Are ChatGPT shared links public? What to check

Sharing a chat feels like handing one person a document. ChatGPT shared links work more like publishing a page with a very long address. On a personal account, anyone who gets that address can read what's on the page, whether you sent it to them or not. The long random address is the only lock, which gives your chat roughly the security of a spare key under the doormat: fine until someone mentions the doormat.

Yes, on a personal ChatGPT account. Anyone with the link can read the shared chat, though it isn't listed anywhere and OpenAI says the pages aren't meant for search engines. You can delete links under Settings, Data controls, Shared links. Deleting one stops new visits but can't recall copies people already saved.

We read the sharing pages from OpenAI, Anthropic, Google and xAI for this post on November 2, 2026, along with the security reports below. We didn't run our own tests on live links, so the behavior described here is what each company says and what researchers reported.

Photo by Alexander Suhorucov on Pexels

What happens when you share a ChatGPT link

Your regular chats stay out of view until you press Share. OpenAI's help page sets out what changes when you do, and the answer depends on your account.

• Personal accounts, such as Free, Plus and Pro: anyone with the link can view the chat. You can't limit a link to named people, and you can't set an expiry date.
• Business, Enterprise and Edu workspaces: only members of that workspace can open it. People outside see nothing. One difference to watch: a workspace link can include messages added after you shared it.

A personal link is a snapshot. Messages you add later stay out of it until you choose Update link. The link doesn't give anyone access to your account, and it's anonymous by default, though OpenAI warns that some older sharing screens may show your name. Check the preview for that, and for what you typed yourself, since a name in your prompt is still a name. Supported images and uploaded files can appear on the shared page too.

Are ChatGPT conversations on Google?

Private chats, no. Shared ones, sometimes, depending on where the link has been. OpenAI's page says shared pages "are not intended for search-engine indexing", then adds that this doesn't make a link private, because anyone who can open it can forward it.

The worry has a history. In 2025, people found shared chats by filtering Google results to chatgpt.com/share, including one person's resume rewrite. Those chats had been made discoverable through an opt-in setting, which OpenAI removed, saying it "introduced too many opportunities for folks to accidentally share things they didn't intend to."

Other tools have followed, by different routes:

• Claude, July 2026. Shared chats and Artifacts turned up in Google searches, reportedly including health records and private company documents. Anthropic said links appear only when someone has posted them somewhere search engines can see. Its help page now says every shared page tells search engines not to list it.
• Grok, August 2025. Forbes counted more than 370,000 indexed conversations, including business tasks and at least one password. Forbes said the Share button gave users no warning.
• Meta Muse, October 2026. Security firm Claroc, which sells tools for monitoring this kind of sharing, found over 6,800 share addresses in public DNS records without using a search engine. Muse built each address from the shared item's title, so a name like loan-application-draft ended up on file for anyone to look up. It may be the first time a good file name has counted against anyone. Claroc opened a few hundred and says most were harmless, but Ars Technica lists a small-business loan application and a revenue review marked for a CEO among them. Meta paused sharing on October 20 and turned it back on two days later, with new addresses for everything shared before the fix. It says the flaw affected shared artifacts, such as documents and images, and not chat sharing, though Claroc and Ars described shared conversations. Claroc says the old titles stay in DNS history.

Researchers have also measured what people put in these chats. A 2026 study of 100,342 shared conversations on ChatGPT, Copilot and Gemini flagged 8% as privacy risks. Most of those exposed identifiers or locations, and 3% exposed login data such as access tokens. In longer chats, 60% of the disclosures came in the last quarter of the conversation, so the part you'd check last is the part to read first.

How to check if your ChatGPT chats are public

Your account keeps the list. Go to Settings, then Data controls, and select Manage next to Shared links. Anything there can be opened by whoever has its address. A personal data export also lists shared conversations, in a file called shared_conversations.json, if you want a record.

For a second check, search Google for site:chatgpt.com/share plus a distinctive phrase from the chat, such as a client's name. An empty result tells you that Google isn't showing that page today. It can't tell you who already has the link.

How to delete ChatGPT shared links

On the same Shared links screen, choose Delete shared link beside one item, or open More actions and choose Delete all shared links. That's how to unshare a ChatGPT chat without losing it, because the original conversation stays in your history.

Some of the other controls don't do what you might expect:

Three things can outlive the link. Anyone who continued your chat has a separate copy in their own account. Screenshots and archive sites are out of your hands. And a dead page can linger in search results for a while, which Google's Refresh outdated content tool can speed up.

If a password, API key or login was in the chat, deleting the link doesn't undo the exposure. Change it. For a key to an AI service, start with what to do when an API key leaks. Our password generator makes a new one in your browser, without sending it anywhere.

If you've shared client work

Freelancers tend to get here through convenience: you send a client a link to a draft, or send a colleague a chat to get their view. On a personal account, that puts the client's details on a page with no sign-in. The rules on client information in ChatGPT start with your contract, and a confidentiality clause may count the link as sharing, whoever ends up opening it. We aren't lawyers, so read the clause or ask one.

Photo by Alena Darmel on Pexels

Three safer habits cover most of it:

• Send the result, and leave the chat out of it. Paste the finished text into a document or email. The San Francisco bar's paralegal section gives legal staff the same advice: don't share conversations by link, and copy into a secure platform.
• Read the whole preview before you copy the link. Long chats pick up names and figures along the way, and the shared version usually carries all of them.
• Use a workspace when a team needs links. Links in ChatGPT Business stay inside the workspace, one of the differences we cover in ChatGPT Business vs Plus. The admins there can still see members' chats.

If a confidential chat did go out, delete the link first. Whether you need to tell the client depends on your contract and what was in it, and our guide on telling clients you use AI has wording for that conversation.

Are ChatGPT links safe to open?

A chatgpt.com address tells you who hosts the page. It tells you nothing about who wrote what's on it, as two cases this year show.

In October, security firm Parameter showed that a booby-trapped shared chat could run its code in the signed-in session of whoever opened it, read a private chat and send data out. "Opening the shared link was the only required action," it wrote about one version. Parameter reported the flaw on October 18, and OpenAI fixed it by October 26, so you don't need to do anything about this one. It was a research demonstration, and nothing we read says anyone was attacked this way.

The second case was a real campaign. In May, Push Security reported Google ads leading to a fake outage notice on a real chatgpt.com link, which pushed visitors to download malware posing as the ChatGPT desktop app. It also describes shared Claude and ChatGPT chats dressed up as install guides, telling people to paste a command into their computer's terminal. In late October, researchers at Gen Digital described two more campaigns on claude.ai that used paid Google ads and fake install guides, Notebookcheck reports.

OpenAI's advice is to confirm the chatgpt.com address and that a trusted sender gave you the link. We'd add two rules: treat an unexpected link as you would any attachment from a stranger, and never paste a command or download an app because a shared chat told you to. Our guide to spotting AI scams covers the messages these links tend to arrive in.

Shared links in Claude, Gemini, Grok and Meta

The rule is the same almost everywhere: treat a public link as public. The differences are in the settings.

So are your Claude chats public? Not unless you created a public link, and Anthropic's page says files you attached aren't included if you do. Gemini shared links do include uploaded images, which viewers can download, and a link disappears when the chat is deleted from your activity, including by auto-delete. xAI's FAQ warns that a Grok link posted publicly may be indexed like any other public page, and Forbes found that uploaded files could be opened from shared pages in 2025.

How we made this: researched and drafted with AI assistance, with any figures checked against their sources and the post reviewed by Hamza before publishing.

Frequently asked questions

Sources