ShareGPT in 2026: What Happened, and How to Share ChatGPT Chats Now

If you searched for "ShareGPT" expecting to install a browser extension and start sharing your ChatGPT conversations, here is the short version: the original project is no longer maintained, and its GitHub repository was archived in September 2026. The one-click tool that defined early ChatGPT sharing has effectively retired.

But ShareGPT is worth understanding — not out of nostalgia, but because it quietly shaped the open-source AI you use today, and because the problem it solved (sharing an AI chat with someone else) now has better, native answers. This guide covers what ShareGPT was, why it mattered far more than a browser extension should, its current status, and exactly how to share a ChatGPT conversation in 2026.

---

ShareGPT at a glance (2026)

• What it was: An open-source Chrome/Edge extension that added a "Share" button to ChatGPT, turning any conversation into a permanent public link in one click.
• Where it lived: sharegpt.com and the open-source repo github.com/domeccleston/sharegpt.
• Status: Deprecated. Not actively maintained since 2024; API discontinued; repository archived (read-only) in September 2026.
• Why it still matters: The conversations users shared became one of the most influential training datasets in open-source AI.
• What to use instead: ChatGPT's built-in share feature for simple sharing; TypingMind for a heavier, multi-model workflow (the option the maintainer explicitly recommended).

---

What ShareGPT actually did

In early 2023, ChatGPT had no way to share a conversation. If you had a brilliant prompt chain or a hilarious exchange, your only option was a screenshot or a wall of copy-pasted text. ShareGPT fixed that with the smallest possible amount of friction.

The workflow was genuinely one-click. After installing the extension, a Share button appeared inside the ChatGPT interface. Clicking it captured the full conversation, uploaded it, and opened a new tab with a clean, permanent URL you could send to anyone — no account required to view it. The shared page rendered the back-and-forth as readable, formatted text rather than a blurry screenshot.

Around that core, ShareGPT added a few community features:

• A public gallery at sharegpt.com/explore where you could browse conversations other people had shared.
• The ability to save favorite conversations for later reference.
• Comments on shared conversations, so a link could become a small discussion.

It installed on both Chrome and Microsoft Edge, and because it was open source, anyone could inspect exactly what it did with their data. For a 2023 utility, that transparency mattered — you were, after all, uploading your prompts to a third-party server to generate the link.

That last detail is the hinge of the whole story. Every shared conversation didn't just become a URL. It became a row in a dataset.

---

Why ShareGPT mattered: the dataset that trained a generation of open models

Here is the part most "what is ShareGPT" explainers skip, and it is by far the most important.

When thousands of people shared their ChatGPT conversations through ShareGPT, the aggregate became a enormous, diverse corpus of real human–AI dialogue: multi-turn, wide-ranging, and far more natural than hand-written instruction data. In 2023, a team from UC Berkeley, CMU, Stanford, and UC San Diego used it to build Vicuna — an open-source chatbot that fine-tuned Meta's LLaMA on roughly 70,000 user-shared ChatGPT conversations collected from ShareGPT.

The result was a landmark moment for open-source AI. Vicuna produced noticeably more detailed, better-structured answers than the earlier Alpaca model, and the team famously claimed it reached a large fraction of ChatGPT's quality — at a time when "an open model anywhere near ChatGPT" sounded implausible. Later Vicuna versions pushed the training set further: after cleaning (converting HTML back to markdown and filtering low-quality or inappropriate samples), the corpus grew to around 125,000 conversations by version 1.3.

Then the story forks in a way worth understanding:

• The original Vicuna training set was never officially released. Citing various concerns, the authors declined to publish it.
• Because the "real" dataset stayed private, community-collected scrapes filled the gap. The widely-circulated "ShareGPT-90K" dataset (an expansion of an earlier ~52K version), assembled via the ShareGPT API, became the practical, downloadable version that countless other fine-tuning projects reused.

That community dataset is ShareGPT's true legacy. If you have used almost any open-source instruction-tuned chatbot from the 2023–2024 era, there is a strong chance ShareGPT data was somewhere in its lineage. A browser extension for sharing links accidentally became foundational training infrastructure for the open-model ecosystem.

It also raised real questions the field is still working through: the data was user-generated ChatGPT output, collected at scale, and reused for training without a formal release or clear consent framework. Modern instruction datasets have moved toward more explicit licensing and provenance for exactly these reasons — but ShareGPT-era data is a big part of why those norms exist now.

---

The current status: archived and unmaintained

Let's be unambiguous, because several directories are not: you should not plan a 2026 workflow around ShareGPT.

• The maintainer stopped actively maintaining the project in 2024 and recommended switching to an alternative such as TypingMind.
• The public API was discontinued, which breaks the programmatic sharing and collection that made the tool useful at scale.
• The GitHub repository was archived by its owner in September 2026 and is now read-only — the clearest possible signal that no further development is coming.

You may still find ShareGPT listed as a live, available product on third-party SaaS directories, sometimes with recent-looking dates. Treat those with skepticism. Many such listings are auto-generated and don't reflect the deprecation the maintainer has confirmed. A read-only repo and a dead API are the facts that matter.

The deeper reason ShareGPT faded is simple: OpenAI built sharing directly into ChatGPT. Once native share links existed, a third-party extension that uploaded your conversation to someone else's server was both unnecessary and a harder privacy sell. The problem ShareGPT invented a solution for got solved at the source.

---

How to share a ChatGPT conversation in 2026 (the native way)

For the vast majority of people, this is the entire answer — and it is built into ChatGPT itself, free, with no extension:

1. Open the conversation you want to share in ChatGPT (web or app).
2. Click the share icon near the top of the chat (commonly an upward-arrow / link icon).
3. Choose to create a link. ChatGPT generates a public URL (on its official share domain) containing a snapshot of the conversation up to that point.
4. Copy and send the link. Anyone with it can read the shared snapshot without logging in.
5. To revoke access, open your shared-links settings and delete the link.

A few things worth knowing so you don't get surprised:

• A shared link is a snapshot, not a live mirror. Messages you add after creating the link are not included unless you generate a new one.
• Anyone with the URL can view it, so don't share conversations that contain personal data, credentials, proprietary code, or client information. A public link is public.
• You can delete shared links at any time from ChatGPT's settings, which disables the URL.

If native sharing covers your needs — and for most people it does — you're done. The sections below are for the workflows the native feature doesn't fully cover.

---

ShareGPT alternatives worth using

ShareGPT really served two jobs: sharing a single conversation (now handled natively) and acting as a lightweight hub for ChatGPT power users (export, organize, revisit). Here are the real options in 2026, limited to tools that actually exist and are maintained.

ChatGPT's native share — the default answer

What it is: The sharing feature baked into ChatGPT, described above.

Why it stands out: No install, no third party, no extra account. Links live on OpenAI's own domain, and you control deletion. It's the most trustworthy way to share a chat because nothing leaves the platform you already used.

Where it falls short: It shares a static snapshot of one conversation. It isn't a workspace — there's no cross-chat organization, tagging, or multi-model view.

Pricing: Free on ChatGPT's free tier; paid tiers add model access, not sharing mechanics.

Best for: Almost everyone who just wants to send one conversation to someone else.

→ View ChatGPT on ToolCenter

TypingMind — the maintainer's recommended successor

What it is: A polished, third-party chat client that puts multiple LLMs behind one interface. You bring your own API key, and it adds the quality-of-life layer ChatGPT's stock UI historically lacked — organized chats, folders, prompt libraries, and export.

Why it stands out: It's the alternative ShareGPT's own maintainer pointed people toward. If you lived in ShareGPT because you wanted to collect, organize, and revisit conversations (not just fire off a link), TypingMind is the closest spiritual replacement, and it isn't locked to a single model provider.

Where it falls short: It's a paid app and requires your own API keys, so there's setup and metered usage cost rather than a flat consumer subscription. That's a feature for power users and friction for casual ones.

Pricing: Paid (license plus your own model API usage).

Best for: Heavy ChatGPT users who want a durable, multi-model workspace instead of a one-off link.

→ View TypingMind on ToolCenter

Poe — many models, shareable by design

What it is: Quora's multi-model chat platform, where you can talk to a range of assistants and build your own bots.

Why it stands out: Sharing is native to the product — both individual chats and custom bots can be shared via link — and you can compare how different models answer the same prompt in one place. It's a low-friction way to share an AI exchange when you aren't tied specifically to ChatGPT.

Where it falls short: It's its own ecosystem, so a shared Poe conversation isn't a shared ChatGPT conversation. Advanced usage sits behind a subscription.

Pricing: Freemium.

Best for: People who want to sample and share across multiple models without juggling separate apps.

→ View Poe on ToolCenter

Perplexity AI — when the chat is really research

What it is: An answer engine that pairs conversational AI with live web sources and citations.

Why it stands out: Every thread is inherently shareable, and shared pages carry the sources behind each answer. If the "conversation" you want to share is actually a piece of research, a Perplexity thread communicates not just the answer but where it came from — something a raw ChatGPT snapshot can't.

Where it falls short: It's optimized for sourced answers, not freeform creative or coding chats. For those, a ChatGPT share link is a better fit.

Pricing: Freemium (Pro tier unlocks stronger models and higher limits).

Best for: Sharing research-style Q&A where citations matter.

→ View Perplexity AI on ToolCenter

---

Is the ShareGPT dataset still relevant?

For end users: no — it has nothing to do with sharing a chat today.

For anyone training or studying models, it's a different question. The ShareGPT-90K-style datasets remain historically important and are still referenced in open-source fine-tuning discussions, but they come with caveats that have aged into real concerns:

• Provenance and consent were informal, which is increasingly a problem as data-licensing standards tighten.
• Quality and safety filtering in the raw scrapes were limited; the famous Vicuna cleaning pipeline existed precisely because the unfiltered data wasn't training-ready.
• It's a snapshot of 2023-era ChatGPT, so it reflects an older model's behavior, not current capabilities.

If you're assembling training data in 2026, treat ShareGPT data as a historical artifact to understand rather than a default to reach for. Newer, explicitly-licensed instruction datasets exist for a reason.

---

How to choose: a quick decision framework

• You just want to send one ChatGPT chat to someone → Use ChatGPT's native share link. Don't install anything.
• You want a lasting workspace to organize and revisit chats across models → TypingMind.
• You want to compare and share answers from several models → Poe.
• Your "chat" is really sourced research → Perplexity AI.
• You found an old tutorial telling you to install the ShareGPT extension → Skip it. The repo is archived and the API is dead; the native feature does the job.

---

The bottom line

ShareGPT is a rare case where a simple tool mattered twice: first as the fastest way to share a ChatGPT conversation, and then — far more lastingly — as the source of the user-generated data that helped open-source models catch up to the frontier. The extension itself has been overtaken by ChatGPT's own share feature and is now formally archived, so there's no reason to install it in 2026.

What survives is the lesson and the alternatives. To share a chat, use the native link. To build a real ChatGPT workflow, use a maintained client like TypingMind. And the next time you hear about an open model "trained on ShareGPT data," you'll know that a one-click browser button is a bigger part of AI history than it ever looked.

Last updated: November 2026. Status reflects the archived state of the ShareGPT project; features of third-party tools are summarized qualitatively — check each tool's site for current pricing.