Working with people
Keep your AI. Share the thinking.
Send someone a briefing your assistant drafts and you approve, whichever AI they use, or none, and get their reply back in your work.
Provenance
Every line says who wrote it.
When an AI drafts the message, “Pam said” stops meaning much. RelayLink labels each part of every message: words the sender wrote themselves, words their assistant drafted and they approved, and assumptions nobody actually stated.
The thread, labeled as Jim reads it
- Pam’s note Can you poke holes in this before Friday? Their own words, not their assistant's
- Her plan Five houses on one street, for a month, before anything is built. Drafted by the sender's assistant, approved word for word
- An assumption in it Neighbors will lend the tools they rarely use. Inferred by her assistant, not stated
- Jim’s reply, by email What about tools that come back broken? Typed in a reply — nobody verified who wrote it
Illustrative · the labels are set as each part is sent
The short answer
How can I tell if a message was written by a person or by AI?
Often you cannot, from the text alone. RelayLink records it instead: the note on every message says who wrote it, the sender in their own words or their assistant with the sender’s approval, and an assumption the assistant worked out is marked as inferred, not stated.
How it works
Labeled as it is sent, not guessed afterwards.
-
Your assistant drafts, you approve
A message your assistant drafted goes out labeled as drafted and approved, never as your own words.
-
Change the wording, and it is yours
Rewrite the note yourself and it is labeled as written by you. Approving the draft unchanged does not count as writing it.
-
Guesses stay guesses
An assumption your assistant worked out is marked as inferred. Only something you actually said is marked as stated.
What you would say
- “Send it, but let me write the note myself.”
- “Which parts of Pam’s message did she write herself?”
- “Is that her assumption or her assistant’s?”
- “What did Jim actually say, in his own words?”
The details
What it does, and what it doesn’t.
What it does
- Labels the note on every message with who wrote it.
- Marks words as the sender’s own only when they actually changed the draft.
- Keeps inferred assumptions apart from stated ones, even when two assistants inferred the same thing.
- Labels a reply typed from an email link as exactly that, with the sender not verified.
- Carries the labels everywhere the message goes: the other person’s assistant, their email and the web page.
What it doesn’t do
- Detect AI writing. It records how a message was made; it does not guess from the text.
- Prove who typed an edit relayed by an assistant. The assistant reports what you typed; reviewing the draft in your account page takes it out of the loop.
- Let a sender choose their own label. The label follows from how the message was made.
Questions
What people ask, answered.
Is this an AI detector?
No. A detector guesses from the text and is often wrong. RelayLink records how each part was made as it is sent, so the label is a fact about the message, not a guess about its style.
What does “their own words” mean on a message?
That the sender changed the draft and wrote the note themselves. Approving an assistant’s wording unchanged is labeled as approved, not as written.
Could an assistant claim I wrote something I didn’t?
That is the case this guards against most. Echoing the draft back never counts as your words. The gap that remains, an assistant reporting an edit you did not make, closes when you review and send from your account page, where no assistant is involved.
Why does it matter who wrote an assumption?
Because the person reading it will act on it. “Neighbors will lend” means one thing if Pam said it and another if her assistant guessed it, and the label tells them which.
Do the labels follow a message everywhere?
Yes. In the other person’s assistant, in the email they receive and on the web page, the labels come from one place in RelayLink, so they cannot disagree.
How are replies sent by email labeled?
As typed in a reply. Email does not prove who wrote something, so the label says so rather than claiming more.
Keep going
Related, and in more depth.
Create your account. Connecting is later.
Create your account now and it is ready right away — nothing to wait for, nothing to redo.