Contents
- 1. What Claude Docs is — a conversation that turns straight into a document
- 2. Three ways to create one (including from Claude Code)
- 3. What it actually looks like — from asking to exporting
- 4. Three ways to edit — type it yourself, ask Claude, or call it in a comment
- 5. Charts and diagrams — it pulls the data, but it will not refresh on its own
- 6. Sharing and permissions — your plan changes a lot
- 7. Export — Word, PDF, Markdown and Google Docs
- 8. Docs, Slides and Design: which one to ask for
- 9. The limits you should know before you rely on it
- 10. What it is good for, and what it is not
- FAQ
When Claude Cowork was folded into chat on September 16, 2026, three creation features arrived in beta at the same time: Claude Docs (documents), Claude Slides (presentations) and Claude Design (visual design). This article is about the first of them, Claude Docs.
In one sentence, it turns what comes out of a conversation into a document you can keep editing. Ask it to "write this up as a spec I can share with the team" and Claude drafts it in front of you, asking about anything it is missing before it starts. What you get is rich text with headings and tables, and a single document can hold several tabs — think of them as chapters in a notebook.
But it is a beta, and what is missing is very clearly missing. There is no version history. Deleting is permanent. On a phone you can ask for a document, but you cannot fix one. This article gives those absences as much space as the features, because that is usually what decides whether you can use it.
The short version — three faces of Claude Docs
Source: Claude Help Center, "Get started with Claude Docs"
WHAT IT DOES
Conversation to editable document
Headings, tables, multiple tabs. Edit it yourself, or call Claude in from a comment
WHO GETS IT
Pro, Max, Team, Enterprise
Not on the free plan. It is a beta, and on Enterprise it is off by default
WHAT IS MISSING
History, recovery, mobile editing
No version history, deletion is permanent, no editing on a phone
1. What Claude Docs is — a conversation that turns straight into a document
What Claude Docs produces is a rich-text document stored in your Claude account. It can carry headings and tables, and one document can hold several tabs. That matters when a long piece of material wants to be split into chapters.
Everything you create is saved to the Artifacts tab. You can find and open it from any conversation, so you never have to work out which chat you wrote that spec in.
You could always get Claude to write a document, of course. The difference is what happens next. Before, you copied the output into another tool and from that point on only humans touched it. With Claude Docs, you and Claude edit the same document. And the Help Center states explicitly that every individual edit is attributed — whether a person or Claude made it.
2. Three ways to create one (including from Claude Code)
There are three entry points.
Three ways into Claude Docs
1. Ask for it in a conversation
"Turn this into a spec the team can share." Claude drafts it in front of you and asks about the gaps before it writes. Starting with /docs, or picking Output → Docs in the message box, does the same thing
2. Start from a template
Pick one from the template gallery in the Artifacts tab and describe what you want to write
3. From a Claude Code session
Turn the session you just worked in into a spec, a runbook or a report. On the desktop app it opens in a side panel; from the terminal you get back a link that opens it on the web
If you write software, the third one is the one that pays. Once you have finished an investigation or a fix in Claude Code, all of that knowledge is locked inside the session; the only way back to it is scrolling the log. Now you can convert it directly into a runbook or a report.
That kills the step everyone puts off: writing things up after the work is done. If you run long-running work through Projects, it also gives you a way to leave a record of what each thread actually did.
3. What it actually looks like — from asking to exporting
A list of capabilities does not tell you much about what you actually do. So here is the shortest complete loop, set just after you have finished fixing one bug in Claude Code.
The shortest loop (four moves)
1. Ask
Write this in the conversation.
Turn the investigation and the fix we just did into a runbook
someone else could follow to do the same thing.
Order it: assumptions, how to reproduce, cause, fix, how to verify.
If any assumption is unclear to you, ask me first.That last line is what makes it work. Claude comes back with questions before it starts writing, so you get far fewer clean documents built on assumptions nobody had filled in.
2. Read what came out
The document is saved to the Artifacts tab. If it is getting long, ask for "the verification steps on a separate tab" and it splits into tabs inside the same document.
3. Fix it (this is the part that is different)
Select the paragraph that bothers you, leave a comment on it, and write this.
@Claude The setup explanation here runs long. Cut it to three lines
and add "cases where this procedure will not work" underneath instead.You never have to describe a location in words — no more "the second paragraph of section 3". Small corrections you can just type yourself, and it saves automatically.
4. Get it out
Choose Markdown from Export and drop it in your repository's docs/, or take the Word file to wherever your company keeps documents. Only at this point can you afford to lose it — as covered below, Claude Docs has neither version history nor recovery.
What makes that loop work is that the "write it up properly" step has disappeared. You used to copy the output of a conversation into another tool, and from there on only humans edited it. Now people and Claude take turns on the same document.
4. Three ways to edit — type it yourself, ask Claude, or call it in a comment
There are three ways to make changes, and knowing which to reach for makes you faster.
- Type it yourself: click into the document and write. It saves automatically and appears immediately
- Ask Claude: say it in the conversation — "tighten the structure and add a section on risks"
- Call it in a comment: select text inside the document, leave a comment and mention
@Claude. Claude can also leave comments of its own to explain something
The third one is where this differs most from other AI writing tools. You can point at the place you want changed, in the place itself. No describing a location in words — no "the second paragraph of section 3".
💡 Collaborative editing is real time. Everyone with edit access can work at once, and every single edit records who made it — person or Claude. Claude acts with the permissions of whoever asked it, so it cannot edit on behalf of someone who only has view access. But as covered below, there is no version history, so knowing who did it does not let you get back to how it was.
5. Charts and diagrams — it pulls the data, but it will not refresh on its own
Ask for "a chart in this document" and Claude pulls the data from your connected apps and embeds it. The same goes for diagrams, graphs and timelines.
They do not update automatically, though. The Help Center is explicit, and it holds even when the figures came from a connected app such as Salesforce or Google Sheets: when you want current numbers, you have to ask Claude to go and fetch the latest data.
This is easy to get wrong. An embedded chart is a snapshot of that moment, not a live dashboard. Build a document around this week's numbers and it will still show this week's numbers when you open it next week. Putting the date in the text is the safe habit.
6. Sharing and permissions — your plan changes a lot
A document is yours alone to begin with. Nobody else can open it until you share it, and whoever you share it with needs a Claude account of their own. Beyond that, how you share depends on your plan, and this is worth checking before you commit to it.
| What you can do | Pro, Max | Team, Enterprise |
|---|---|---|
| Share by link | Yes (and only this) | Yes |
| Share with named people or groups | No | Yes |
| Share with the whole organization | No | Yes |
| Share outside the organization | — | No (not possible) |
There are three permission levels: viewer, editor and owner. Editors can read, edit, comment and export. Only owners can delete and rename.
There is no comment-only permission. In other words, you cannot hand a document over saying "please do not change it, but do tell me what you think". Ask someone for a review and they become an editor. For anything meant for outside review, that matters.
7. Export — Word, PDF, Markdown and Google Docs
You pick the format from Export. There are four: Word (.docx), PDF, Markdown (.md) and Google Docs.
On top of that, you can ask Claude to turn the same content into a Claude Slides presentation, which lets you settle the substance as a document first and build slides afterwards. As how to make slides with AI also covered, starting straight from slides tends to leave the structure wobbly, so that order earns its keep.
Markdown export is quietly important. It means the output can live as a document in your repository, so a runbook written up from a Claude Code session can go straight into docs/.
8. Docs, Slides and Design: which one to ask for
All three arrived in beta together, which makes it hard to know which one to ask for. The shape of what you are making decides it.
| Feature | What it makes | Where it fits | Export |
|---|---|---|---|
| Claude Docs | Documents (rich text, multiple tabs) | Specs, runbooks, reports, meeting notes | Word, PDF, Markdown, Google Docs |
| Claude Slides | Slides | Anything you show while reporting, pitching or explaining | PowerPoint, PDF |
| Claude Design | Visual designs and prototypes | Screen concepts, diagrams, visuals | — |
When you are unsure, ask "is this to be read, or to be shown?" Something to be read in order is Docs; something you put on screen while you talk is Slides; something whose form is the point is Design.
💡 The order that works is Docs first, Slides second. Anything you build in Claude Docs can be converted into a Claude Slides presentation on request. Starting straight from slides tends to leave the structure wobbly, and that is not an AI-specific problem — get the argument straight as a document, then turn the settled version into slides, and you finish sooner. How to make slides with AI recommends the same order.
One more thing: the September 16, 2026 merge made Claude Design work inside any conversation. You no longer have to move to a dedicated screen the way you used to. All three can be called from the same conversation.
9. The limits you should know before you rely on it
Being a beta, it is very clear about what it does not have. Here are the items the Help Center lists as known limitations, ordered by how much they hurt in practice.
What is missing, in order of impact
No version history
Nothing to roll back to. A paragraph a collaborator deleted is not something you can restore yourself
Deletion cannot be undone
There is no trash to restore from. Once it is gone, it is gone
No comment-only permission
Ask for a review and the reviewer can edit
You cannot fix it on mobile
On iOS and Android you can still ask for a document in a conversation and read it, but you cannot edit, pick templates or change sharing settings
No sharing outside the organization
This one is about Team and Enterprise. Not usable for external review
Some org settings rule it out
Not yet available under CMEK, ZDR and HIPAA. There is also no audit log of actions inside a document
⚠️ "No history" and "deletion is permanent" are worse together than apart. This is a document several people can edit, with neither history nor recovery behind it. Anything you would hate to lose should be exported and kept somewhere else. Markdown export makes a repository the cheapest place to put it.
10. What it is good for, and what it is not
Once you take the limits into account, the line is fairly easy to draw.
| Good fit | Poor fit |
|---|---|
| Writing up something you settled in a conversation and handing it over | Documents you grow over years (no history) |
| Turning Claude Code work into a runbook or report | External review (no sharing outside the org, no comment-only role) |
| Drafting here, then exporting to Word or Google Docs to finish | Documents that need auditing (no log of in-document actions) |
| Reports built around a snapshot of some numbers | Material that must always show the latest (charts do not auto-refresh) |
| Settling the substance as a document before making slides | Editing on your phone while travelling (mobile cannot edit) |
Put simply, it suits the short run from "make it" to "hand it over" to "get it out". What it is not yet good enough for is being the place things live — no history, no recovery, no audit log. Keep the storage somewhere else and use Claude Docs as the workbench between draft and finished copy: right now, that is where it fits best.
Summary
Claude Docs turns what comes out of a conversation into a document you can keep editing. It is a beta for Pro, Max, Team and Enterprise, and the free plan does not get it. Documents carry headings, tables and tabs; you and Claude can both edit them; and you can summon Claude with @Claude in a comment. You can export to Word, PDF, Markdown and Google Docs, and convert to Claude Slides.
And the most-overlooked strength is that you can create one from a Claude Code session. Work you have already finished becomes a runbook or a report, which removes the step that always gets put off.
On the other side, there is no version history, deletion cannot be undone, you cannot fix anything on mobile, and Team and Enterprise cannot share outside the organization. Start using it as the place your documents live without knowing those four, and it will cost you later. Use it as a workbench and export the finished product. For now, that is the way of working that fits the limits best.
FAQ
Q. Can I use it on the free plan?
A. No. Claude Docs is a beta for Pro, Max, Team and Enterprise; the free plan is not included. Watch out, because the default is the other way round depending on the plan — on Pro, Max and Team it is on by default and you can switch it off under Settings > Capabilities, while on Enterprise it is off by default and nobody can use it until an owner enables it under Organization settings > Artifacts. Note that Claude Docs also depends on Artifacts being switched on in the first place. For organizations running CMEK, ZDR or a HIPAA-ready configuration, it is not available yet.
Q. Can I get back a document I deleted by mistake?
A. No. Deletion is permanent and there is no recovery mechanism. There is no version history either, so you cannot roll back to an earlier state. Export anything you would hate to lose to Markdown or Word and keep it somewhere else.
Q. Can I have someone outside the company review a document?
A. Team and Enterprise cannot share outside the organization. Pro and Max can share by link, so you can hand it over — but with no comment-only permission, the reviewer ends up able to edit it. If external review is part of the plan, exporting to PDF or Word is the safer way to send it.
Q. How is this different from Claude Slides or Claude Design?
A. The shape of what they make is different. Claude Docs makes documents, Claude Slides makes slides (exportable to PowerPoint and PDF), and Claude Design makes visual designs and prototypes. Ask yourself whether it is to be read or to be shown. Section 8 has the full comparison.
