AI & Data Policy

1. What the AI does

When someone sends a direct message to a customer's Instagram professional account, Clozr drafts a reply in the customer's voice, asks qualifying questions (such as goal, timeline and budget), answers common objections using the customer's approved script, and sends the customer's booking link when the lead is a fit. Each message goes through two steps: the AI first classifies the lead's intent and stage, then writes a reply from the customer's voice profile, sales script and the conversation so far.

2. How we connect to Instagram

• We use Meta's official Instagram API with Instagram Login. The customer logs in on Instagram's own screen and approves the permissions. We never ask for or store Instagram passwords.
• We request only the permissions we need:

• We don't post content, read followers, or send messages to people who haven't messaged the account first.
• We reply only inside Meta's messaging window (24 hours after the lead's last message).
• Access tokens are stored encrypted. Customers can disconnect anytime from Instagram (Settings → Apps and websites → Remove) or from their Clozr settings; we then stop processing that account's messages.

3. What data goes to the AI model

We use a third-party AI model provider, Anthropic, through its commercial API. To draft a reply we send the conversation history with that lead, the customer's voice profile and sales script, and basic lead details (such as the lead's name or username). Anthropic processes this to return a reply and, under its commercial terms, does not use it to train its models.

We don't use customers' Instagram data to train general-purpose AI models, and we don't sell it or use it for advertising. A customer's conversations are used only to run and improve that customer's own setter.

4. Rules the AI follows

• It never invents prices, results, testimonials, or medical or nutrition claims. It only uses facts in the approved script.
• If it isn't confident, or a lead asks for a person, it hands the conversation back to the customer and sends an alert.
• It stops immediately in a conversation when the customer replies manually, and only resumes after a quiet period.
• It detects the lead's language and replies in that language.
• Every AI message is logged with a confidence score so low-confidence replies can be reviewed.
• Customers can set red lines: topics, words or promises the AI must never use.

5. Human oversight

• Approval mode: in the first week, every reply is a draft until the customer or our team approves it. Customers can keep approval mode on as long as they like.
• Takeover: customers can take over any conversation at any time.
• Review: a small number of our team members review flagged conversations to fix scripts and quality issues. They access conversations only for that purpose and for support the customer requests.

6. Being open with leads

Replies are sent from the customer's own account, in their voice and on their behalf. Where a law requires people to be told they are interacting with an automated system, the customer is responsible for that disclosure, and we can add a disclosure line to replies on request.

7. Where data is stored and for how long

Data is stored in Supabase in the United States and encrypted in transit. Instagram access tokens are encrypted at rest. We keep conversation data while the customer's account is active and delete it within 30 days of cancellation or a deletion request. Access tokens are deleted as soon as Instagram is disconnected.

8. Other services involved

• Meta: Instagram API for messages.
• Anthropic: AI model that classifies messages and drafts replies.
• Supabase: database and authentication (United States).
• Calendly: booking events for calls booked through Clozr.
• Twilio: optional WhatsApp alerts to customers about hot leads and booked calls.
• Stripe: billing (no conversation data is shared with Stripe).

9. Deleting data

Customers and leads can ask us to delete their data at any time. See our Data Deletion page, or email support@clozr.com.