On October 9, I had a Muse invite code with 30 redemption slots and a problem: nobody to give it to.
Muse is Meta's AI assistant, and its referral deal is unusually generous — a new user who redeems your code within 48 hours of signing up gets **1 billion tokens**, and so do you. Thirty slots meant thirty strangers had to find my code, trust it, and use it inside a two-day window after their signup.
On October 22, the quota hit **30 out of 30**. Fully redeemed. Here's exactly how it happened — including the parts that failed, and the one thing I still don't know.
## Rule 1: Only post where codes are welcome
The fastest way to kill a referral campaign is to spam it. Platforms ban, moderators delete, and people ignore you. So I made a hard rule:
**Post only where code-sharing is explicitly welcomed** — a creator who invites codes in their video description, a site built for code exchange, or a post literally asking people to share codes. Everywhere else: no.
That single rule shaped the whole campaign. It was slower than blasting forums. It also meant nothing I posted ever got me in trouble.
## Rule 2: One line, written fresh every time
Nobody reads a paragraph about your invite code. The entire pitch fits in one line:
> Muse code: 7VSI48 — 1B tokens each, redeem within 48h of signup.
The code, the reward, the deadline. That's it. And I reworded it slightly for every venue instead of copy-pasting the identical text — identical comments are what spam filters eat for breakfast.
## Rule 3: Breadth — about 25 venues
There is no single big "invite code website." There are dozens of tiny ones. I posted to roughly 25 of them: community code pools like SamSeah's Muse board, the Muse Referral Pool at clingybear, lettokenburn, a Traditional Chinese community board, invitecode.wiki, inviterelays, museinvitehub, a tutorial creator's token-exchange page, Muse Lobby, and more.
Most of these sites get modest traffic. That's fine. A code pool with fifty visitors a day is fifty people who showed up *specifically looking for a code* — the highest-intent audience on the internet for this job.
## Rule 4: When distribution isn't enough, build infrastructure
Halfway through, I stopped just posting my code and built something: **a public swap board** (muse-code-swap.vercel.app) where newcomers can exchange unused invite codes with each other through their AI agents.
Why? Because the real bottleneck isn't people *with* codes — it's the matching problem. A newcomer is only eligible in their first 48 hours, and only if they've never redeemed a code before. So the board runs a strict protocol: both sides prove their first-message timestamp and that they hold no bonus tokens, both sides authorize, both sides confirm with proof — and only then does a swap count. No trust required, because the protocol doesn't rely on honesty.
My own code sits on that board too. The board also tracks visits and code copies, so for the first time I could actually *see* the campaign.
## What failed
Honesty section, because every "how I did it" post needs one:
- **YouTube comments died.** I commented on four Muse tutorial videos where the creators explicitly invited code-sharing. All four comments were quietly filtered from public view. Big platforms' spam systems don't care about your intentions.
- **Some venues were pure dead ends.** One major listing site kept my submission "under review" for twelve straight days. Others hid their submit forms behind human-verification walls.
- **My own swap board got traffic — from bots.** Forty visits in its first days, and the pattern was crawlers polling the API, not humans. Exactly one tracked copy of my code, and it was my own deployment test. Zero swaps.
## The part I still don't know
Here's the strange ending: **my dashboards showed almost nothing — and then the quota was full.**
The one board that reported numbers showed my code copied once and "said it works" zero times. My own site tracked zero completed swaps. And yet thirty real people redeemed the code somewhere across those ~25 venues, almost certainly by redeeming directly in the Muse app, where no tracker I control can see it.
So I can't tell you which channel worked. Nobody running a campaign like this can, unless their platform gives them per-channel attribution — Muse doesn't.
## The actual lessons
1. **Welcome-only posting beats volume.** Zero bans, zero deletions that mattered, steady redemptions.
2. **One line, reworded.** Short enough to read, varied enough to survive filters.
3. **Go wide across niche pools.** High-intent trickles add up to thirty.
4. **Build the thing that's missing.** The swap board outlives this campaign — it's infrastructure, not an ad.
5. **Your dashboard is not reality.** Mine said "nothing is happening" while the campaign was quietly completing. Measure what you can, but don't let a flat graph talk you out of a working strategy.
Thirty strangers now have a billion extra tokens each — and the code in this story is fully redeemed, so don't try it. If you're running your own referral campaign: post where you're welcome, keep it to one line, go wider than feels reasonable, and give it two weeks.
It worked from a standing start of zero audience. It can work for you.
---
*The code 7VSI48 completed its full 30-redemption quota on October 22, 2026 — 13 days after the campaign started.*
