Four record types have their own assistant, on an AI Assistant tab inside the record:
- an event
- a location
- a program
- a performer
Each one answers about that record and nothing else. There is no assistant on contacts, messages, bookings or assets.
What each one will answer
| Record | It can answer about |
|---|---|
| Event | The event's own details — its date, status, location and capacity · The bookings attached to it, with committed value counted over confirmed and completed bookings only |
| Location | The location's own record and its spaces · Activity at that location |
| Program | The program's own details — its schedule, status, category and the targets set for it · The events and bookings within it |
| Performer | Your organization's saved relationship with them · Your organization's booking history with them |
That is the whole list for each record. Ask something outside it and you get a plain refusal that names what the assistant can answer, rather than an answer stretched to fit.
Every answer ends with a Based on… line naming exactly what it was derived from — including which bookings were counted. Two figures that sound alike often count different things, and that line is how you tell them apart.
What it will not answer
It will not tell you what to do. Ask "what should I do about this?" and it says so:
I can tell you what this record's data shows, but not what to do about it. Suggestions — with the signal and threshold behind each one — live in the Recommendations panel on your dashboard.
If an answer drifts into advice on its own, it is withheld rather than shown to you, and the same signpost appears.
It will not answer about a record you cannot see. If you do not have access to a record, the reply is the same whether the record exists or not — "That event isn't available to you." — so nothing can be learned from asking.
It will not show you a figure it cannot trace. Every figure in an answer is checked back against your records before you see it. If one does not match, the whole answer is withheld:
The answer contained a figure that doesn't match your data, so it wasn't shown. This has been logged.
Answers are drafts to check, not advice
Every assistant carries this line, always — not only when something has gone wrong:
Answers describe this record's own data and are drafts for you to check. Figures are traceable to your records, not fact-checked — and this assistant doesn't advise on what to do.
The distinction in the middle of that sentence is the important one. Traceable means the number came out of your records. It does not mean the number was attached to the right thing in the sentence. Read an answer against the record it describes.
"Not enough recorded yet"
If a record is too thin to say anything honest about, you get a refusal naming the gaps rather than a padded answer. On an event, for example:
Not enough recorded yet
There isn't enough recorded against this event yet to answer questions about it. Once it has bookings, or a capacity and expected attendance, there'll be something real to say.
Below it is a list of what is missing.
This is deliberate. A record with nothing in it can still produce a fluent-sounding paragraph, and a fluent paragraph about an empty record is worse than being told the record is empty — you would have no way to tell. The refusal is decided before anything is generated, from your data, not by the model.
Each record type has its own wording. A quiet event inside a busy organization gets a sentence about that event, never one claiming your organization has no activity.
What a view-only member sees
No ask control at all — no input box, no button. Reading generated content and producing it are separate permissions, and a member who has the first but not the second is told so plainly rather than handed a control that will refuse them.
If ARRIVE AI is not available to your organization yet, the panel says so and why.
Limits
ARRIVE AI has two limits, in two places at once:
- Per organization: a monthly AI allowance. Each organization has an allowance for the calendar month, measured in what its AI requests actually cost. Settings → AI shows about how many uses are left. It resets on the 1st of each month.
- Per member: 25 requests in any 24 hours.
These are set by ARRIVE, not by your organization. ARRIVE may adjust these limits.
Two things worth knowing about how they count:
- Questions the assistant can't answer still count. If a question is off-topic, or there isn't enough recorded yet, it still uses one of your 25. One exception, on the Ask bar only: if your organization has no events and no bookings yet, the bar gives a plain count instead — for example, "MyCity Venue has 0 locations, 2 team members and no events or bookings yet." That answer usually does not use one of your 25. A refusal uses little or none of the organization's monthly allowance.
- The organization's allowance is shared with every other ARRIVE AI feature that generates — a daily briefing or a report summary draws on the same allowance.
When the organization's allowance runs out you are told: "Your organization has used this month's AI allowance. It resets on (the date)." When you reach your own 25, you're told it frees up over the next 24 hours.
What this does not do yet
- There is no conversation. Each question is answered on its own; the assistant does not remember the previous one.
- There is no assistant on contacts, messages, bookings or assets.
- Answers cannot be saved, shared or exported from the panel.
- The assistant cannot compare two records, or answer about a record other than the one you are on.
Related: Where ARRIVE AI appears · Turning ARRIVE AI on and off · What ARRIVE AI keeps