Search and research — AI Basics
AI BasicsWhat can you use it for?

Search and research

Some chatbots search the internet for you, and tell you what they found. Like a fast librarian who fetches books and sums them up. Handy, but you still open the book. Learn how to search with AI, and how to check what it found.

Time 20 minutesChecked on 28-10-2026Read first Audio and video
1/12

First, what do you think?

A chatbot searched the web and gives you an answer with a link. Is the answer right? Have a guess.

2/12

What is AI search?

The whole story in three sentences. Some chatbots can search the internet. They put what they find on the whiteboard, and write an answer from it. The answer comes with links to the sources.

Think of a fast librarian. You ask a question. The librarian fetches a few books, skims them, and tells you what's in them. Handy. But what the librarian tells you is a summary. To be sure, you open the book.

  • Search. One maker says search lets a model answer with up-to-date information, beyond its knowledge cutoff. The answer comes with citations: links to where it found things.
  • Deep research. Google's Deep Research searches hundreds of websites by itself. Within minutes, it writes a report of several pages.
  • Check it. Google adds to Deep Research: check the answers.
3/12

What the predictor does

The predictor reads the search results, and predicts an answer. That's the rule.

Search puts fresh pages on the whiteboard. That helps a lot: the answer can be about today. But writing the answer is still predicting. Google writes that predictions often contain mistakes. So a summary of a source can still get something wrong.

The link is your way in. It shows where the predictor got its information. If you click it, you can see if the source really says that.

When does a chatbot search? One maker gives a list. Recent news, current prices, scores, and things that change. For stable facts, maths or writing, it answers without searching.

Remember

The librarian fetches and sums up. You open the book.

4/12

Try it yourself

Three search answers. What do you check, and how?

Everything runs in your browser. Nothing you type is sent anywhere.

5/12

How do you search well with AI?

Four steps. Try them today on a question about something in the news.

Ask it to search, with sources

Ask it to search the web and give its sources. Then you know the answer isn't only from the training.

Open at least one source

Click a link. Find the sentence the answer is based on. Does the source really say that?

Use deep research for big questions

For a big question, like comparing options, deep research can save hours. Read the report as a start. Check the parts you'll act on.

Compare with a trusted source

For anything important, look at an official or trusted site too. A government site, an encyclopedia, the company itself.

6/12

In practice

Three people use the fast librarian.

Case 1The news presentation

Situation

Mila (13) gives a presentation about this week's news. She asks a chatbot to search. It gives a neat summary with three links.

What most people do

Copy the summary.

Why that doesn't work

The summary is a prediction. One detail may not match the source.

What works better

Click the links. Find the facts she'll use. Put the real sources on her slide.

Why that helps

Her facts come from the source, not from the summary.

Case 2The energy contract

Situation

Robin (40) wants a cheaper energy contract. He asks a chatbot to compare prices.

What most people do

Pick the cheapest one in the answer.

Why that doesn't work

Prices change often. Without search, the answer is old. With search, the summary can still mix things up.

What works better

Use the answer to find options. Then check the prices on the suppliers' own sites.

Why that helps

The chatbot saves searching. The supplier's site has today's price.

Case 3The medical question

Situation

Gerard (58) has a new diagnosis. He uses deep research and gets a long report with forty sources.

What most people do

Trust it, because it looks so thorough.

Why that doesn't work

Forty sources look impressive. But the report is still predicted text, and Google says: check it.

What works better

Use the report to prepare questions. Discuss them with his doctor.

Why that helps

He goes to the appointment well prepared. The doctor knows his situation.

7/12

When not to

  • Not when you know the right websiteTimetables, opening hours, your bank.Go to the website itself.
  • Not as the last word on health, money or lawSummaries can be wrong.Check with an official source or a professional.
  • Not for stable knowledgeMaths or how something works doesn't need a search.Ask without search, or use a book.
8/12

By situation

Use AI search to find sources. Then read the sources, and cite those. Not the chatbot.

9/12

Myths and mistakes

ClaimTrue?Why
With a link, the answer is rightNoThe answer is a summary. Check that the source says it.
More sources means more trueNoForty sources still lead to one predicted report.
AI search replaces looking things upNoIt helps you find sources. You still read them.
Search is always neededNoFor stable facts, maths or writing, it isn't.
10/12

Related topics

Read first

Next

  • Codeprogramming with AI, even if you can't.

Related

11/12

Sources

  • documentationGoogle. Gemini Deep Research.Deep Research searches hundreds of websites by itself and writes a report of several pages within minutes. Google adds: check the answers.
  • explainerGoogle for Developers. LLMs: What's a large language model?.A language model predicts the next token, like a very large autocomplete. A token can be a word, part of a word or one character. Training needs a huge amount of text. Predictions often contain mistakes.

All sources of this course →

12/12

Test

You get 7 questions about the whole page. Get 5 right, and you've mastered the topic.

About the tools in this course. Every tool is described in the same way, in alphabetical order, without a recommendation. Names, prices and features change often: the maker's own site is always right. This site was built with the help of Claude, one of the tools described here.