The smart clipboard for the AI era. Paste a long Claude, ChatGPT, Cursor, or Gemini answer, mark every passage that needs work, and pass the whole batch back to the model in one paste — each note anchored to the exact words it is about. It is a station your drafts pass through, not an archive they live in.

Why does the model keep rewriting the wrong paragraph?

Because the chat box can carry your sentence but not your finger. Typed as prose, fix the third bullet — no, the second gives the model nothing to point at, so it rewrites a paragraph that merely sounds close. Mark the actual passage and the note ships with its verbatim quote and the heading above it, so the model lands on one target instead of guessing.

How do I send five corrections at once instead of giving up at three?

Mark all five on the rendered answer, then hit Copy — every note bundles with its quote into one block you paste back. The writing tax that made points four and five not worth it is gone, so they ship in the same turn. One paste, five anchored fixes, no second round.

Can I see what the model changed between versions?

Yes — that is the Auto-Diff. Paste the model’s new version and PassbackAI lines it up against the last one and highlights exactly what changed, so you confirm the fix landed without re-reading the whole answer. It is the clearest proof that drafts here are in transit, not filed away.

Reviewing a long AI answer takes forever — does this even help?

Yes — that’s the quiet half of the job. The answer lands as a wall of dense text in a cramped chat, and re-reading it to find what’s off is the real time sink. PassbackAI opens it as a real document with a table of contents, and the Auto-Diff shows what changed since the last version — so you review what matters instead of reading all 1,800 words again.

Claude wrote something I don’t fully understand — can I do anything with that here?

Yes, and it’s the same move as fixing it: mark every passage you don’t follow and send explain these back in one paste, instead of asking one question at a time and losing the thread. PassbackAI doesn’t decide whether the output is correct or safe to ship — it just makes it effortless to point at exactly what you don’t get and let the model account for it.

Where does my draft go — is it saved anywhere?

It waits locally in your browser while you work the task, and rolls off the bench on its own as newer drafts come in (pin one to keep it longer). Nothing is uploaded, archived, or sent to a server — the text is ephemeral by design; the habit is what’s durable.

Is it private? Do I need an account?

No account, no email. Your document lives in your browser, and the single network call is an anonymous page-view ping that never carries your content. Routing a doc to a reviewer is the one explicit, opt-in step that stores it on our backend — everything you do locally stays local.

How does sharing work with no server?

One click encodes the whole document — text and your marks — inside the link itself, in the URL fragment the browser never sends to a server. Your colleague opens it with no login, adds their own layer of notes, and passes it back. Add a password and the link unlocks only for the people you tell.

Does it work with Claude, ChatGPT, Cursor, and Gemini?

Any chat that takes Markdown in and out is a fit — paste the answer, mark the fixes, copy the block back into the same chat. There is also a Chrome extension that loads your copied answer in one click, so you skip the paste step entirely.

How do I connect this to Claude?

Connect the PassbackAI MCP server and Claude does the carrying: it routes a document to you for review and pulls your marked-up feedback back on its own, no copy-paste. Add the URL in Claude’s Settings → Connectors and authorize it — the Connect to Claude page walks it through. Routing a doc is the one opt-in step that stores it on our backend; everything you do by pasting locally still never leaves your browser.

Can I turn a brief into questions before the model guesses?

Yes — the /passback skill reads any brief, PRD, or transcript and weaves ONE interactive document from it: every unclear point gets the widget that fits it (a single choice, a multi choice, a ranking, an open question), and the settled thinking stays prose you can comment on. Paste it in, answer each point in ninety seconds, and pass the answers back so the model proceeds on your decisions, not its assumptions. Full note at /blog/extract-open-questions-from-any-text.

Can I work on more than one draft at once?

Yes — your recent drafts sit on the bench (the last twenty, most-recent first), and you switch between them from the top bar. Each waits locally while you work it; pin the one you are not done with so it stays put.

What about Lovable, v0, or Cursor specifically?

Same move, with per-tool nuance. For AI builders, quote the exact strings and prefix change only what I quoted — the AI-builder playbook has a per-tool note for v0, Replit, Base44, Bolt.new, Sticklight, and Lovable. For Cursor, switch to chat mode and use paired-passage feedback — the Cursor playbook covers it.

Yes — free, $0, with unlimited documents, comments, and exports, and no account to create. There is no commercial tier and no paid plan to upsell you to, because the thing it replaces — typing feedback into a chat box — was never a billable problem. It is a station your drafts pass through, not a subscription you manage.

Mark up an AI's answer like a document — send the fixes back in one paste.

Claude, ChatGPT, Cursor, and Gemini hand you a wall of text that's mostly right. Instead of retyping corrections into a tiny chat box, open it as a real document, highlight every line that needs work, and copy the whole batch back at once — fixes, cuts, and "explain this" included.

Free — $0, unlimited, no account. It runs in your browser — nothing to meter, nothing to paywall.

Chat box vs PassbackAI

The same long answer, two ways. (The right column carries one comment — scoped to the whole column, not retyped per cell.)

The job	The chat box	PassbackAI
Send five fixes at once	you give up after three	five notes, one paste
Re-read a dense answer	scroll a tiny field	a real document, with a contents rail
Ask what you don't follow	one question at a time	mark them all, explained in one paste
Point at the exact line	one text field — you can't	every note anchored to its words
Loop in a teammate	paste elsewhere, lose the thread	one link, no sign-up
Keep it private	it lives on a server	your draft stays in your browser until you choose to share

1 · Mark every fix where it lives

Highlight a line, drop a note, keep reading. Each note stays pinned to the exact words — and remembers the heading above it, so the model learns what to change and where. The tenth fix is as cheap as the first.

Take a sentence the model oversold:

Per my previous correspondence, kindly be advised the deliverable shall be furnished by end of week.

One Delete, struck through — no retyping "cut this" down a long page.

2 · Send it all back in one paste

Hit Copy and every note bundles with its quote into a single block your model can act on — nothing scattered across a chat, nothing lost:

# Edits:
- ...furnished by end of week.

# My Feedback:
> the deliverable shall be furnished
Give a real date, not "end of week".

3 · Paste the next version — see what changed

Drafts move fast, and you go text-blind. Paste the model's new version and PassbackAI lines it up against the last one and lights up exactly what it changed — so you confirm the fix landed without re-reading the whole answer.

4 · Or hand it to a teammate — one link, no sign-up

Need a human in the loop? Share the whole document as a single link — empty for a colleague to fill in, or already marked up. No account for you, none for them.

One click turns the doc into one link. No sign-up, ever — for anyone.
Send it blank or fully commented; they open it and keep going right from your marks.
Sensitive? Add a password and the link unlocks only for the people you tell.

5 · Private by design, not by policy

Your documents live in your own browser's localStorage. Nothing is uploaded unless you choose to share it; the network traffic is anonymous same-origin analytics with no document payload — plus, if a document embeds a video, the video itself loads from YouTube — open devtools and check. When you share, the entire document rides inside the link itself, in the URL's hash fragment — the one part of a URL browsers never send to any server. An optional password encrypts the link end to end (AES-256-GCM via WebCrypto, PBKDF2 at 600K iterations).

The marks above are live. Highlight any sentence on this page and a note opens right where you marked — that's the whole tool, working on itself.

Paste an answer Read the manifesto →