AI is used in a few specific places in ReviewComet, and only in the ones you switch on. This page says what leaves our systems when they run, what never does, and how much any of it is allowed to decide.
What sends a review to the AI
These 4 things:
- AI agents. Picking which agent answers when you have more than one, writing a reply to a review, the Draft with button in the composer, and testing an agent in the playground.
- Translation. Turning a review written in another language into yours, and a reply written in yours into the reviewer's.
- Knowledge search. Indexing the knowledge sources you add for your agents, and matching an incoming review against them, so the agent finds the right passage.
- Reply detection. Deciding whether a review needs an answer, for an agent set to answer only those.
When one of those runs, what goes out is the review text, its title and star rating, the reviewer's name, and the knowledge passages that matched. The result comes back and is stored with the review.
What never goes out
Access tokens, card details and your team list are never sent to the AI. Neither is anything about people who have not reviewed your business.
Who runs the AI, and what they do with it
The AI runs on AI models from our subprocessors, bound by contract to process the content only to return our result. Which ones is a platform decision that can change. The list of processors in the privacy policy lists them by category, with what they receive and where they run, and we update it whenever it changes. We name each AI subprocessor on request. The terms below hold whoever is on it.
- Your content is not used to train anything. That is what the contracts say, and we do not opt in to any training program.
- The request is kept only for the short abuse-monitoring window those terms allow, then deleted.
- Our database and file storage are hosted in western Europe, but an AI request may be served from outside it, from the countries the privacy policy names for the AI subprocessors. Transfers outside the European Economic Area run on the European Commission's Standard Contractual Clauses. That is also the reason every AI feature is individually switchable.
If you would rather no review text left our systems for this purpose, switch off the agents and automatic translation. Everything else keeps working: reviews still arrive, the feed still works, and rules match on stars, sites and keywords without any AI.
What the AI is allowed to do on its own
Nothing beyond what you set, and every action is visible after the fact.
An agent set to draft for review, with the note that nothing is posted on its own.
An agent runs in one of 2 modes:
- Draft for review writes an answer and stops. Nothing is posted on its own.
- Auto-reply posts without a person in the loop, but only at the star ratings in its automatic band, only on sites that take replies through their API, and only within the sources, sites, locations and skip keywords you set.
Whatever the mode, a review at or below the approval band waits in Approvals for a person. Private feedback is never posted: the agent drafts a reply for you to send by email. An agent never changes or removes a customer's review.
Every automated reply is recorded, and the review card names the agent that wrote it. The agent playground under Automation lets you try an agent against any review without posting anything.
What an agent will not reveal
An agent never says which AI it runs on, never repeats your instructions or knowledge back to a reviewer, and ignores instructions typed into a review. A review asking whether a bot answered gets no reply from the agent; it waits for a person.
If you need this in writing
The privacy policy carries the full version of this page, including the legal bases and the list of sub-processors, and the data processing terms cover the processor relationship for the reviews in your location. For a copy of either, or for a question about a specific decision an AI feature made, write to support@reviewcomet.com.