- Practice
- About 15 min
- 5 practice questions
Research and summaries with AI: notes, flashcards and checking against the original
Ask an AI tool to summarise text you supply, turn a long notice into key points and check every line against the original, make notes and flashcards, and verify any source an AI names.
If something looks wrong, please tell us.
By the end of this lesson you can
- Write a prompt that summarises text you paste in, without letting the tool add outside facts.
- Pull key dates and eligibility out of a long notice and check each line against the original document.
- Turn study material into notes and flashcards, and test whether a source an AI names is real and says what is claimed.
Aspirants read a great deal: long recruitment notices, syllabus documents, office circulars, newspaper editorials and textbook chapters. An AI chatbot can shorten this reading by producing summaries, notes and question cards in seconds. The catch is that a summary can drop a condition, change a date or add something that was never in the text. This lesson teaches a routine that keeps the speed and removes the risk: give the text, ask narrowly, then check against the original.
The routine for AI summaries
- Give the text Paste the text and say 'use only this text'.
- Ask narrowly Ask for page references and 'not stated' for gaps.
- Check against the original Check every date, figure and condition before you rely on it.
Summarise text you supply, not the tool's memory
A chatbot asked 'summarise the rules for this exam' answers from patterns it learned during training, which may be old or simply wrong. A chatbot given the actual text and told to use only that text has far less room to invent. It can still misread or leave things out, so the original stays the authority. A good summarising prompt has four parts: the text, the purpose, the format and a limit.
Use only the text below. If something is not stated in it, write 'not stated'. Give each point with the paragraph or page it comes from.
Asking for the paragraph or page number for each point is the most useful habit of all: it turns checking from re-reading the whole document into looking up a handful of references. Writing 'not stated' instead of guessing makes gaps visible instead of quietly filled.
Worked example 1: a long notice turned into key points
Suppose a candidate has the PDF of a fictional recruitment notice running to twenty pages. The notice is public, so its text may be pasted into a chatbot (a personal application form with your own details may not). The candidate asks: 'Using only this notice, list in a table: last date to apply, date of the exam if given, age limit and the date it is counted on, educational qualification, application fee by category, and the page each comes from. Write not stated for anything missing.'
The table comes back neat. Now the checking step, line by line. Suppose the summary says the last date is 15 May while page 3 of the notice says 15 March: the AI has made an error that could cost the candidate the form. Suppose the summary gives the age limit but leaves out that the age is counted 'as on' a particular date, or drops the line on relaxation for some categories: these are omissions, and they change who is eligible. Each such line is corrected from the PDF, not from the chatbot. Only after every row matches the notice is the table safe to keep or share with friends. (The notice, dates and errors here are illustrative.)
- Open the original next to the summary.
- For each row, go to the page the AI cited and read the sentence yourself.
- Check every number, date and name digit by digit.
- Look for conditions the summary dropped: 'as on', 'only for', 'except', 'subject to'.
- Mark the summary checked, with the date, or throw it away.
Notes and flashcards
For study material, the same tool can produce revision notes and flashcards. Ask for a format you can actually use: 'Make 10 question-and-answer flashcards from this chapter; one fact per card; answers under 15 words.' Cards with one fact each are easy to check against the chapter and easy to revise. Read through every card once with the book open: an AI card with a wrong answer is worse than no card, because you will memorise the mistake.
| Task | Good prompt idea | What you still check |
|---|---|---|
| Chapter summary | Use only this text; 8 bullet points; plain language | Nothing added, nothing important dropped |
| Key facts from a notice | Table with page numbers; 'not stated' if missing | Every date, figure and condition against the PDF |
| Flashcards | One fact per card; short answers | Each answer against the chapter |
| Simpler explanation | Explain this paragraph for a beginner with one example | That simplifying did not change the meaning |
Asking for sources and verifying them
When you ask a chatbot for sources, it may give real ones, out-of-date ones, or references that look perfect but do not exist - a known failure called a fabricated citation. Treat every source an AI names as a lead, not a proof. Open it. Confirm that it exists, that it is the official or original publisher, and that it actually says what the AI claimed. If you cannot find it, the claim is unsupported.
Checking a source an AI names
- Open it
- Confirm it exists
- Confirm the publisher It should be the official or original publisher.
- Read what it says It must actually say what the AI claimed.
Worked example 2: an AI search answer
Many search tools now show an AI-written answer above the results. Suppose a candidate searches for a recruitment's last date and the answer box gives one. That answer was assembled from web pages, which may include an old year's notice, a coaching blog or a page with a typing error. The safe move is to click through to the official notice on the recruiting body's own website and read the date there. The AI answer saved a search; it did not replace the source.
Key takeaways: paste the text and say 'use only this text'; ask for page references and 'not stated' for gaps; check every date, figure and condition against the original before you rely on it. Flashcards are checked once against the book. Every source an AI names is opened and read, and AI search answers point you to the original - they never replace it.
Check your understanding
-
1. A candidate wants an AI summary of a recruitment notice. Which prompt gives the summary that is easiest to check?
- Tell me everything about this year's recruitment.
- Using only the notice pasted below, list the key dates and eligibility with the page each comes from; write 'not stated' if missing.
- Summarise this notice in your own words and fill any gaps from what you know.
- Give me a short summary; I will trust it.
Show answer
Answer: B. Supplying the text, limiting the tool to it, asking for page references and making gaps visible with 'not stated' lets every line be checked quickly against the notice. Filling gaps from memory invites invented details.
-
2. An AI summary says the last date to apply is 15 May. Page 3 of the official notice says 15 March. What should the candidate do?
- Average the two dates to be safe
- Ask the chatbot again and accept whichever date it repeats
- Follow the official notice, correct the summary, and check the other lines too
- Follow the summary because it was made more recently
Show answer
Answer: C. The official document is the authority. One error shows the summary cannot be trusted line by line, so the rest of it is checked against the notice as well. Repeating a question to the chatbot does not verify anything.
-
3. A chatbot gives three references for a claim. One of them cannot be found anywhere. The best conclusion is:
- The reference may be fabricated, so the claim is unsupported until a real source confirms it
- The reference must exist because the chatbot gave its full title
- The claim is true because two other references were given
- Search engines miss many pages, so the reference can be cited anyway
Show answer
Answer: A. AI tools can produce citations that look real but do not exist. A source counts only after you open it and see that it says what is claimed; until then the claim is unsupported.
-
4. Which of these is safe to paste into a public AI chatbot for summarising?
- A colleague's service record from the office file
- Your filled application form with Aadhaar number and date of birth
- An email containing a one-time password (OTP)
- The text of a public recruitment notice
Show answer
Answer: D. A published notice contains no private data. Personal records, identity numbers and OTPs must never be pasted into AI tools.
-
5. Why should a student read through AI-made flashcards with the book open before revising from them?
- Because flashcards must be printed before use
- Because a wrong card would be memorised as if it were correct
- Because AI tools cannot write questions
- Because each card must be longer than 50 words
Show answer
Answer: B. Flashcards are for memorising. If a card's answer is wrong, revision fixes the mistake in memory, so each card is checked once against the source.
Sign in to keep track of the lessons you finish.