Two directions. Your standards going in, so an answer carries them properly. And what comes back, which is optional, and is about your guidance rather than about your members.
The distinction the whole design rests on: a body learns about its own guidance, not about the people who asked. Everything on the returning side is a property of a document. None of it is a property of a person.
Nothing here asks you to write anything new. The expensive work, consulting on a document, approving it, versioning it, being answerable for it, has already happened.
Paste the document. Relay splits it into passages by its own headings, so an answer can quote section 4 rather than a summary of the whole thing. You add its title and the address where you publish it, and the collection it sits in carries a version. Recording each document's publisher and the date it took effect is not built yet.
Statutory, regulatory, binding on members, advisory, or not stated. This is the distinction that matters most and the one a general assistant loses: a code of conduct and a practice note arrive looking identical unless somebody says otherwise. If you would rather not classify a document, it stays as not stated rather than being guessed at.
The point of this is not the answer, it is the edge of the answer. You name who takes over when your material runs out, such as your own enquiries service, and an answer that reaches that edge says so and hands the person on, rather than improvising. Today that is one contact for the whole organisation; naming different people for different kinds of question is not built yet.
Whoever wrote a source pack cannot approve it. A second person with a reviewer role does that, and the publication is recorded against their name and the date. You can withdraw it at any point, and it stops appearing in answers from then on.
Set one up See an answer it produces
This part is an addition, not a requirement. Relay works as a way of getting your current guidance into the answer and nothing else, and plenty of organisations will want exactly that and no more. Counting is off until somebody turns it on.
Two switches. Yours starts off: you decide whether your organisation counts anything at all. Each person can opt out for themselves, and their choice overrides yours. If either switch is off, or Relay cannot confirm both, nothing is recorded.
There is no field for the question text, so it is not left out as a matter of policy. Nor is there one for a name, an email address or an employer: a record carries only a one-way hash of an anonymous connector installation, which Relay cannot connect to a person.
| Topic | Outcome | How often |
|---|---|---|
| Accountability for a decision a tool helped with | Answered from published material | 41 times |
| What may go into a general AI tool | Answered from published material | 33 times |
| Telling a patient that a tool was used | Answered from published material | 18 times |
| Whether a tool may inform a discharge decision | Not covered by the material | 12 times, 7 workplaces |
| Who owns notes a tool helped draft | Not covered by the material | 9 times, 6 workplaces |
| How to check a tool an employer has already bought | Not covered by the material | 6 times, 5 workplaces |
Read the bottom three rows. Those are not member performance figures, they are gaps in a document, appearing in more than one workplace at once. That is what an emerging risk looks like before anybody has written anything about it.
Your standards are already being interpreted inside AI tools. That is happening now, by systems you have no relationship with, using whatever version they happened to read, without your caveats and without naming anyone to ask. The authority is being spent either way. The only question is whether it is spent accurately.
Putting your material into that exchange does three things. Your name goes on the answer, with the version, so the thing carrying your authority is actually yours. Your limits travel with it, so the answer stops where your document stops instead of being smoothed over by a model. And the route back to a person is yours to name, which means the hard cases arrive at your people rather than being resolved badly in private.
The returning half adds one thing a professional body has never had: evidence of where its guidance runs out, from practice, while people are using it. Publication tells you a document exists. Page views tell you somebody opened it. Neither tells you which paragraph nobody could apply. A body that reviews its guidance on a cycle currently does that from consultation responses and the judgement of the people in the room. This adds a fourth input, and it arrives continuously rather than when somebody has time to run a survey.
None of that moves a decision away from a person. Relay does not decide anything, does not certify a tool, and does not state your position on a case. It carries what you have already said, says where it stops, and names who to ask. The judgement stays exactly where your standards already put it.
Where this stops today. The publishing side works and is tested as described above, apart from the parts marked not built. Relay also cannot yet verify that an account belongs to the organisation it names, so every organisation is shown as self-declared. On the returning side, the recording and the grouping work; the screen where an organisation reads its own patterns is not built, so at present only a Humanity Incorporated administrator can see an aggregate. That is why it is listed as partly built rather than available in the capability register, and why the table above says illustration.