So far, everything you've done has happened inside a chatbot, answering from what it learned in training. This Thing is about something different: what happens when AI goes and looks things up for you.
Traditional search gives you a list of links. Then the work falls on you: clicking through, scanning for the bit that actually answers your question and piecing it together from fragments scattered across different sites. It works, and it's worked for decades. But it's slow.
AI search reads those pages for you, pulls out what's relevant and gives you an answer with its sources, so you can see where each claim came from. And there's a step beyond that. Deep research tools make a plan, run lots of searches, read dozens of pages and come back several minutes later with a structured report.
You'll try both. You'll put the same question to a chatbot and an AI search tool, then hand a bigger question to a deep research tool and judge what comes back. The skill to build here isn't just finding answers faster. It's checking them. If your work involves finding things out or pulling information together, this Thing will change how you do it.
From search engine to answer engine
Search engines find pages for you to read. Answer engines read the pages for you.
Traditional search engines are really finding engines. They find pages that might contain your answer and leave the reading to you. AI search tools are closer to answer engines. They do the reading and present what they found.
AI Overviews: the one you've already met
If you've noticed a block of text above the normal results when you Google something, that's an AI Overview: Google's attempt to answer your question without making you click through to a website. It's handy for a quick question, and you can usually expand it to see the links it drew on. And if you just want the old-fashioned list of links, the Web filter under the search box hides the AI summary.
AI Mode: a conversation with Google Search
Google has also added AI Mode, a separate tab in Google Search (or go straight to google.com/ai). You can ask longer questions, follow up as you would with a chatbot and see links to its sources. It's free and you don't need a new account. Google labels it as experimental and warns that it may make mistakes, which is a useful reminder for everything in this Thing.
ChatGPT search and other tools
ChatGPT search is built into ChatGPT, including the free plan. When a question needs up-to-date information, ChatGPT searches the web and shows citations you can click to open the original source, plus a Sources button listing everything it drew on. To make sure it searches rather than answering from memory, choose Search from the tools in the message box. Gemini and Microsoft Copilot search the web in a similar way.
There are also dedicated AI search tools. The best known is Perplexity, which gives answers with numbered citations for each claim. It has a free plan and is worth a look if you'd like something built purely for search, although far fewer people use it than Google or ChatGPT.
How AI search works (and why sources matter)
An AI search answer is only as good as the pages it read, and how well it read them.
Under the surface, AI search does roughly what you'd do yourself, only faster. It turns your question into one or more searches, picks out promising pages, reads the relevant passages and writes a summary that links each part to where it came from. This is called grounding: the answer is anchored to real, current web pages rather than to whatever the model happens to remember.
That matters because of something you met in Thing 2. A chatbot answering from memory is predicting plausible-sounding text, and sometimes it confidently makes things up. Its knowledge also stops when its training ended, so recent events and fast-changing facts are where it's weakest. AI search reduces both problems. Put simply, a chatbot gives you an answer from memory. An AI search tool gives you an answer from the web, right now, with receipts.
Why the receipts matter
Grounding reduces the risk of errors. It doesn't remove it. An AI search answer can still go wrong in several ways:
- The source is wrong. If the AI finds a poor page, you get a poor answer. When AI Overviews first launched, some repeated jokes from discussion forums, including a suggestion to use glue to stop cheese sliding off pizza. Google acknowledged that some overviews had featured "sarcastic or troll-y content from discussion forums".
- The AI misreads the source. It might take a quote out of context, mix up two pages or state something more strongly than the original does.
- The citation doesn't support the claim. A link next to a sentence looks reassuring, but it only helps if the page actually says what the sentence says.
- Something important is missing. AI search can only read what's public and easy to find. Paywalled research, very recent developments and voices that aren't well represented online may not appear at all.
- Everything sounds equally certain. A claim backed by several strong sources and a claim resting on one blog post are written in the same confident tone.
Two more things are worth knowing. The way you ask shapes what you get: "Why is a four-day week better?" invites a one-sided answer, while "What does the evidence say for and against a four-day week?" invites a balanced one, so your skills from Thing 3 still apply. And if you ask the same question twice, you may get different sources and a different answer, because the tool searches afresh each time.
Deep research: when you need more than a quick answer
Deep research tools make a plan, read dozens of sources and write you a report.
Deep research tools are more ambitious than AI search. Give one a question and it creates a research plan, runs search after search, follows leads from one source to another, reads full pages rather than snippets, cross-references what it finds and writes up a structured report with citations. It's a bit like sending an assistant to the library on your behalf. It takes several minutes, sometimes longer, which feels slow next to a chatbot's instant reply but is remarkably fast compared with doing the same work by hand.
The tools
Two deep research tools are free to try with a limited number of uses. Perplexity has one too, called Research, but its free allowance is very small.
Gemini Deep Research (Google). Choose Deep Research from the plus menu in Gemini's message box and enter your question. Gemini shows you its research plan, the angles it intends to investigate, and you can edit it before clicking Start research. The finished report opens in a side panel and can be exported to Google Docs. It's included on the free plan with usage limits, and at very busy times it may be temporarily unavailable to free users. It can also work with files you upload and, if you allow it, your Gmail and Google Drive. You won't need that for the activity, and it's worth thinking carefully before switching it on (Thing 15 covers privacy in depth).
ChatGPT deep research (OpenAI). Choose Deep research from the tools (the plus button) in ChatGPT's message box. It may ask a few questions to clarify what you want before it starts, and it tends to produce long, detailed reports. Free accounts get a small number of deep research reports, with more on paid plans, and a counter shows how many you have left.
What deep research is good at (and what it isn't)
Deep research is very good at getting you up to speed on an unfamiliar topic quickly: the main arguments in a debate, the options for a decision or what's changed in a field recently. Because these tools follow chains of references, they also surface sources you might not have found yourself, such as government publications, specialist reports and research papers.
They have the same blind spots as AI search, only bigger. Paywalled research, internal documents and very recent developments are often missing. They work best on questions with researchable answers, such as "What does the evidence say about X?", and much less well on original analysis or anything that hasn't been published. They find and organise what's already out there. They don't create new insight.
The biggest risk is a false sense of completeness. A long, neatly structured report in a confident tone feels authoritative, which makes it tempting to trust it wholesale.
Which tool for which job
- A chatbot is best for ideas, drafts, explanations and thinking things through, where up-to-date facts aren't the point.
- AI search is best for a quick, current answer with sources you can check in a couple of minutes.
- Deep research is best for a bigger question with several parts, when you'd otherwise spend an hour or more reading around.
Checking AI research: the supervisor mindset
You didn't do the research. You delegated it. Now it's your job to review it.
Here's an idea that runs through the rest of this programme: the supervisor mindset. When you use an AI research tool, you're not the person doing the research. You're the person who delegated it. The AI is a fast, tireless member of your team who has gone away and pulled something together, and now it's your job to review the first draft.
If you've ever supervised anyone, you know the feeling. You wouldn't forward a new colleague's first draft to your director without reading it, and an AI research summary deserves the same treatment. In some ways reviewing is harder than researching from scratch, because you need enough knowledge to spot problems in something polished and confident. That's why the activity asks you to pick a topic you already know well.
A quick checking routine
You don't need to check everything. Focus on the claims that matter most:
- Check it answers your question. AI tools sometimes answer a slightly different, easier question than the one you asked.
- Click the citations. Open the source and find the passage. Does it actually say what the AI claims?
- Ask who the source is. Is it an official body, a respected publication or an expert? Or is it an anonymous forum post, a company selling something or a page with no author?
- Check the dates. Is the source current enough for your question? Rules and prices change.
- Find better coverage. For any fact or figure that really matters, look for a second, independent source. One AI summary is a starting point, not something you'd cite.
- Look for what's missing. Whose perspective isn't here? What would someone who disagrees say?
These habits draw on SIFT (stop, investigate the source, find better coverage, trace claims back to the original), a well-known approach to checking information online that you'll find in the resources below. Thing 13 goes deeper into hallucinations, and Thing 16 looks at checking what's real more broadly.
Resources to explore
Google's own introduction to AI Mode, one of the tools you can use in the activity. It's free in Google Search and lets you ask follow-up questions and explore the sources behind each answer.
Google's help page on how AI Overviews work, why they "can and will make mistakes" and how to see plain web results instead.
OpenAI's help article on how ChatGPT searches the web, how to switch search on yourself and how to see the sources behind an answer.
Google's overview of how Deep Research works, from building a research plan to browsing sources and writing the final report.
OpenAI's guide to deep research, covering how to start a report and how usage limits work.
Mike Caulfield's short, practical method for checking information online. It works just as well on AI answers as on anything else you read.
Activity: the research showdown
In part 1, you'll compare a chatbot answering from memory with an AI search tool and check the sources. In part 2, you'll hand a bigger question to a deep research tool and review the report like a supervisor.
What you'll need: Google Search (AI Mode is free and needs no account), any chatbot, a free Google account (for Gemini Deep Research) or a free ChatGPT account (for ChatGPT deep research) and a document for your notes.
- Choose your topic. Pick a personal interest or hobby you know well, so you can spot problems in the answers. Keep the question general, with no names, personal details or anything confidential, and avoid questions with a single factual answer. Some examples:
- "What should I look for when buying a second-hand electric bike?"
- "What's the best way to grow tomatoes in a cool, wet summer?"
- "How have the rules of [your sport] changed in recent years, and why?"
Part 1: quick answers (about 20 minutes)
- Ask a chatbot from memory. Open any chatbot (ChatGPT, Gemini, Claude or Copilot) and ask your question, adding: "Answer from what you already know. Don't search the web." Save the response.
- Ask an AI search tool. Ask exactly the same question in Google AI Mode (choose AI Mode in Google Search, or go to google.com/ai) or ChatGPT search (start a new chat and choose Search from the tools in the message box). Ask at least one follow-up question. Save the response and note how many sources it links to.
- Check three sources. Pick three specific claims from the AI search answer and open the source linked to each. Note whether the source actually says what the AI claims, who published it and how recent it is. Then check one claim from the chatbot's answer yourself with an ordinary search. Did either answer get something wrong, or miss something the other included?
Part 2: a deep research report (about 30 minutes, including waiting time)
- Write a bigger question. Broaden your topic into a question with several parts, the kind that would take you an hour or more to research by hand. Use what you learned in Thing 3: say who the report is for and what you want from it. For example: "I'm a keen amateur cyclist thinking about buying my first e-bike. Write a briefing that explains the main types, what matters most when buying second-hand, common problems to watch for and the UK rules I need to know. Flag where sources disagree." Free plans only allow a few reports, so get your question right before you submit it.
- Run your report. Use one of these routes:
- Gemini Deep Research: go to gemini.google.com, choose Deep Research from the plus menu in the message box and submit your question. Read the research plan. Does it cover what you'd expect? Click Edit plan if you want to adjust it, then Start research.
- ChatGPT deep research: go to chatgpt.com, choose Deep research from the tools (the plus button) in the message box, submit your question and answer any clarifying questions.
Don't connect your email, documents or other accounts, because the web is all you need. When the report is ready, save or copy it (Gemini can export it to Google Docs).
- Evaluate the report like a supervisor. Read it carefully and note your answers to these questions:
- Does it answer the question you actually asked?
- Based on what you already know, is anything wrong, oversimplified or presented as fact when it's really opinion?
- Click through two or three citations. Do the sources say what the report claims, and do they look credible?
- How recent are the sources, and does that matter for your topic?
- What's missing? What perspective or evidence would you expect that isn't there?
- How much more checking would you need to do before relying on it?
- Reflect. Write a short paragraph comparing all three approaches. When would you reach for a chatbot, AI search or deep research? What will you always check before relying on AI research?
If you can't use a deep research tool (for example, you've used your free allowance or Gemini is busy), try the other tool or come back later. If neither is available, run your bigger question through Google AI Mode instead, ask at least three follow-up questions to build a fuller picture and evaluate the result in the same way. For a third point of comparison, you could also try Perplexity's Research mode, although its free allowance is very small.
Your output
A document containing your questions, the chatbot and AI search answers with notes on the three sources you checked, your deep research report (copied, exported or screenshotted), your evaluation and your reflection.
Claim your Open Badge
Submit your research showdown document via cred.scot. Include your questions, both quick answers with your source checks, your deep research report, your evaluation and your reflection on when you'd use each approach.
Submit your AI search comparison, deep research report and evaluation to claim this badge via cred.scot.
Claim now