Short answer
Generative AI is AI that creates new things: text, images, audio or video. ChatGPT, Claude and Gemini are the best-known chat apps for text, and the program inside them that actually writes the sentences is called an LLM (large language model). An LLM has read a huge amount of writing and learned which words tend to follow which. When you ask a question, it builds an answer one word at a time from those patterns. That makes it strong at writing, summarizing and translating, and weaker at getting exact facts right.
Generative AI vs. LLM: what is the difference?
"Generative AI" is the broad label. It covers AI that writes text, draws pictures, and makes music or video. An AI that sorts photos into categories or filters spam does not create anything new, so it is not generative.
An LLM is the kind of generative AI that works with language. ChatGPT, Claude and Gemini are the apps people use. Inside each app is a language model (GPT, Claude and Gemini models). The app is the chat window and the file upload button. The LLM is the engine that produces the answer.
Generative AI
- Meaning
- All AI that creates new text, images, audio or video
- Example
- A text writer, an image generator
LLM
- Meaning
- A program built to understand and produce language
- Example
- The engine inside ChatGPT, Claude and Gemini
Chatbot (app)
- Meaning
- A service that lets you use an LLM in a chat window
- Example
- The ChatGPT, Claude and Gemini apps
How it makes an answer: it keeps writing the next word
An LLM practiced on a very large amount of text from the web, books and documents. The practice was simple: given some words, predict what is likely to come next. When you ask a question, it does not look up the answer in a mental encyclopedia. It picks the most natural next word, then the next, until the answer is done.
A rough comparison is the next-word suggestion on your phone keyboard, made far larger and far more capable. The comparison has limits. A keyboard suggests a word or two. An LLM looks at the whole conversation and can write several paragraphs, summarize, or translate. Nobody typed in rules for those skills. They emerged from training.
Two things follow from this. First, asking the same question twice can give slightly different answers, because there is a little randomness when it picks words. Second, it can write smooth sentences even when it does not know the facts. That is what people call hallucination.
How is it different from a search engine?
A search engine finds web pages that already exist. Generative AI writes a new answer to your question. So you use them for different jobs.
- ChatGPT, Claude and Gemini now include web search. When it is on, the app reads search results and summarizes them. In some apps it is on by default. In others you turn it on yourself.
- For recent news or exact numbers, turn search on and open the source links.
What it does
- Search
- Finds existing pages
- Generative AI
- Writes a new response to your question
Result
- Search
- A list of links with clear sources
- Generative AI
- One finished answer, sometimes with no source
Good for
- Search
- Today's weather, store hours, official sites
- Generative AI
- Rewriting, summarizing, brainstorming, simpler explanations
When it fails
- Search
- Shows an outdated page
- Generative AI
- Makes up something that sounds right
What it does well and what it does not
- Does well: drafting emails and documents, summarizing long text, translating, explaining hard topics simply, brainstorming, building tables, writing spreadsheet formulas or simple code.
- Be careful with: exact numbers and dates, quotes that need a source, recent news, and legal, tax or medical decisions. The AI can be wrong and still sound sure.
- Cannot do: see files on your computer unless you upload them, or confirm bookings and appointments for you. Unless the app has a connection feature you set up, it works only inside the chat.
- Memory: it remembers what is in the current chat, but often not earlier chats. Some apps have a separate memory setting, so check yours.
A good first thing to try
Do not overthink it. Hand over one small task you already do and see what comes back. Paste the text below as is.
Explain in five sentences why my cat suddenly runs around the house at night, in words a ten-year-old would understand. If you are not sure about something, mark it "not certain".
The explanation will probably be good. Whether a symptom needs a vet visit is a question for a vet.
Before you start
- Do not type passwords, personal ID numbers or confidential work material. How your chats are used depends on the service and your settings.
- Double-check numbers, sources, laws and health information on official sites.
- If the first answer is not right, ask again: "shorter", "more formal", "as a table". Refining through conversation is the normal way to use these tools.
Related terms
- Training cutoff
- The date of the last material the LLM learned from. Without search, it does not know about anything after that.
- Prompt
- The question or instruction you type to the AI.
- Multimodal
- The ability to work with more than text, such as photos and audio.
FAQ
Does AI actually think?
Not the way people do. It produces the most fitting words based on patterns it learned. The results can look like reasoning, which is why it is useful for so many tasks.
Are ChatGPT, Claude and Gemini the same thing?
They are the same kind of service, but the company, the models and the features differ. Pick one to start, and compare with another if you want.
Is what I type used to train the AI?
It depends on the service and your settings. Check the data controls in settings, and avoid typing sensitive information in the first place.
Can I use them for free?
All three have a free option with usage limits. Limits and models change often, so check each service's own help pages.
Keep going
Checked against sources on 2026-11-05. Sources: Anthropic, Best practices for prompt engineering, Google, Gemini Apps help: upload files. How we write