Data Pledge

1. The short version

A call to one of our customers is answered by an AI receptionist built on a small set of specialized providers - one that understands what a caller says and drafts a reply, and one on each side of speech (turning a caller's voice into text, and turning a reply back into speech). Their whole job is to answer that one call. None of them keep what they process in order to get smarter from it, what we keep afterward is kept for a reason and on a schedule you can check, and only a limited, named set of people can ever read it back. The three sections below expand each part of that sentence.

2. No training on your calls

We do not train any AI model on your calls, your callers' words, or your knowledge base, and the providers that understand, answer, and voice your calls do not train their general-purpose models on them either.

• The AI model. The model that reads a call and drafts the receptionist's reply runs on a commercial, pay-as-you-go API, not the free consumer chat product the same company also sells. That distinction matters because it is the API path, specifically, whose provider currently states that content submitted through it is not used to train their models unless a customer separately opts in to a named research program - we have not, and nothing about how we call it opts us in by accident.
• Turning speech into text, and text into speech. Live audio on the phone system that answers the overwhelming majority of calls is handled by our telephony carrier's own, direct integration with each speech provider - not a separate account of ours - and that carrier's own published documentation for the product states plainly that the underlying model is not trained on customer data. We also keep a second, backup phone system that takes over on the rare occasion the primary one is unreachable, and a small, separate account of our own that renders sample voices when you preview one in your dashboard; we now opt that account out of training too, in its own settings, matching the primary path above. As with any such setting, it covers what is sent through the account from the day it was turned on, not before.
• Searching your knowledge base. Answering "what are your hours" by finding the right sentence in what you told us to say requires converting a caller's question into something searchable. The provider behind that step publishes the same rule as the AI model above: nothing submitted through its business API trains its models by default.
• What this pledge does not cover. Phone carriers and payment processors keep the right, under their own standard terms, to use the traffic on their networks to detect spam, fraud, and abuse, and that can include training the detection tools they use for it. That is how robocall and card-fraud screening work on those networks, and it is separate from the AI models that understand and answer your callers.

None of this is a special deal we negotiated - it is each provider's own standing policy for the business API products we use, verified rather than assumed. A full, named list of the providers behind the Service is available on request; see privacy@coveredline.com.

3. What we keep, and for how long

• Call history and recordings are deleted automatically on a nightly schedule once they pass a window tied to your plan - see the retention row on our pricing page for the current numbers. Recording itself is off unless you turn it on.
• Messages, bookings, leads, imported contacts, and chat conversations are kept so your history and reports stay complete. They are not deleted on a timer; you can remove some of them yourself from your dashboard, and you can ask us to delete any of them.
• Leaving. Cancelling stops the receptionist from answering; it does not by itself erase what we hold, so you can pick back up later. Email us to have your account's data deleted and we will delete or de-identify it within 30 days of confirming the request, 45 days at the outside, short of what we must keep for legal, security, or billing reasons. Backup copies roll off on our providers' own schedules after that.

4. Who can actually read a transcript

• Every business's data is isolated at the database row level - not by convention, by how the database itself is configured - so one customer's calls are not reachable through another's account, ever.
• Inside your own account, access follows the role you gave someone. Whoever you invite as a team member can read a call to help you run the business; nobody you have not invited can see into your account at all.
• Our own staff do not read call content in the ordinary course. On the rare occasion someone here needs to, it is to help with a support request you raised, to debug a problem, or to look into misuse - and for nothing else.
• The AI model reads a call live, to answer it - that is the one thing it is for - and, per the section above, does not keep what it read in order to learn from it afterward.
• A webhook, a connected automation, or an exported report never carries the transcript itself, however you have those set up - by design, those surfaces carry the outcome of a call (booked, a message taken, and so on), never the words of it.

If you are a caller rather than a business we serve and you have a request about your own information, please contact the business you called - we act on their behalf, and they can reach us to act on it too. See Privacy Policy for the full version of this.

5. Where this fits

This page is a plain-language summary, not the contract - Privacy Policy and Terms of Service are what actually governs, and either one controls if the two ever read differently. Some businesses need to know more than this page promises today before they can sign up with us in good conscience; if that is you, write to privacy@coveredline.com before you do; we would rather tell you plainly where we stand than have you find out after.

Questions about anything on this page: privacy@coveredline.com.