You're Talking to Software

RuleInside builds AI voice and chat agents that answer and place calls for online stores. This page explains what that means for the people who talk to them.

You are speaking to an AI

If you would rather not speak to an AI, end the call at any time; you are never obliged to continue, and ending it costs you nothing. You can also ask to be put through to a person. The agent will not claim to be human if you ask it directly.

Messages on WhatsApp

A message you receive from a store on WhatsApp may have been written by an AI agent, by a person, or chosen from a pre-approved template. Which one depends on how the store has set itself up:

• An AI agent may reply to you in the conversation. It is set up to tell you it is an AI, and it will not claim to be human if you ask it directly. Depending on the store’s settings it may reply with a voice note, which is synthesised speech, not a recording of a person.
• A store may send you a pre-approved template - a fixed message, written by the store and approved in advance, such as an order or delivery update. No AI writes these.
• A member of the store’s staff may write to you directly from their own WhatsApp Business app. We receive a copy so the store has one history, but a person wrote it.

If you are unsure, ask whether you are talking to an AI.

You can stop at any time. Stopping costs you nothing and does not affect any order you have already placed. There are three ways:

• Reply STOP. Where the store’s WhatsApp number is connected to RuleInside directly, a message that says only STOP stops the store’s WhatsApp messages and its calls to you through us. Reply START later to receive its WhatsApp messages again; calls stay stopped. You may receive one short message confirming each.
• Tell the store directly, by phone or email, that you do not want further messages. A message typed into the chat is not treated as a request to stop unless it says only STOP.
• Write to us at info (at) ruleinside.com and we will record it against your number, so the store cannot message or call you again through us.

If you are not sure how the store is connected, write to us: that always works.

Not every reply is read by a person. Where an AI agent handles the store’s messages, the agent reads and answers your reply. Where the store answers from its own WhatsApp Business app, its staff see your reply there. Otherwise your reply is stored (see below) but may not be read. If you need an answer, contact the store directly.

How long we keep the text of your message. We keep the text of WhatsApp messages for 180 days and then delete it automatically. After that we keep only the record that a message happened: who, when, which direction and what type. This is the same however the store has connected WhatsApp. Where the store’s messaging runs through our voice-agent provider, ElevenLabs also holds the conversation, for the period in section 6 of our Privacy Policy. Message text is not used to train AI models.

WhatsApp itself is operated by Meta, under its own terms and privacy notice.

Calls are recorded

Calls handled by our agents are recorded. The agent tells you so in its opening words or, where the store has chosen it, is instructed to tell you in its second or third reply. In that second case, if you object to the recording, the agent is instructed to end the call. If you tell the agent that you do not want to be recorded, we delete our recording and transcript of that call, normally within two days. A store that keeps its own copy of its calls is responsible for that copy. To have the recording of any other call deleted, ask the store you spoke to.

Recordings and transcripts are used to operate the service, to handle complaints, and for quality. They are not used to train AI models. How long they are kept, and how to ask for them to be deleted, is in our Privacy Policy.

What the law requires, and who owes it

The EU AI Act (Regulation (EU) 2024/1689) draws a line between the provider of an AI system (whoever develops it and puts it into service under their own name) and the deployer, who uses it. Both apply here, and the duties differ.

RuleInside is the provider. Under Article 50(1) we must design the system so that people are informed they are interacting with an AI. Under Article 50(5) that information must be given "in a clear and distinguishable manner at the latest at the time of the first interaction", and must meet applicable accessibility requirements. That is why the disclosure is in the agent's opening words rather than buried on a policy page.

There is an exemption where the situation is "obvious" to a reasonably well-informed person. We do not rely on it. The European Commission says it should be read restrictively, and someone who dialled a business telephone number has no reason to assume software will answer.

The store using the agent is the deployer. It has its own obligations, including telling callers about recording where local law requires it, and complying with the rules on marketing and unsolicited calls where the people it contacts are located. That obligation is written into our Terms of Service.

Marking of AI-generated audio

Article 50(2) of the AI Act requires providers of systems that generate synthetic audio to ensure the output is marked in a machine-readable format and detectable as artificially generated.

The speech in our agents is generated by ElevenLabs, who embed an inaudible watermark (SynthID) in the audio they generate. Where our audio reaches you over a channel that preserves it, that marking is present.

Automated decisions

The agent can book appointments, look up orders, and record what you asked for. It does not make final decisions about you.

Refunds are never processed automatically: a person at the store approves them. Contracts, prices and anything else that binds you are decided by a person. If a decision matters to you, ask for a human to handle it.

What the agent can get wrong, and what we do about it

The agent is software and it can be wrong. We would rather tell you how than leave you to find out.

Known limitations

• It can produce an answer that sounds confident and is not correct, the failure usually called a "hallucination".
• It can mishear, particularly on a poor line, with background noise, or with an unfamiliar accent.
• It can miss irony, sarcasm or an indirect answer, and take it literally.
• It can misjudge what you want and take the conversation in the wrong direction.
• It does not give legal, financial or medical advice, and should not be asked for it.

What limits the damage

• Nothing that costs you money happens automatically. Refunds are never processed by the agent; a person at the store approves every one.
• It will not read out your personal or order details on the strength of your phone number. Caller ID can be faked, so it is never accepted as proof of who you are. Without verified identification you get general information only.
• You can ask for a person at any point, and the agent will arrange it rather than talk you out of it.
• Outbound contact fails towards silence, as described below.

If something the agent tells you matters (a price, a delivery date, a policy), check it with the store before relying on it.

If we call you

Some stores use our agents to call people and to send WhatsApp messages, as well as to answer them.

Before a call, the system checks four things independently: whether you have opted out, whether you are on a do-not-call list, whether consent was recorded for you specifically, and whether it is within the calling hours where you are. The store sets those hours; unless it changes them, they are 8:00 to 21:00 your local time. A cart reminder call, which a store places only if you ticked the call box at its checkout, is never outside 8:00 to 21:00 in the time zone of your phone number (or the store’s, where the number does not show one), whatever the store’s hours, and it opens in English.

Before a WhatsApp message, the system always checks the first three: opt-out, do-not-call and consent. Whether it also waits for your calling hours depends on the message:

• Marketing sent automatically, such as a campaign or an automatic follow-up, waits for your calling hours. If we do not know your time zone, it waits for the store to add it, and if that has not happened after 7 days, the message is not sent.
• Marketing a member of the store’s staff sends to you personally, by hand, is sent when they send it.
• A code or offer you asked for in a WhatsApp chat is sent straight away if it is within 24 hours of your request. Later than that, it waits for your calling hours.
• Messages that are not marketing, such as an order or delivery update, are not held for the calling hours.
• After you write STOP or START, you may receive one short message confirming it. It is the only message we start without these checks, because it confirms your own choice.

If a check fails, or cannot be completed, you are not contacted. The system is built to fail towards not contacting you, and an error in the checks produces silence, or a delay, rather than a call or a message.

If you tell an agent you do not want to be contacted again, that is recorded and honoured. We keep that record even if you later ask us to delete your data, because deleting it would destroy the proof that you asked, and you could be contacted again. That is the one thing we deliberately keep.

The store that instructed the call is responsible for having a lawful basis for it, and for complying with the marketing and calling rules where you are.

Questions

Write to info (at) ruleinside.com.