Magi Copilot: An AI Marketing Tool Built for Reviewable Changes

Most AI marketing tools make changing a draft easier. That creates another problem: once AI can rewrite copy, replace images, resolve feedback, and pull in new evidence, teams need to know what changed before they approve it.

Magi Copilot works inside an active marketing asset and proposes changes for review before they become part of the document. Recent updates make those changes easier to preview, attribute, trace, reverse, and support with evidence.

Magi Copilot is Magi’s AI marketing assistant for reviewing and refining briefs, outlines, drafts, and templates inside the workspace where the work already lives.

What Is Magi Copilot?

Magi Copilot is an AI marketing assistant built around Magi's custom editor, the same workspace where you turn a brief into a finished marketing asset.

Copilot starts with a relevant layer of brand context and can pull more from BrandOS when the task calls for it. It also has access to the active document, chat history, attachments, and the relevant brief or outline at each turn. Based on these inputs, Copilot can evaluate a request against what’s already on the page, not just the latest prompt.

Copilot can also help when creating templates. In “Template Copilot“, you can paste a Figma URL or describe the template you need, then create a reusable starting point for future marketing assets.

Since your copy, sources, brand rules, layout, and review decisions stay in one place, Copilot can help you improve the work without breaking the page. It’s worth noting that it cannot publish or approve content for you. It’ll only make changes after you apply them.

It works beside the document because its job is to assist you while you work, not take on a task and return later with an answer. That’s what Magi’s Agents are for.

How Does Magi Copilot Make Drafts Publish-Ready?

Preview AI-Assisted Changes Before You Accept Them

Copilot proposes changes inside the active document, so marketers can compare the current version with the suggestion before applying it.

Select a passage, tell Copilot what needs to change, and review the original against the proposed version. You can apply the suggestion, discard it, or continue working on it.

That same experience now works more reliably for images.

When Copilot suggests image alternatives, each option renders in the editor’s ghost preview. Switching between alternatives updates the preview before anything is accepted.

We’ve also unified how alternatives are merged into proposed edits, reducing cases where the applied result differs from what appeared in the preview.

Because the edit is limited to that passage, you do not have to replace the rest of the draft or paste in a new block of copy.

Say you have a comment from an editor asking you to fix something in a passage. You can select the “Resolve with Magi“ option to get Copilot’s help in resolving it. When you choose the “Resolve with Magi“ option, it opens the referenced block in Copilot with the reviewer’s feedback already attached.

How this is used: A content marketer spots a claim in a product post that overstates the available evidence and asks Copilot to revise that sentence without weakening the surrounding argument.

Brings the Right Context Into the Document

Ask Copilot to look beyond the draft when you need product details, customer proof, or internal source material. It can search your knowledge base, ideas, the web, and connected tools such as GitHub and Notion for relevant suggestions.

You can also add any specific files, ideas, voice and tone specifications, or sources in Magi’s Knowledge Base.

Attachments are limited to three knowledge sources and three files, so Copilot stays grounded in the material you choose.

Copilot can also add citations where the evidence supports the claim inside the draft. You can cite sources already in Magi’s Knowledge Base or add a relevant external link.

How this is used: A product marketer needs customer proof for a launch post and asks Copilot to search the knowledge base for a relevant testimonial. Copilot identifies where it belongs, explains its role, and adds a citation to support the claim.

Checks if a Passage Belongs in the Draft

You can ask Copilot to explain how a passage supports the argument, fits the audience, and advances the structure. It can then recommend whether you should keep it, revise it, or cut it. Copilot follows your intent and asks for clarification when a request is ambiguous or conflicts with the document’s constraints.

You can also run a named team skill (Skills are your team's own instructions, saved once and callable in any document), such as a humanizer or editorial audit, on a specific section or the full draft. You can type a slash in the chat to view the skills available for your workspace:

How this is used: A content lead runs the team's humanizer skill on a dense section, then asks Copilot whether the revised version still supports the article's argument.

Refines Images Without Leaving the Draft

Use Copilot to edit or regenerate an image, compare new directions, swap in approved brand assets, or adjust the crop, placement, text overlay, alt text, and surrounding layout. You can refine the visual and the copy from the same conversation.

How this is used: A product marketer asks Copilot to give a launch hero more whitespace for the headline, update the overlay text, and keep the rest of the visual treatment intact.

Resolves Brand Audit Findings

Brand Audit Agent checks the draft against your configured brand rules and flags potential issues for review. Copilot prioritizes the findings, explains each proposed fix in plain language, and shows the original alongside the proposed change.

You can apply or discard each change manually, or select a group of proposed changes and have Copilot apply them together.

When you use Copilot to resolve brand violations, applied resolutions remain recorded as Copilot-assisted changes.

How this is used: Before publishing a launch article, a brand lead selects the highest-severity findings, reviews the original and proposed versions, and applies only the changes that preserve the product's intended meaning.

AI-Assisted Edits Now Keep the Marketer as the Author

When a marketer accepts a Copilot suggestion, the resulting version is attributed to that marketer and marked AI assisted in version history.

Previously, Copilot activity could create versions attributed to “Magi.” In some cases, identical “Generated by Magi” snapshots could also appear even though the document itself had not changed.

The updated attribution records both parts of the interaction: Copilot contributed to the edit, and the marketer decided to accept it.

That distinction becomes increasingly important as AI takes on more of the editing work inside a shared document

Review AI Changes With Version History, Diffs, and Restore

Trying an AI suggestion shouldn’t put the current draft at risk.

Magi’s editor now includes a version-history rail that updates as collaborators save. Teams can preview previous versions, name important versions, and see better author attribution across human and agent-created work.

Where available, Yjs-powered diffs show what changed between versions. Older documents use a plain-text fallback.

Restoring an earlier version is also non-destructive. Magi snapshots the current document before the restore, preserving the version you were working on.

Together, preview and version history give teams visibility on both sides of an AI-assisted change: what Copilot is proposing before it is accepted, and what happened to the document after it was applied.

Edit Citations Alongside AI-Generated Copy

A factual edit can change more than the sentence. It can change the evidence required to support it.

Marketers can now add, edit, and remove citations directly in context, attach multiple sources to a selected passage in one action, and cite charts as well as prose.

Chart citations are included in the document’s references section on export. Source search now includes titles, excerpts, and source-type badges to make the supporting material easier to inspect.

Copilot can already work with the knowledge and sources surrounding the active document, including selected files, ideas, and connected knowledge. The richer citation workflow gives the marketer more control over which evidence remains attached as the copy changes.

Stop and Resume Copilot Requests Without Losing Partial Work

Long Copilot requests don’t have to run to completion.

Copilot and Template Copilot requests can now be stopped while they’re running. Partial output remains available, and Magi marks the request as Interrupted by user so there’s no ambiguity about why the response ended.

Streaming can also recover after a connection drops. Sequence-based deduplication reduces repeated output after reconnection, while throttled rendering helps keep longer responses from slowing down the browser.

We’ve also fixed a production issue that could cause the same Copilot response to be delivered twice following reconnects or deployments. The underlying streaming system now uses safer ownership and resume behavior to prevent duplicate responses.

These are reliability changes, but they affect a simple expectation: stopping or losing a connection shouldn’t mean losing the work already completed.

Copilot Now Retains Relevant Tool History During a Conversation

An editing conversation accumulates useful context.

Copilot may search for evidence, inspect workspace knowledge, or use another tool while working through a document. A later request may depend on something it found several turns earlier.

Relevant tool calls and their results are now preserved correctly in Copilot’s conversation history, allowing that earlier work to inform subsequent requests.

This reduces cases where Copilot loses track of research or actions it already completed during the same conversation.

Better Handoffs Between Copilot, Comments, and Manual Editing

AI-assisted editing still involves reviewers and direct changes from the marketer.

Comment resolution has been improved with Copilot-specific handling for referenced feedback. When an editor leaves a comment on a passage, the referenced block and feedback can be brought into Copilot while the marketer works through the resolution.

We’ve also improved manual editing for LinkedIn content. A single click now places the caret and allows direct editing, with more reliable selection behavior across social and read-only modes.

The result is a cleaner handoff between a reviewer identifying an issue, Copilot proposing a change, and the marketer making the final adjustment themselves.

How Does Magi Compare With Jasper, ChatGPT, and Claude?

Jasper Canvas and Grid help teams create campaigns and scale structured marketing workflows. ChatGPT Projects and Tasks organize contextual work and scheduled prompts, while Claude Artifacts and Cowork combine editable artifacts with delegated work. Magi Copilot closes a different gap: helping you diagnose and resolve brand issues, validate claims, and make careful edits in the draft you're working on.

What’s New in Magi Copilot?

Recent releases improve how AI-assisted changes are previewed, attributed, recovered, and grounded inside active marketing documents.

• Image suggestion previews: Compare image alternatives in the document before accepting one.

• AI-assisted attribution: Accepted suggestions remain attributed to the marketer, with an AI-assisted signal in version history.

• Version history: Preview, diff, name, and safely restore previous document versions.

• Citation editing: Add, edit, or remove citations, attach multiple sources, and cite charts.

• Stop and resume: Interrupt Copilot requests while retaining partial output and recover streams after a disconnect.

• Streaming reliability: Reduce duplicate Copilot responses after reconnects and deployments.

• Conversation history: Preserve relevant tool calls and results for later requests in the same conversation.

• Review and manual editing: Improved comment resolution, selection behavior, and direct LinkedIn editing.

  
  

What’s Next for Magi Copilot?

We’re continuing to improve the copy quality of Copilot’s proposed changes, with the aim of reducing the cleanup required after a suggestion is generated.

We’re also expanding the context Copilot can work with through additional integrations. As goal-setting and analytics workflows become available elsewhere in Magi, that context can eventually inform the decisions Copilot makes while helping marketers review content.

These capabilities are in development and aren’t part of the live Copilot product today.

What Should an AI Marketing Tool Do During Review?

AI marketing tools are getting increasingly capable at changing content. That makes the review layer more consequential.

Teams need to see a proposed change before accepting it. They need to know who approved it, inspect the evidence behind factual claims, understand what changed between versions, and recover earlier work when an edit doesn’t hold up.

That’s the direction behind the latest Copilot releases.

Copilot can propose the change. The marketer decides whether it deserves to stay.

New to Magi? Book a demo to review and refine a live marketing asset with Copilot.

Already using Magi? Open an asset or template and launch Copilot from the editor.

Frequently Asked Questions