At breakfast, one assistant helps you research a profile and shape the angle. After lunch, another is better suited to a structural edit. The first conversation holds six abandoned openings, a source you rejected, and a joke that does not belong in the piece. The second assistant does not need that room. It needs the editorial brief that came out of it.
Preserve the audience, voice, and claims. Leave the chat behind.
Split the writing process at real editorial boundaries
A practical division looks like this:
- Research: gather leads, questions, and source references for you to verify.
- Outline: turn the checked material into a sequence with a clear promise to the reader.
- Draft: write against the approved outline and named voice constraints.
- Edit: test structure, clarity, repetition, tone, and claim support without reopening every decision.
The stages do not need four assistants. Two assistants can alternate. What matters is that each stage ends with a documentable editorial state rather than “continue where the other model stopped.”
Write the handoff for a cold editor
Tell the next assistant:
- who the reader is and what they should understand or do;
- the piece’s purpose, format, length, and publication context;
- the approved angle and outline;
- verified claims with source titles, URLs, and relevant passages;
- voice rules expressed as observable choices;
- words, claims, and approaches to avoid;
- decisions already settled and questions still open;
- the exact edit you want now.
“Keep my voice” is weak. “Use short declarative openings, no rhetorical questions, and keep technical terms when they are the precise word” is reviewable. So is “do not strengthen this source’s correlation into causation.”
Do not carry the whole chat for atmosphere. A transcript preserves every false start while obscuring which version you chose. AI context does not transfer; a briefing reconstructs the portion the editor needs.
Move the briefing, not the manuscript
RelayLink carries selected text. It does not attach a manuscript, source folder, comments, tracked changes, or browser state. Include a document title, version, section name, or authorized artifact URL. The destination assistant needs its own permitted access to the actual draft and sources.
Connect both assistants to the same RelayLink account and address the package to your own email. Draft it, review every field, then confirm. The model never sends on its own. The package creates no contact request, email, or magic-link page, needs no second account, and remains unread until one of your assistants pulls it. You can also inspect it in the account portal.
Sending a briefing to yourself does not synchronize sessions. It gives the next assistant only the briefing you approved.
Keep authorship and fact-checking with you
An editor model may flatten a deliberate voice, restore a rejected claim, or make a sentence more certain than its source. Compare its changes with the brief and the manuscript. Verify quotations, names, dates, links, and factual claims against primary material.
RelayLink’s provenance helps distinguish wording you authored from wording you approved, but it does not decide whether the prose is accurate or sounds like you. Provenance labels are a record of process, not a substitute for editorial judgment.
When this workflow is a poor fit
Keep the work in one place when the next task depends on dense inline comments, layout, tracked changes, or a source set the second assistant cannot access. The handoff is also wrong when a publisher, employer, client, or source agreement forbids the destination provider from receiving the draft or research.
For confidential interviews, unpublished material, or identifiable notes, remove what the next stage does not need. If the safe briefing no longer contains enough to do the edit, stop rather than treating your approval as permission the underlying agreement never gave.