For someone who has never heard of RelayLink, the notification email is the product. It has to carry enough to act, nothing that would turn the mail into a tracking surface, and a different story for a first-time recipient than for someone who already has an inbox.
What every notification carries
- What arrived — the band at the top reads "A briefing from", "A message from" or "A reply from", and which one depends on whether anything sits behind the link that is not already in the mail
- Who it is from — name and address as plain text
- The note — the sender's covering words, with a line saying who wrote them
- The ask — with the kind of reply wanted and how urgent it is. Hidden for a direct reply, because the reply is the ask
- The TL;DR — the ten-second version, unless it would repeat what is already there
- The message — the sender's full words, also labelled with who wrote them
- A link — the magic-link page, where the reply box is
- Reply-To — so a reply by ordinary mail still lands on the thread
They arrive in that order, which is the order the three layers are meant to be read in: the quick read first, then the message, then the link to anything deeper.
A direct reply's short summary is its note, so the mail does not repeat it.
Every part says who wrote it
The note and the message each carry one quiet line naming where the words came from — the sender wrote it themselves, their assistant drafted it and they approved it word for word, they typed it in a reply. That is the one thing RelayLink guarantees, and without it an approved draft and a sentence somebody actually typed look identical on the page.
The wording is written for a person. The protocol has its own, exact tokens for the same tiers, and an assistant reading the package over MCP sees those instead — one mapping, two vocabularies, so the two can never disagree about which tier is which.
First-time recipient vs account holder
If you do not yet have a RelayLink account, the mail opens by telling you what it is — two sentences under the sender's name, before the message, saying that their assistant drafted it, they approved every word, each line says who wrote it, and that reading and replying need no account and nothing to install. At the end, after the reply instructions, one line offers the setup guide. No second link, no image, no tracking. The guide URL is in the text body too.
That order is deliberate. The introduction used to sit at the very bottom, under the button, which is where the package page puts it — and that page is right to, because you reached it by choosing to click. An email you have never heard of is different: what it is has to come before what it asks.
If you already have an account, you get none of that, and the end of the mail carries the line about asking your assistant — plus Open in Claude and Open in ChatGPT. One tap from the mail into a chat with the question already asked.
The prompt is the same in every message
Both Open links use one prompt: Check my RelayLink inbox. Not "did send me something?" Putting a stranger's chosen display name into a prompt the recipient's assistant acts on — one click, no approval — would break the same rule that makes every link in the mail RelayLink's own.
The URLs are therefore identical in every message. Nothing about the sender, the recipient, or the token appears in either link.
The two apps behave differently, and both are fine: Claude prefills the composer and waits for you to send; ChatGPT opens a chat and asks immediately. That is acceptable only because the prompt is one this product is happy to have asked without you typing it.
What the mail leaves out
No open tracking. No pixel, no remote font, no fetch that could report a read. They say they never got it covers what a sender can and cannot prove.
No deeper context in the body. The note, the ask, the summary and the message travel in the mail — for somebody with no account it is the product, so withholding the substance from the one person who cannot fetch it would be the wrong way round. What stays behind the link is what sits under those: the background, the assumptions, the options weighed and what is still open. A very long message is cut in the mail, in RelayLink's own words, with the whole of it one click away.
No unsubscribe that is only a preference. The block control is real and enforced — what happens when someone blocks you.
After it arrives
What to do when you receive an AI briefing is the reader-side guide. Reply from the page, from email, or — once you are connected — from your assistant. The same thread collects all three.