Connecting an AI assistant to your email takes two clicks. Understanding what you just agreed to takes longer, and most people skip it. These are the questions we would ask of any vendor. We have answered each one for Pranan, and linked the page where you can check.

1. Can it send email as me?

This is the biggest line. A tool that only drafts can embarrass you in private. A tool that sends can do it in public. Look at the permissions it requests from Google or Microsoft, not at the marketing page.

Pranan: it creates drafts for you to review and does not request the gmail.send scope. To be exact, the gmail.modify scope it uses to create drafts and apply labels would technically allow sending. Pranan does not use it that way. Slack is different: if you connect Slack, Pranan can post a reply as you when you press Send on its review card. Details here.

2. What does it store, and for how long?

"We do not store your email" is rarely true for a tool that learns from it. Ask what is kept and when it is removed.

Pranan: it stores message body text and metadata such as sender, recipients, subject and date, because it needs them to classify, learn your writing patterns and draft. Formatted HTML copies are removed after about 90 days. The plain text and metadata are kept while your account is active. More here.

3. Is my email used to train models?

There is a difference between a model being trained on your data and your data being sent to a model to produce an answer for you. Ask about both.

Pranan: no model is trained or fine-tuned on your email, and your voice profile is never used for other users. Message content is sent to the AI providers listed in the privacy policy, only to classify and draft for you.

4. Who else processes it?

Every AI tool relies on other companies: model providers, hosting, email delivery. A vendor should be able to list them.

Pranan: the full list of service providers, what each one processes and where, is in section 6 of the privacy policy.

5. How are my account credentials protected?

The tokens that give a tool access to your mailbox are the most sensitive thing it holds.

Pranan: Google OAuth tokens are encrypted with AES-256-GCM before storage and the service runs over HTTPS.

6. What independent assurance is there?

Ask for audit reports and certifications, and be wary of logos without a report behind them.

Pranan: it is not currently represented as SOC 2 certified. If you need a security review, the security page explains how to send your questionnaire and what evidence is available.

7. Can I leave, and what happens to my data?

You should be able to disconnect, export and delete without emailing anyone.

Pranan: when you disconnect Gmail, it deletes your stored tokens, revokes its Google access and deletes the email and calendar data it stored from that account. Deleting your account removes everything else from production systems, and backup copies expire within 30 days. You can request an export first.

One more thing to do yourself

No setting protects mail that should never pass through an AI service at all. Pranan has a never-draft list that stops it drafting for a contact, but that contact's mail is still synced, stored and classified. If a conversation must stay out entirely, keep it out of the connected mailbox.

Ask these seven questions of every tool on your shortlist. Any vendor worth trusting will answer them plainly.