep Research, Memory function and Connectors.
Weaknesses (ChatGPT)
- Sometimes explains things in a cumbersome and overly detailed manner.
- Occasionally hallucinates.
- Sora 2 only for paying users.
Safety (ChatGPT)
- Conversations are stored by default.
- Data is used for training by default (can be disabled via privacy settings).
- Sora-generated videos have visible watermarks and C2PA metadata.
Educational value (ChatGPT)
- Can assist with homework and explain complex topics in an easy-to-understand way.
- Deep Research supports scientific research and source work.
- Can help teachers create exercises and teaching materials.
Classification (ChatGPT)
- The best-known and most widely used AI model worldwide.
- Good all-rounder for text, images, code and videos.
- Particularly strong at creative tasks, explanations and autonomous workflows.
Try ChatGPT: https://chatgpt.com/(opens in new tab)
Gemini (Google)
Gemini is Google's multimodal AI model based on Gemini 3 Pro (as of February 2026). Gemini is seamlessly integrated into the Google ecosystem and can be found everywhere: Google Search, Chrome, Gmail, Google Drive, Maps, YouTube and other Google services. With the Google AI Ultra subscription, additional features such as Deep Think mode and Veo 3 video generation are available.
Age rating (Gemini)
Recommended for ages 13 and up
Access (Gemini)
Web, app, integration into Google services (with Google account)
Strengths (Gemini)
- Multimodal model with native processing of text, images, audio, video and code.
- Seamlessly integrated into Google services.
- Deep Think mode for complex scientific and mathematical queries.
Weaknesses (Gemini)
- Creative tasks may take a back seat to a focus on facts.
- Advanced features only available with a paid subscription.
- Tends to give relative and cautious answers.
Safety (Gemini)
- Parental control possible via Google Family Link.
- The privacy policies of Google services apply.
- Chat history must be disabled to prevent data usage for training.
Educational value (Gemini)
- Suitable for school research and scientific topics.
- Can explicitly search for trustworthy sources (double-checking by hand is still recommended).
- Structures complex topics in an understandable and systematic way.
Classification (Gemini)
- Focus on facts, science and reasoning.
- Ideal for Google users thanks to seamless integration.
- Particularly strong in research and multimodal understanding.
Try Gemini: https://gemini.google.com/app(opens in new tab)
Meta AI – formerly Meta Llama (Meta)
Meta's AI assistant, integrated into WhatsApp, Instagram, Facebook and Messenger, is based on Llama 4, Meta's open-source language model with native multimodality (text and image). Meta AI has been available in Switzerland since March 2025 as a European version with limited functionality (compared to the US version) for data protection reasons (GDPR): text chat only, no image generation and no memory function. It can be recognised directly in the Meta apps by the blue circle icon or can be contacted by writing to @MetaAI.
Age rating (Meta AI)
Depending on the respective platform, usually from 13 years of age
Access (Meta AI)
In WhatsApp, Instagram, Facebook, Messenger or as a standalone app (meta.ai)
Strengths (Meta AI)
- Low threshold for access.
- Strong in social contexts and everyday conversations.
- Free for all Meta users.
Weaknesses (Meta AI)
- European version severely restricted.
- Multimodal features only available in the USA.
- Little in-depth explanation, focused on everyday questions.
Safety (Meta AI)
- Meta's privacy policy applies.
- Data is used for training purposes (can be objected to).
- European version trained with European user data.
Educational value (Meta AI)
- Answers questions about everyday knowledge.
- Not very structured for educational purposes.
Classification (Meta AI)
- Practical as an aid within Meta apps.
- Geared towards social and everyday interactions.
Try Meta AI: https://www.meta.ai/(opens in new tab)
Claude (Anthropic)
Claude is Anthropic's AI assistant, known for its thoughtful responses and nuanced conversation skills. The current main model is Claude Opus 4.6 (as of February 2026). Claude Sonnet 4.5 is considered the world's best coding model. Claude specialises in code, text comprehension, ethical discussions and complex reasoning tasks.
Age rating (Claude)
Recommended for ages 18 and up
Access (Claude)
Web, API, Claude Code, Office integration (Excel, PowerPoint), partnerships (e.g. Notion)
Strengths (Claude)
- Autonomous code development.
- Extended thinking for complex contexts.
- Agent teams to work on large projects in parallel.
Weaknesses (Claude)
- Less well known than ChatGPT and Gemini.
- No image generation available.
- Reluctant to answer speculative questions.
Safety (Claude)
- Transparent data protection.
- Data is not used for training by default.
- Takes security and ethical AI development seriously.
Educational value (Claude)
- Strong at text analysis, summaries and essay feedback.
- Extended thinking suitable for academic problem solving.
- Promotes critical thinking through balanced, differentiated presentation.
Classification (Claude)
- Specialises in coding and complex reasoning tasks.
- Integration into tools such as Office or Notion makes Claude a productive work assistant.
- More thoughtful and reflective than other models.
Try Claude: https://claude.ai/(opens in new tab)
Mistral AI
Mistral AI is a French open-source language model and a European alternative to US providers such as OpenAI. The current model family: Mistral 3 with Mistral Large 3 and Ministral 3 (as of February 2026). The language model focuses on multilingualism (especially European languages), GDPR compliance and data sovereignty.
Age rating (Mistral AI)
No information
Access (Mistral AI)
Web (Le Chat), API, Amazon Bedrock, Azure, Hugging Face, IBM WatsonX
Strengths (Mistral AI)
- Open source: models can be downloaded free of charge and integrated into your own systems.
- Multimodal and multilingual.
- Ministral 3 runs on edge devices, also ideal for offline use.
Weaknesses (Mistral AI)
- Still little known compared to US competitors.
- Proprietary models such as ChatGPT are even more powerful.
- Relatively young language model.
Safety (Mistral AI)
- GDPR-compliant data protection guidelines.
- Focus on data sovereignty, making it suitable for businesses.
- Open-source transparency enables independent security audits.
Educational value (Mistral AI)
- Suitable for European educational content.
- Model can be adapted for teaching/learning purposes thanks to open-source technology.
- Strong in languages and translations.
Classification (Mistral AI)
- Most important European alternative to US models.
- Ideal for data protection-sensitive applications.
Try Le Chat: https://chat.mistral.ai/chat(opens in new tab)
Perplexity (Perplexity AI)
Perplexity is an AI-powered search engine with real-time web search. The language model specialises in fact-based research rather than conversation and provides accurate, source-based answers. Pro users can choose between several AI models. For even more targeted research, there are Deep Research and Focus Modes functions.
Age rating (Perplexity)
From 13 years of age
Access (Perplexity)
Web, app, browser extension, API
Strengths (Perplexity)
- Real-time web search with source references and direct links.
- Deep Research creates comprehensive reports with numerous sources in just a few minutes, exportable as PDF.
- Focus Modes tailor search results, e.g. only scientific sources.
Weaknesses (Perplexity)
- Perplexity is not suited to creative tasks such as storytelling or brainstorming.
- Can sometimes be overwhelming with lots of details and sources.
- Pure fact transfer, without deeper explanations or didactic preparation.
Safety (Perplexity)
- Data on servers in the USA, shares information with third parties for web searches.
- Incognito mode (with account) available, searches are not saved.
- Pro users can choose the underlying model – data protection varies depending on the model.
Educational value (Perplexity)
- Educational value (Perplexity)Conveys a scientific approach through consistent referencing and citations.
- Helps with research for school projects or similar.
- Academic Mode is popular with upper secondary school pupils and students.
Classification (Perplexity)
- Ideal as a research tool, less so as a chatbot.
- Source-based research with direct web access.
- Alternative to Google search for complex research questions.
Try Perplexity: https://www.perplexity.ai/(opens in new tab)
DeepSeek
DeepSeek is a Chinese open-source AI model consisting of two main components: DeepSeek V3 for everyday use and DeepSeek R1 as a reasoning model that shows its thought processes. DeepSeek attracted attention because it was developed at very low cost. Based on open source and MIT licensing, DeepSeek can be hosted locally. However, there are data protection concerns due to data storage in China and the possibility that the state can access this data. For this reason, DeepSeek has already been banned in several countries.
Age rating (DeepSeek)
No information
Access (DeepSeek)
Web, app, API, local installation possible
Strengths (DeepSeek)
- Strong in mathematics and coding.
- Open source and free (can be hosted locally and without cloud dependency).
- Offers chain of thoughts, ideal for learning problem-solving strategies.
Weaknesses (DeepSeek)
- Censors certain topics in accordance with Chinese law and provides one-sided answers.
- Factual accuracy weaker than Western models.
- Recent frequent server overloads.
Safety (DeepSeek)
- Critical, as data is stored on servers in China; Chinese law requires companies to share data with authorities.
- Several European countries have banned DeepSeek due to GDPR concerns.
- Local installation bypasses the Chinese servers, but security vulnerabilities in the model itself remain.
Educational value (DeepSeek)
- Chain-of-thought shows the solution paths transparently.
- Censorship and political bias limit educational value.
Classification (DeepSeek)
- DeepSeek is problematic in terms of data protection law.
- Suitable for mathematical and code queries, but not for sensitive data.
- Open source alternative for tech-savvy users.
Try DeepSeek: https://www.deepseek.com/(opens in new tab)
Apertus
Apertus is the first complete Swiss AI language model. It was developed by EPFL, ETH Zurich and CSCS. What is particularly noteworthy about Apertus is the high degree of transparency regarding training data and technical development of the model. However, Apertus is not a finished chatbot, but rather a basic model for developers and researchers.
Age rating (Apertus)
No information available
Access (Apertus)
PulicAI, Swisscom Business Platform, Hugging Face, local installation
Strengths (Apertus)
- Complete transparency with regard to training data, architecture and code.
- Developed in accordance with the Swiss Data Protection Act, GDPR and EU AI Act.
- Multilingual with a focus on Switzerland.
Weaknesses (Apertus)
- Even less mature than commercial models.
- Makes mistakes in rare languages.
- Requires technical know-how for installation and use.
Safety (Apertus)
- High compliance standards, takes into account the Swiss Data Protection Act, GDPR and EU AI Act.
- Fully locally hostable, full data control.
- Personal data is removed before training.
Educational value (Apertus)
- Ideal for research and teaching.
- Valuable resource for language research.
- Model can be further developed for own projects.
Classification (Apertus)
- Not Swiss ChatGPT, but research infrastructure and basic model for further developments.
- Technological sovereignty.
- Open-source role model: Shows how ethical, transparent AI can be developed.
Try Apertus (PulicAI): https://publicai.io/(opens in new tab)
Microsoft 365 Copilot (Microsoft)
Copilot is Microsoft's AI assistant and is therefore directly integrated into all Office applications. The AI is based on GPT or Claude models (as of February 2026) and is directly integrated into workflows. Copilot Chat is free for all Microsoft 365 users, but premium features such as Agent Mode cost extra.
Age rating (Microsoft 365 Copilot)
No information
Access (Microsoft 365 Copilot)
Integrated into Microsoft 365 (Word, Excel, PowerPoint, Outlook, Teams, OneNote), web, app
Strengths (Microsoft 365 Copilot)
- Seamlessly integrated into Office tools, so no app switching is necessary.
- Agent Mode autonomously performs complex multi-step tasks.
- Work IQ learns your working style and adapts to your habits and preferences.
Weaknesses (Microsoft 365 Copilot)
- Copilot accesses all data for which you have permissions (including accidentally shared files).
- Premium features are subject to a fee.
- Not available as a separate chatbot.
Safety (Microsoft 365 Copilot)
- Data is not used for training.
- Respects your permissions, but only if you have set them correctly.
- Companies should carefully check SharePoint/OneDrive permissions to prevent Copilot from accessing sensitive data.
Educational value (Microsoft 365 Copilot)
- Helps with text creation in documents, data analysis in Excel, or designs in PowerPoint.
- Automates repetitive tasks such as formatting, email drafts, etc.
- Copilot does the work, but hardly shows how it does it: limited learning effect.
Classification (Microsoft 365 Copilot)
- Less of a chatbot and more of a work assistant, which is why it is integrated into existing Microsoft tools.
- Ideal for Microsoft 365 users.
- Check permissions before activation to avoid the risk of oversharing.
Try Copilot: Copilot Chat included in Microsoft 365
Topic
Basic rules: How do I prompt?
Although language models are designed and very good at understanding natural language and responding accordingly, clear structure and language in prompts help to obtain accurate responses from AI.
When prompting, always adopt a ‘concrete and specific’ approach. In other words, formulate exactly what you want to know. If you ask vague questions, the generative AI will give vague answers. And you don't want that; you want clear, useful statements. Right? More prompting hacks:
Assign roles to the AI
What writing style should the AI use to respond? It's best to assign it a role so that it strikes the right tone, for example: ‘Respond as a teacher’ or ‘Explain in a way that a 10-year-old child can understand’.
Tip for writing tasks: Have the AI perform a style analysis of an existing text in advance so that it can train itself to use the desired style. Only then should you give it the actual task you want it to answer – in the style it has now been trained to use.
Specify the target format
In what form should the answer be presented? As continuous text with subheadings, or as a table with columns for “Places”, “Sights” and “Restaurants”? Or perhaps even as a graphic or code? Tell the AI.
Let the AI think step by step
A generative AI usually understands and responds better if you give it only one clear task per prompt. If you want to know more after that, keep asking questions. During the conversation, the AI gradually grasps the entire context and you can have all your questions answered one by one.
For more complex tasks, it can also help to explicitly ask the AI to think step by step. Add phrases such as ‘Think step by step’ or ‘Explain your reasoning’ to your prompt. The AI will then show you its thought processes and often provide more thoughtful, accurate answers.
Many models now have a reasoning mode available. If you activate this, the AI will automatically think longer and show you its thought processes transparently.
Examples can help
If you want to receive information in a specific structure, subtopics at a certain altitude, or in a predefined style, provide the AI with an example in your prompt so that it understands you. This will save you time-consuming explanations and make it easier to get better results.
Be patient and willing to experiment
Chatting with a language model always means receiving answers that are not very helpful. The AI may also misunderstand you sometimes. That can happen. Rephrase your question; perhaps you can divide the task into two steps? Talk to the AI and tell it when and what it has misunderstood and what it should do instead.
Maintain a critical attitude
A language model works by always finding the most probable next words or answers. However, this does not necessarily mean that this information is correct. Most language models therefore issue a justified warning (usually near the input field): ‘I may make mistakes. Please check my answers.’ Please adhere to this.
Use images and documents
Many models can not only process text, but also analyse images, PDFs or other files. Take advantage of this and upload the file if you want the AI to recognise, explain or describe something in it. Upload a document that you want the AI to summarise or translate into another language. This will save you a lot of time typing or explaining.
Switch to reasoning mode for highly complex problems
When you present tricky problems (maths problems, logic puzzles or ethical dilemmas) to the AI, some now offer special reasoning modes such as DeepThink, etc. The AI then thinks longer, questions itself critically and shows its thought processes transparently. It asks itself intermediate questions and proceeds very systematically before answering. This often makes the result more accurate than if the AI answered quickly.
What if the AI forgets the beginning?
Generative AI has limited memory. If a chat becomes very long, it may not remember the beginning. In such cases, you can start a new chat and summarise the most important points from the old chat so that the AI has the full context again.
Language models and data protection:
Do not share personal data with AI
Please be aware that when using generative AI, the safest approach in terms of privacy is not to share any personal data with AI in the first place. Personal data can include information such as names, addresses or telephone numbers, but also photos of yourself or others. Please note the following:
- Avoid personal information: The safest way to prevent personal data from being used in AI training and thus being disseminated is to either create requests without any personal data or to censor it before sending.
- Data storage on the model's servers: The data provided to the AI in the form of a prompt is processed and stored on the model's servers. This means that, by default, this data is also used for further training of the AI models.
- Activate your right to object to training: To prevent your data from being used by the company for AI training, you can usually object (opt out) in the respective data protection settings, e.g. here: ChatGPT(opens in new tab), Google Gemini(opens in new tab), Meta Llama(opens in new tab) and Claude(opens in new tab).
Topic
Taking prompt skills to the next level
Are you already experienced with prompting and want to get even more out of your AI's responses? Then try out the following prompting techniques.
Have it explained to you from different perspectives
Not only can AI integrate metaphors into its responses, it can also vary the tone and pitch of an explanation. Use this to have a topic explained to you from multiple perspectives.
Example: Answer the question ‘Should smartphones be allowed in class?’ from the perspective of a teacher, a 10-year-old pupil and his mother.
Not only can AI integrate metaphors into its responses, it can also vary the tone and pitch of an explanation. Use this to have a topic explained to you from multiple perspectives.
Challenge the AI to discuss with itself
AI can also simulate a discussion by taking both sides equally and finding both arguments and counterarguments for a thesis and bringing them into the conversation.
Example: Simulate a discussion between two media experts on the thesis: ‘Teaching critical media literacy in schools is more important than teaching technical skills.’ Expert A takes the position that critical thinking forms the basis on which technical skills are built. Expert B argues that without sound technical knowledge, critical thinking and media literacy are not possible. Structure the discussion as a dialogue with 3-4 contributions per person. Take into account current challenges such as AI-generated content, disinformation and algorithmic filtering. Both experts should contribute arguments and examples from educational practice.
Feedback loops with your virtual teammate AI
Although AI has no feelings and sometimes no opinion of its own, it can still be a great partner for brainstorming or feedback loops. Ask the AI to engage in an iterative process and exchange ideas on your topics digitally, just as you would with a team member.
Example: Prepare the AI for the exchange by informing it: ‘I am going to present an idea to you. You will then give me feedback on how I can develop and optimise it. I will then improve my idea. We will repeat this process three times.’
Have the AI play a (historical) role-playing game
AI can slip into roles and bring stories or fictional scenes to life for you. This type of prompting is ideal for bringing historical figures or situations to life.
Example: Simulate a conversation between Marie Curie and Albert Einstein in 1925 about the future of physics. Marie Curie should defend her research on radioactivity, while Einstein asserts the importance of his theory of relativity. Take into account the state of knowledge at the time and the social circumstances for scientists, especially for a woman in science.
How good do you think your own answer is, dear AI?
You can also ask the AI to rate its own answer on a scale of 1–10. It should also explain what it could have done better. This helps you to identify and classify the strengths and weaknesses of the answer.
What information do you need from me, dear AI?
For example, if you want to plan a trip and want the generative AI to suggest sightseeing tips or routes, ask the AI in advance what it needs to know about you in order to plan optimally. Answer all questions individually, in a structured manner, and provide all necessary details (no personal data!). This will help you optimise the suggested trip so that it matches your travel dates, preferences and budget as closely as possible.
Use prompt frameworks such as TIDD-EC
Prompt schemata (known as frameworks) such as TIDD-EC support structured prompting. TIDD-EC is composed of the initial letters of its components:
- Task Type: Give the language model a clear idea of what type of task it should perform.Instructions: Describe specific steps or guidelines for the language model to follow.
- Do: What actions will help the AI perform the task?
- Don't: What elements should the AI avoid when responding to the prompt?
- Examples: Provide concrete examples of the desired type of results or responses.
- Content: Provide specific data (note: only general data, no personal data!) that the AI can work with, use in its response, or refer to.
Example: ‘Explain to me in simple terms the health benefits of green tea. Refer to scientific findings and current research results. The answer should be easy to understand and not contain any complicated technical terms. An example would be: “Green tea can help strengthen the immune system.”’
How to avoid common prompt errors
If the AI isn't giving you the answers you want, these could be the reasons why:
- Vague prompts: An AI needs specific questions in order to give specific answers. Imprecise questions result in imprecise answers.
- Too little context: If you give an AI too little background information, it can easily misunderstand you. Examples and sources can help.
- Giving up too soon: The first answer is usually not the best. Ask follow-up questions and give the AI feedback on what you think is good or bad.
- Too contradictory: Be consistent and uniform in your instructions. If you first write ‘please be thorough and detailed’ and then ask for ‘short and concise’ answers, this will confuse the AI.
- Expecting too much: Don't overwhelm the AI with too many sub-questions in one and the same prompt. It's better to ask questions gradually and be aware of how it works and what it can do.
- Too familiar: AI can make mistakes. Therefore, double-check important facts and technical or historical topics.
Topic
Which AI mode suits my prompt?
Formulating the prompt correctly is only half the battle. Many AI tools today offer additional settings that greatly influence the result. Depending on the task, you can choose whether the AI accesses current web content or only its trained knowledge. The latter is faster, but may be out of date for current topics. With tools such as Perplexity, the ‘Deep Research’ function allows you to limit the search to academic sources.
The choice of model itself also plays a role: lighter models respond quickly and are well suited for simple everyday questions. For complex tasks, it is worth choosing a more powerful model or activating the so-called reasoning mode (standard, long, pro). In this mode, the AI ‘thinks’ longer before responding, thus arriving at more thoughtful results.
In short, if you invest a little time in getting to know the model's settings, you'll get much more out of the results in the end.
Topic
Application examples for families
The field of application for generative text AI is immense. In addition to providing daily support in a wide variety of professional sectors, language models can of course also be helpful in private and family life.
In a family context, it could provide support in the following areas:
Learning and homework
Is your child stuck on a maths problem? Generative AI can provide alternative explanations if the explanations in the teaching material haven't quite clicked yet. For example, you can ask the AI to provide a clear example to illustrate the topic.
Example: Explain photosynthesis to me in a way that a 10-year-old child would understand, using a comparison with a factory.
Of course, AI does not replace your own thinking – rather, it supports it. Like a patient tutor who is available around the clock.
Creative writing
Thanks to its capabilities as a language model, generative text AI can be a wonderful aid to creative writing. This does not mean that it does everything for you, but rather that it serves as a source of ideas or brainstorms possible topics with you, for example. It can also help you if you find it difficult to get started on a text, you can try out different styles with it, or it can give you feedback when you want to put the finishing touches to your text.
What kinds of texts can AI write? Well, anything you want. Perhaps a bedtime story for your 6-year-old daughter?
Text-based games
Fancy a unique game night? Generative text AI can also create interactive adventure stories with you and your family. Together, you can explore fantasy worlds, solve puzzles or investigate as detectives. The AI responds to your decisions (with the appropriate task at the beginning) and continuously develops the story.
Example: Embark on a family adventure in space, where we explore a newly discovered planet as a team of explorers. It should be child-friendly and offer decision-making opportunities that incorporate and take into account physical and scientific topics.
Everyday organisation
‘Mum, Dad, what are we having for dinner tonight?’ We've all heard it before – the same question every night. Generative AI can provide inspiration here too, for example with a meal plan for the whole week. You can tell it in advance about any food intolerances or specific requests and ask it to create a meal plan and shopping list before you go shopping.
Example: Create a weekly meal plan for our family of four with vegetarian dishes that are easy to prepare and loved by children.
Beyond menu planning, AI can also help with other aspects of everyday organisation, such as creating a template for a chore schedule. Or providing you with ideas for a rainy Sunday.
Multilingualism
Generative AI can also be an important resource for language learning. It usually translates individual words or entire texts flawlessly (depending on the model). Compared to a conventional translation tool, AI can be helpful when it comes to classifying or comparing cultural nuances of a language.
Example: Translate the following three German idioms into Spanish and explain whether there are similar idioms in Spanish culture.
Parent guide: What should parents bear in mind when it comes to AI?
How can I support my child in the rapidly changing world of AI? What should I pay attention to and what does my child need from me?
Topic
What are the opportunities and
limitations in education?
Yes, artificial intelligence is also changing the educational landscape. But how can we make meaningful use of its potential? Where do we need to recognise its limitations? An overview for teachers and parents:
Opportunities
AI as a learning companion – not a replacement
Generative text AI can act as a patient learning companion, available 24/7 to answer questions at different levels. It can also offer alternative learning approaches or adapt to the individual pace of learners. In short, AI can provide support where personalised assistance is needed.
However, it does not replace the teacher or the independent learning process. Rather, it is a tool that can promote critical thinking, problem-solving skills and creativity – when used correctly. Human interaction, the educational relationship and collaborative learning in a group remain central elements of education.
When does AI really help with learning?
AI unfolds its full potential when its use is specifically tailored to the current learning phase:
- When preparing for a topic, AI can, for example, impart background knowledge.
- When working on a new topic, AI can provide clear explanations of difficult concepts.
- As an aid to follow-up work, it can provide feedback on a completed task or identify gaps in understanding. Based on this, it generates exercises to fill in knowledge gaps.
As a flexible learning companion, AI is therefore particularly helpful and valuable for independent learning.
Boundaries
What text AI cannot (yet) do
Artificial intelligence is what it is: a technical construct. It cannot therefore establish genuine learning relationships or respond emotionally to pupils. It currently lacks an understanding of the educational process and developmental psychology. It may learn this in future. However, it is doubtful whether it can also learn the powers of observation or human sensitivity of teachers.
AI also (still) has its limitations when it comes to complex, open-ended questions that require interdisciplinary thinking or creativity. It can impart knowledge, but it does not train users to think independently. Education is a complex and social process – one in which AI can be a valuable tool, but can never completely replace humans.
Beware of AI hallucinations
So-called hallucinations occur when an AI provides convincing but factually incorrect information. This happens because the AI always spits out the most probable answers to the question asked – which do not necessarily correspond to reality. Such hallucinations can range from subtle inaccuracies to completely fabricated facts. Research in specialist areas requiring specific knowledge or on topics for which there is little information available online is particularly tricky.
Consequently, a critical attitude towards AI-generated answers is a core competence in media education. Students should be trained to question and examine content (both text and images or videos) and, where necessary, verify it using other sources. In this way, the AI weakness of hallucination can potentially become a learning opportunity for a core 21st-century skill: critical media evaluation.
Recognising and avoiding the dangers of generative AI
All that glitters is not gold. This also applies to generative AI. What are the downsides of AI? What risks should you be aware of?
Topic
Language models as AI agents
Generative AI has evolved from simple chatbots to capable assistants. Instead of just answering questions, modern AI systems can now plan, divide and independently carry out tasks. And they can often do so across multiple steps and using various tools, data sources and applications. In this context, we refer to AI agents or agentic AI.
What is an AI agent?
An AI agent is not a single super model, but a system consisting of several elements:
- a powerful language and reasoning model,
- tools such as web access, file editing and code execution,
- connectivity to apps and data,
- as well as clear control and security mechanisms.
How does this work in practice?
Let's illustrate with an example: You formulate a specific goal, say:
‘Plan dinner with my best friend Daniel today.’
Your AI agent interprets this goal and breaks it down into several subtasks:
- Check availability: When are you and Daniel free?
- Determine location: Where are you right now?
- Find options: Which restaurants are nearby?
- Consider context: What is the weather like? Are there any preferences?
- Prepare reservation: Can reservations be made online?
The AI agent collects all relevant information, compares options and prepares an actionable solution. Critical actions such as bookings or submitting forms are only carried out after you have given your explicit approval.
Where are AI agents used?
AI agents can provide the most support where recurring goals, clear processes and accessible data come together, for example:
- Everyday organisation: Planning appointments, trips, shopping, projects. AI agents can provide support from planning to effective preparation.
- Knowledge and office work: researching, structuring, creating and maintaining documents, presentations or spreadsheets.
- Education and learning: creating learning plans or learning resources from different sources.
- Business: automating support, sales or operations workflows where AI agents perform defined tasks.
Opportunities and risks of AI agents
AI agents have the clear advantage of saving time by automating task chains. Recurring processes are executed consistently and there are low barriers to entry for complex activities such as research, evaluation or planning.
The disadvantages include over-automation, for example when results are accepted without being checked. Or when missing information leads to a misinterpretation of context or incorrect assumptions by AI agents. Security and data protection issues are also risks associated with AI agents, as is potential manipulation through content.
Examples: Where are AI agents already in use?
- ChatGPT (OpenAI) – Agent Mode in the chatbot: available at chat.openai.com or in the ChatGPT app, ChatGPT's AI agent can perform multi-step tasks, create files, code or analyses, but cannot access external websites.
- ChatGPT (OpenAI) – Operator: As a specialised web agent, Operator has been available since January 2025 and primarily handles forms or triggers orders (after prior consultation).
- ChatGPT (OpenAI) – Atlas: Atlas has been available as an AI browser since October 2025. It has an integrated agent mode that can access external websites.
- Claude (Anthropic) – Agent Teams & Code: With Agent Teams, you can have multiple agents work on large projects in parallel. With Agent Code, developers can delegate complex code tasks.
- Microsoft 365 Copilot – Agent Mode: Copilot goes from being a simple assistant to an autonomous helper and can perform tasks independently across multiple apps. You remain in control and confirm important steps.
- Gemini (Google) – Agentic Capabilities: Gemini 2.0 performs multi-step tasks using the Google ecosystem: Search, Drive, Gmail, Calendar. Particularly strong at web research and data analysis.
Useful links
Further content
Would you like more information on the topic of text AI? We have compiled the most important blog posts and links here.
Videos
How to integrate ChatGPT into your daily life.
How to ask the best ChatGPT queries.
Which AI tool did you already use in your everyday school life?
Other interesting topics
Ask Marcel
Marcel is a trainer at Swisscom. He is available to answer any questions you may have about AI.
Marcel
Trainer at Swisscom