Help center

Guides and workflows for AcaTrove

Find setup instructions, feature guides, and admin documentation for researchers, teams, and institutions.

AI-Powered Document Editing

How the Research AI Panel proposes, previews, and applies edits to your documents.

Professor
Graduate Student
Postdoc
All Users
4 min read

AI-Powered Document Editing

The Research AI Panel can do more than answer questions. On pages with editable content -- such as the Write workspace or Manuscript Mode -- it can propose changes to your document, show you a redline preview, and apply edits with a single click.

How It Works

When you type a prompt that the AI classifies as an edit request, the panel enters editing mode instead of chat mode. The workflow follows three stages: propose, review, and apply.

1. Propose

Type an editing instruction in the panel's text area. Examples:

  • "Improve the clarity of this paragraph."
  • "Rewrite the introduction to emphasize the research gap."
  • "Fix the grammar and tighten the prose."
  • "Add a transition sentence between these two sections."

The AI detects that your prompt is an edit request (as opposed to a question) and routes it to the document editing system. An intent badge appears showing "Edit" along with the target scope.

If the AI misclassifies your intent, click "Ask instead" to re-route the prompt as a question.

2. Review

The AI streams its proposed changes in real time. Two things happen simultaneously:

  • Explanation: An assistant message appears in the chat area explaining what changes are being made and why.
  • Redline preview: The document editing system generates a diff between your original content and the proposed revision. Insertions appear highlighted in green. Deletions appear with strikethrough in red. The document itself is not modified yet.

A "Pending document edit" card appears in the panel with the AI's change summary. This card tells you exactly what will change before you commit.

3. Apply or Discard

Once the AI finishes streaming its proposal, you have two choices:

  • Apply -- Click the Apply button to write the AI's changes into your document. The edits are saved automatically.
  • Discard -- Click Discard to reject the proposed changes. Your document remains untouched.

If the save fails after applying (for example, due to a network issue), a "Retry save" button appears. You can also click "Revert to pre-AI version" to undo the applied edits entirely.

Selection-Based Editing

You can target a specific portion of your document by selecting text before submitting your prompt:

  1. Highlight the text you want the AI to edit in the editor.
  2. Type your instruction in the Research AI Panel (e.g., "Make this more concise").
  3. The AI will edit only the selected text, leaving the rest of your document unchanged.

If no text is selected, the AI applies edits to the entire document. For whole-document edits, there is a 20,000-character limit to ensure quality. If your document exceeds this limit, select a specific section to edit.

Supported Content Types

Document editing works on pages that register an editable document with the Research AI Panel. Currently supported contexts include:

  • Write workspace -- The main writing editor at /write.
  • Manuscript Mode -- Section-by-section editing at /write/manuscript, including agentic critique runs. The AI can also help shape your manuscript's research question, argument, and structure.

The panel automatically detects whether an editable document is available. If not, edit-intent prompts will be treated as questions instead.

Tips

  • Use selection-based editing for targeted improvements. Selecting a single paragraph and asking "strengthen the argument here" produces more focused results than editing the whole document.
  • Review the redline preview carefully before clicking Apply. The green and red highlights make it easy to see exactly what changed.
  • If you are unsure about a proposed edit, click Discard and rephrase your instruction with more specific guidance.
  • The AI considers your research context (linked projects, labs, publications) when making edit suggestions, so linking your document to relevant sources improves the quality of AI edits.