What Not to Share With ChatGPT in India: 7 Things Students Should Never Paste

In India, never share these seven things with ChatGPT, Claude or Gemini: your Aadhaar or PAN, bank, UPI or card details, your home address with your phone number, medical reports, a full resume with its contact block, anyone else's personal details, and confidential work material. Chats are private by default — only a chat you press Share on becomes public.

Written by the AI Academia team, Kolkata. Last updated: July 2026.

Over the weekend of 25 July 2026, a Reddit post set off a small panic: a site-restricted Google search could return conversations people had shared from Anthropic's Claude. Private chats were never exposed, and the Google results were gone within about two days — but the headlines had done their work. Here is the accurate version, and the more useful question underneath it: what not to share with ChatGPT in India in the first place. (Still deciding which free tools to use? That is a separate guide.)

Can other people see my ChatGPT, Claude or Gemini conversations?

No. Conversations in ChatGPT, Claude and Gemini are private to your account by default. Nobody can browse them, nobody can guess their web address, and they do not appear in search results. Exactly one thing changes that, and it is a button you press yourself: Share.

Clicking share does not send a private message. It publishes a copy of that conversation as a web page, at an address anyone holding the link can open — no login, no permission check. Anthropic said during the July episode that its share links "are not guessable or discoverable unless people choose to share them themselves," per Fortune. True — and also the whole problem. A link that only works if you have it becomes a public page the moment somebody posts it where a crawler can reach.

One clarification: this article is about ordinary chatbots that produce text, not about AI agents that take actions on your behalf — a different topic, covered in our guide to AI agents that take actions.

What actually happened with shared Claude chats on Google in July 2026?

The July 2026 Claude incident was a discoverability failure, not a break-in. A Reddit user noticed on the Saturday that a site-restricted Google search returned shared Claude conversations and Artifacts. TechCrunch reported that 404 Media broke the story on Monday morning, and that by Monday afternoon the same test searches returned nothing. What was in those pages was not trivial: TechCrunch found medical records and clinical trial results carrying patient names, children's names and phone numbers, internal company documents and employee reviews, and both The Next Web and Fortune reported cryptocurrency wallet keys among the material.

Three facts keep this in proportion:

• Private chats were never involved. Hackread stated it flatly: the affected pages were not private chats taken from user accounts. Fortune and The Next Web reported the same.
• No confirmed total was published. TechCrunch noted that no number of exposed chats was independently confirmed, so anyone quoting a figure is guessing. A similar Claude episode in 2025 involved just under 600 conversations, per TechCrunch and The Next Web.
• It was cleared fast — but only from search. By that Monday afternoon the same test searches returned nothing, per The Next Web. Hackread notes that reporting never established whether Google, Anthropic or both took them down, and Fortune reported that already-exposed pages stayed reachable for anyone holding the direct link.

That last point is the uncomfortable one: a page you published cannot really be unpublished. If judging what is safe to paste feels like guesswork, that is normal — most people learn it by getting it wrong once. Our Generative AI program works through these tools on real projects, which is where "can I paste this?" gets answered properly instead of in the abstract.

Why did shared AI chats end up in Google search results at all?

Shared AI chats ended up in Google because of a common technical mistake — one that explains why "we blocked search engines" is not the same as "this cannot be found." Claude's share pages were blocked from crawling in robots.txt while also carrying a header telling search engines not to index them. That combination does not work. As Search Engine Journal explained, a noindex instruction only works if the crawler is allowed to read the page, and a blocked page can still be indexed if other pages link to it. Block the crawler and you block it from seeing your "do not list this" note. So every share link posted on a forum or an open group chat became a listable page.

This is not unique to one company: Search Engine Journal notes that OpenAI pulled shared ChatGPT chats from search in August 2025, and Google blocked Bard transcripts in 2023. The lesson is simpler than the engineering — treat every share link as a publication.

What not to share with ChatGPT in India: the 7 things students should never paste

What not to share with ChatGPT in India comes down to seven categories, and the same seven apply to Claude and Gemini. The reasoning is not that the company will misuse them — it is that a chat you might later share, screenshot or leave open on a lab computer is itself a copy of everything you put into it.

Do not paste this	Why it matters in India	Do this instead
1. Aadhaar or PAN number, or a photo of either	These unlock most Indian services, from SIM cards to bank KYC — and unlike a password, you cannot change them.	Type a placeholder such as "my ID number" and fill the real value in later.
2. Bank account, UPI ID, card numbers, OTPs, crypto seed phrases	Crypto wallet keys were among the things found in the indexed pages, per The Next Web.	Never, in any form. If a task seems to need these, the task is wrong.
3. Full home or hostel address with your phone number	Separately they are ordinary. Together they point at one person behind one door.	Give the city only. "A hostel in Kolkata" is enough context.
4. Medical reports, prescriptions, test results	Medical records carrying patient names surfaced in July 2026, per TechCrunch.	Retype just the question — "what does a high TSH reading mean" — with no name, no report.
5. A complete resume with the contact block attached	Your name, phone, email, address and date of birth in one file is the exact bundle needed to impersonate you.	Delete the contact block first, or use a tool that builds the file in your browser.
6. Anyone else's personal details	A friend's marksheet, a classmate's number, a client's name. Documents with the names and phone numbers of primary-school children were among the exposed pages, per TechCrunch.	Replace real names with "Student A" and "Student B" first.
7. Employer or internship confidential material	Source code, contracts, customer lists, unreleased numbers. Internal documents and employee reviews turned up in the July pages, per TechCrunch, and most offer letters make this a disciplinary matter.	Ask the general version of the question, with your company's specifics removed.

Number five has an easy fix. If the only reason you were pasting a full resume into a chatbot was to get help writing it, our free Resume Builder assembles the whole document inside your own browser — so your name, phone number and address never reach a chat box at all. It costs nothing and needs no payment details.

How do I check and delete my own shared AI chat links today?

You can audit every AI chat you have ever shared in ChatGPT, Claude and Gemini in about five minutes — each product keeps its own list of shared links, in a different place. These paths were checked against each product's official help pages in July 2026; menus get renamed, so look for the same idea if a label differs.

Product	Where your shared chats are listed	What to click
Claude	Settings, then Privacy — a Manage option sits beside shared chats, per the Claude Help Center.	Unshare each link you no longer want live, or switch the chat from Public back to Private.
ChatGPT	Settings, then Data controls, then Manage beside Shared links, per Privacy International.	Delete links one at a time, or use the menu to delete all at once.
Gemini	Settings and help, then Your public links, per Google's Gemini Help.	Use the delete icon on individual links, or Delete all links.

One point worth getting right: deleting the chat and deleting the link are related but not identical jobs. On ChatGPT, deleting the original conversation also invalidates its shared link, per DeleteMe's ChatGPT privacy guide. On Gemini, Google's Gemini Help says a public link stays live only for as long as the associated chat is saved in your Gemini Apps Activity, and is deleted with it. Neither of those undoes a publication, though: anyone who already opened, saved or copied the page still has their copy. So go to the shared-links list and delete links deliberately, rather than assuming a tidy-up of your chat history covered it.

Are my AI conversations used to train the model, and can I turn that off?

On free consumer plans, your AI conversations are often used to improve the model unless you say otherwise — and every major product gives you a switch. In ChatGPT, open Settings, then Data controls, and turn off Improve the model for everyone, per Privacy International. In Claude, the equivalent toggle sits in Settings under Privacy and is worded as an offer to help improve Claude, per Privacy International's Claude guide. Gemini frames it as history: Google's help page says Gemini Apps activity older than 18 months is auto-deleted by default, that you can change that to 3 or 36 months, and that even with activity off, conversations are kept for up to 72 hours.

India also has its own data protection regime now — the Digital Personal Data Protection Act, 2023, whose Rules were notified in November 2025 and are being phased in over 18 months, per S.S. Rana & Co. It governs how companies handle your data, not what you volunteer into a chat box.

What is my AI chat privacy checklist for this week?

This AI chat privacy checklist takes about fifteen minutes and needs no technical background.

1. Audit your shared links in all three products, using the paths above. Delete every link you have no reason to keep live.
2. Turn off model training where you can — ChatGPT's Data controls, Claude's Privacy settings, Gemini's activity controls. Two minutes each.
3. Search for yourself. Put your name and phone number into Google in quotes. If something is public, you want to know before an interviewer does.
4. Adopt one rule for pasting. Would I be comfortable if this exact text appeared on a public web page tomorrow? If not, edit it first.
5. Strip identifiers by habit. Real names become Student A, companies become "a mid-size firm", phone numbers and addresses go. Answer quality almost never drops.
6. Never share a chat that contains an uploaded document.
7. Keep work and study separate. No internship code or client names in a personal account. Ask your manager which tool is approved.

Where does structured learning help with using AI at work?

Every step above is free, and plenty of students do it alone. But if working this out alone feels overwhelming — especially judging what is safe to paste at an internship — that judgement builds faster with someone experienced checking your thinking. AI Academia is a Kolkata-based institute (since 2023, ISO 9001:2015 certified, MSME registered), and our programs are built around using these tools on real projects, which is where "can I paste this?" stops being theoretical. Our mentors are working professionals at companies such as Amazon, Accenture, Myntra, and Airtel. Plans start at Self-Learn ₹4,999 (as of 2026 — confirm the latest at aiacademia.in), and Live Help is ₹11,999 (as of 2026 — confirm the latest at aiacademia.in). We hold a 4.8★ rating from 200+ reviews and have trained 1,300+ students, and we offer project and interview support. We will never promise a guaranteed job — nobody honestly can — but using AI tools without creating a problem for your employer meaningfully improves your odds.

The bottom line on AI chat privacy

The July 2026 story was frightening in the retelling and ordinary in the facts: no private conversations were exposed, the problem was pages people had chosen to publish, and the search results were gone in about two days. The uncomfortable part is not what the companies did. It is that people had put Aadhaar-grade personal information, medical records and other people's phone numbers into a chat box, then pressed a button labelled Share without registering that it meant publish. Spend fifteen minutes this week clearing your shared links — then keep the one habit that beats every privacy setting. What not to share with ChatGPT in India is not really a list to memorise: before you paste anything, just ask whether you would be comfortable seeing it on a public web page tomorrow.