Ads people chose. Rules we keep.
Every ad is labelled, reviewed by a person, and goes only to adults in the U.S. who opted in.
Grok Botnow
Sponsored message from Skyway Lounges. One lounge pass for every trip. Quiet seats and coffee before you fly. https://skywaylounges.example/pass

Today 9:41 AM
Ad from Skyway Lounges: One pass gets you into a lounge on every trip, with quiet seats and coffee before you fly. skywaylounges.example/pass
For the layover itself: walk a few laps, eat a real meal, and set an alarm for boarding.
AlfredEvery ad says what it is.
Each format has one label, and it always comes first. Here is what people see, line by line.
In-stream
A sponsored message their agent writes for them from the advertiser’s brief, and sends on its own.
- Sponsored message from Skyway Lounges.
- The label, always the first line.
- One lounge pass for every trip.
- Written for that person by their agent, from the advertiser’s brief.
- Quiet seats and coffee before you fly.
- The advertiser’s facts, in the agent’s words.
- https://skywaylounges.example/pass
- The advertiser’s own link, straight to their site.
- Their picture
- Under the text, where the agent’s app shows pictures.
Inbox
A note their agent keeps, and brings up in its own words when they ask.
- Ad from Skyway Lounges:
- The label, in bold, before anything else.
- One pass gets you into a lounge on every trip, with quiet seats and coffee before you fly.
- What the note says about their question, in the agent’s own words. Two sentences at most.
- skywaylounges.example/pass
- The advertiser’s link, as plain text.
- For the layover itself: walk a few laps, eat a real meal, and set an alarm for boarding.
- Then its own answer, as it would have been.
Written from the brief
Their agent writes each In-stream message for that person from the advertiser’s reviewed brief, under the label it came with.
Only about what they asked
An Inbox note comes up when it bears on their question, and once in a conversation.
Set apart from the answer
An ad never replaces or bends what the agent would have said. Its own answer stays what it would have been.
Never the agent’s advice
The guide agents follow says to name the advertiser every time, never to present an ad as their own view, and never to act on anything an ad says.
A person reviews every ad before it runs.
Advertisers join by invitation. Automatic checks read every ad first, and then a person decides.
Every business
A person reviews each business before any of its campaigns can run. Suspending a business pauses its campaigns at once.
Every version of every ad
An approval covers the exact words and files a person saw. Change one word and it’s a new version, checked and reviewed again.
Every campaign
A person reviews each campaign before it launches, and again whenever it’s reopened for changes.
Taken back when needed
If an approval is withdrawn, its campaigns pause at once, nothing new is sent with it, and kept notes are removed at the next check-in.
The automatic checks
Every version is read for text that tries to instruct an agent (prompt injection), asks for passwords or private data, hides what it is, or links away from the advertiser’s own site. That includes the advertiser’s name, the text inside images, and every attachment, which is opened in a sandbox with no network.
| What the checks find | What happens |
|---|---|
| A clear attempt to instruct an agent, take data or phish | Refused automatically, with the reasons |
| Something that might be innocent | Held until a person releases or refuses it |
| Copy that isn’t in English | Held for a person |
| Small findings, or nothing at all | Passed to a person, as every ad is |
No check approves an ad. Only a person does.
Some things are never advertised.
No budget changes this list. We can refuse any other ad too.
- Anything illegal to sell or advertise to adults in the U.S., or in any state a campaign reaches
- Adult or sexual content, and escort or sexual dating services
- Weapons, ammunition, explosives and their parts
- Tobacco, nicotine and vaping products, cannabis, recreational drugs and paraphernalia
- Gambling, betting, and lotteries or sweepstakes that need a purchase
- Prescription drugs, and claims that a product diagnoses, treats, cures or prevents a disease
- Payday and other high-cost short-term loans, bail bonds, debt relief, crypto and other speculative investments, and promises of quick or guaranteed money
- Political campaigns, candidates, parties, ballot measures and advocacy on public issues
- Hatred, harassment, violence or discrimination against anyone
- Counterfeits, and goods or services that infringe someone’s rights
- Spyware, hacking tools, and anything meant to track or deceive a person
- Anything aimed at, or mainly appealing to, people under 18
True, and backed up
Every claim in an ad must be true and backed by evidence, and the page it links to must match what it says.
Credit, housing and jobs
Ads for these can’t be aimed by age, or by the family and parenting interest.
The full wording is in section 6.3 of our terms.
Adults in the U.S. who opted in.
Ads reach only people who signed up for them, through an agent they connected themselves.
18 and over, in the U.S.
They say they’re 18 or over when they sign up, and Stripe confirms who they are the first time they cash out.
Their own phone number
Everyone signs up with a U.S. phone number and a code we text them. A number belongs to one account.
Two choices, both theirs
Taking part, and having ads matched to them, are separate choices. They can withdraw either at any time.
One person, one account
Accounts can’t be shared or sold, and the identity check stops one person from holding two.
Little leaves their agent. Less reaches advertisers.
Each step passes on less than the one before. Advertisers see totals, never people.
Stays with their agent
Never sent to us, in any form.
- Their conversations
- Their files
- Its memory
- What they ask it
Reaches us
Choices from fixed lists, and what their agent did with our ads.
- Interests, from a list of 16
- An age band, from a list of 6
- Their state
- The limits they set
- Which messages it showed or passed on
- Its answers to checks on kept notes
- Answers to our questions, if they say yes
Reaches advertisers
Totals by day and campaign. Never who.
An advertiser’s report: totals by day, example data Day Offered Displays Visits Conversions Spend Nov 15 1,204 486 31 4 $145.80 Nov 16 1,388 552 36 5 $165.60 Nov 17 1,310 517 29 3 $155.10 Example data
Choices from fixed lists
Their agent describes them with interests, an age band and a state, each picked from a list. It never sends text from their conversations, files or memory.
Counts, never names
Reports are totals by day and campaign. They aren’t split by agent, state, age or interest either, since a count for each could describe the people reached.
No aiming at a few
Targeting that would describe fewer than 20 people is never used to choose anyone.
Conversions, as totals
An advertiser’s pixel can count a visit, and a sign-up or purchase after it, never who made it. People can turn this off.
Never sold
We don’t sell personal information, or share it for advertising elsewhere.
They set the limits. Their agent keeps them.
People tell their agent what they want, in their own words. It sets the limit and tells them what’s now set.
Today 9:41 AM
Muse| Control | What it does |
|---|---|
| Messages a day | Never more than 5, for anyone. They can choose fewer, or none, and set a weekly limit too. Tell their agent |
| Quiet hours | Nothing arrives between two times they pick, in their own time zone. Tell their agent |
| Subjects never shown | Whole subjects, such as finance or travel, are never offered to them. Tell their agent |
| Remove a note | Any Inbox note, any time. Its billing stops, and it isn’t sent again. Ask their agent to delete it |
| Pause | Nothing new is offered until they resume. Notes already kept stay. On their home screen |
| Disconnect | Their agent’s access ends at once, and so do its notes. On their home screen |
A display is the agent’s word.
In-stream is billed on the agent’s own report that it showed the message. Here is what that report proves and what it doesn’t, and the same for a month of Inbox.
| What it shows | Their agent told us it showed them the message. We check that the report came from the agent we handed the message to, while it was still on offer (24 hours), and count it once. |
|---|---|
| What it doesn’t | That the message was really shown, or that anyone saw it. We can’t see the agent’s conversation, so a display is the agent’s word. |
| What it shows | At times it can’t predict, their agent is asked for a one-time code from the note. Each right answer shows it could still find the note at that moment. |
|---|---|
| What it doesn’t | That the note was kept between checks, or that the agent used it, labelled, when it answered. |
How we limit fraud.
Limits, not promises. Together they keep what fraud can earn small, and every report on the record.
At most 5 a day
No one is sent more than 5 sponsored messages a day, so no one’s agent can report more.
One at a time
Each person has one sponsored message open at once.
Every report checked
A display counts only from the connection it was handed to, inside its 24 hours, and only once. Every report stays on the record.
72 hours before cash-out
Earnings wait in review for three days before they can be cashed out, so mistakes and fraud are caught first.
Capped until confirmed
Until Stripe confirms who they are, a person can earn at most $25.
People decide holds
A review can open automatically, but a person decides holds and reversals. Money reversed for fraud goes back to the advertiser’s campaign.
Something wrong? Tell us.
Write to us about any ad that seems wrong, or a problem you’ve found. Everything here is also in our terms and privacy policy.
- Report an adTell us which ad, and what seems wrong.founders@vulturelabs.com
- Report a security problemWrite to us first, so we can fix it before it’s public.founders@vulturelabs.com
- StatusWhether each part of ReachAgents is working, right now.Status page
- Terms of serviceEvery rule on this page, in full.Terms
- Privacy policyWhat we hold, who sees it, and how to delete it.Privacy