IntegrAuth Academy
Identity, security and practical AI — free, from zero. Byte-sized lessons, real diagrams, hands-on labs, zero fluff. Two programs, each with its own exam and certificate.
English is hard? Your browser can translate this page into Tamil, Telugu, Kannada, Malayalam or Hindi. Tech words stay in English so you learn the real names.
How?
- Chrome on phone: ⋮ menu → Translate
- Chrome on computer: right-click → Translate to …
- Edge: the translate icon in the address bar
Research and fact-checking with AI
Priya asks an AI about new city rules for e-bike batteries. She gets an answer with 5 links. Only 1 of them is a source she can use. Your goal: check a source before you trust it.
🧪 Interactive lab — enable JavaScript to play with this one.
- An AI can answer in 4 ways. Find out which one you got.
- No links? It answered from memory. A forward has no source at all.
- A link to a source is a Citation. It is a claim, not proof. Open it and find the sentence.
- Check the facts that matter most: numbers you will repeat, facts your choice depends on, and surprises.
4 ways an AI can answer
| Way | Where the answer comes from |
|---|---|
| Memory | What it learned in training. It may invent facts. It knows nothing after its knowledge cutoff. |
| Web search | A few pages it read fast. It can pick weak or old pages. |
| Deep research | Dozens of searches. You get a long report with many claims to check. |
| Notebook | Only the files you upload. It is only as good as those files. |
Check the source
Open the link. Then find the sentence with the find tool: Ctrl+F, or ⌘+F on a Mac. Does it say the same thing, with the same number? Check the date, too.
Next, leave the site. See what other sites say about it. This is called Lateral reading. Last, find where the claim started, like the law or the study itself. That is the Primary source.
Now match each question to the best way.
🧪 Interactive lab — enable JavaScript to play with this one.
Forwarded messages
A forwarded message names no source. Look for a hurry, a link, an OTP request or free money. Then check the official source.
Now check five messages.
🧪 Interactive lab — enable JavaScript to play with this one.
Go deeper: why links fail
A link feels like proof. It only says, "this page backs me up." Tests show how often that is wrong. In March 2025, a Columbia University team tested 8 AI search tools. Each tool got a short piece of a news story. It had to name the story. More than 60% of the answers were wrong. Some tools gave broken or made-up links. In October 2025, journalists in 18 countries checked over 3,000 AI news answers. 45% had at least 1 serious problem, most often with sources. Tools improve, but the habit stays: check the link.
A made-up link with a real-looking title is a hallucination. A real page that the AI read wrong is harder to spot. It is just as wrong. Don't ask the AI if its own link is real. You only get another smooth answer. Open the page yourself.
Go deeper: read like a fact-checker
In a 2019 study, Stanford researchers watched experts judge websites they did not know. History experts and students stayed on the page. They looked at the logo and the "About" page. Those are easy to fake. Professional fact-checkers opened new pages. They read what others said about the site. They judged better, and faster. Try it: search the group's name plus "funding" or "criticism".
Facts change as they are retold. A study says "200 riders in one city". A news story says "most riders". An AI summary of a blog about the story is even further away. So always ask where the claim started.
Many libraries teach this as SIFT: Stop. Investigate the source. Find better coverage. Trace the claim to where it started.
Your question matters, too. "Is it true that…?" invites the AI to agree with you. This is sycophancy. Ask "What does it say?" Then ask, "What is the best evidence against this?"
Go deeper: deep research and notebooks
As of October 2026, most big assistants can search the web. Examples are ChatGPT, Gemini, Claude, Microsoft Copilot and Perplexity. Several have a deep research mode. Google's notebook tool was called NotebookLM. In July 2026, Google renamed it Gemini Notebook. Search engines also show an AI summary above the links. The same checks apply.
Deep research saves hours. But now your job is checking. Tell it what you need: the question, the dates, the kinds of sources. Ask it to show where sources disagree. You can't check every line of a long report. So check the claims that matter most.
A notebook answers only from your files. It can still read a passage wrong, so click through on anything important. Is a file private? First check that the tool is approved for it (safe AI at work).
Go deeper: when a search engine or a librarian is better
- You know what you want. For opening hours or an official form, search or go to the official site. It is faster, and it can't reword the answer wrongly.
- It just happened. Go to news sites. An AI answer may use early, incomplete reports.
- The exact words matter. For a law or a contract, read the original text.
- It isn't on the open web. Try Google Scholar, or ask a librarian. Many libraries help for free.
A page can hide text that gives orders to an AI that reads it. This is prompt injection. So judge the source, not just the summary.
Ask an AI with web search about a topic you know well. Open every link it gives. How many really support the answer?