One agent answers everything that reaches it, and there is nothing to decide. The moment a location has 2, something has to pick which one a review goes to, and that is what this page covers.
What you need: at least 1 agent. The routing only becomes visible once a second one exists.
Why more than one
Most businesses do well with 1 agent. A second one earns its place when 2 kinds of review need a different voice:
- a warm agent for praise, and a careful one for complaints;
- one agent per brand, where a location covers 2 brands;
- an agent that knows your delivery policy for reviews that mention delivery.
If the only difference is the star rating, you do not need 2 agents. The rating bands already treat a 1-star review differently from a 5-star one. See Rating bands.
What each agent answers
Every agent has a What this agent answers block in its Reach section.
- Kinds of review. Pick from pricing, support, shipping, complaints, praise, other questions and buying signals, or leave it on Everything, the default.
- Topics. Your own words: a product, a dish, a branch name. A review that mentions one comes to this agent before any kind or catch-all is considered.
- Use as fallback. One agent per location can hold this. It takes anything no other agent covers.
The What this agent answers section showing the kinds of review chips, the topics field and the fallback switch.
The Sites, Locations and Assigned sources lists come first: an agent never answers outside them. An agent pointed at specific sources wins a tie over one that covers the whole location.
The answering order
The Agents list is the order your agents answer in. The card at the top answers first, and each card carries its place: Answers 1st, Answers 2nd, and so on. Drag a card by its grip, or use the up and down buttons on the card, to change it.
The order only decides between agents that matched equally narrowly. A topic match still beats a declared kind, and a declared kind still beats the catch-all, whatever the order says. An account-wide agent from an agency answers after the agents the location owns.
The order a review is matched in
- A topic match wins first.
- A declared kind is next. Complaints beats Everything, wherever the 2 sit in the list.
- Everything is the catch-all, and the list order decides between 2 agents that are both on it.
- The fallback agent takes it only when nothing above matched.
- Nobody answers when nothing matched and no fallback is set. The review stays in the feed like any other unanswered review.
See which agent would answer, before it runs
Once a location has 2 or more enabled agents, a Which agent would answer? panel appears next to the playground. Type or paste a review, press Check the routing, and it names the agent that would take it, the reason, and every other candidate with why it did not.
The Which agent would answer panel with a sample review, the chosen agent and the reasoning underneath.
Nothing is posted and no reply goes anywhere.
What this costs
Setting kinds and topics, ordering the list and switching the fallback are free.
Working out which kind a review is can cost a small number of credits, and only when it has to. With 1 agent, or when every candidate is on Everything or narrows by topic alone, nothing is read and nothing is charged. Otherwise 1 reading runs per review, the first time it arrives. The reply itself costs whatever an agent reply always costs.