What not to automate

AI drafts. You decide. You send.

Don’t set up anything that sends messages to real people without a human approving each one, at least until you’ve checked the rules below.

1. Recording and transcribing calls: get consent first

• Some US states and many countries require everyone on a call to agree to be recorded; others require only one party. Rules also apply to AI note-takers that record or transcribe.
• Safest habit: say it at the start (“I’d like to record this for notes. Okay with you?”) and honor a “no.”
• Never feed a transcript into an AI tool unless you had consent to record and your company allows it.
• Check where the AI tool stores recordings and transcripts, and whether it trains on them.

2. Email: know the basics

• In the US, the CAN-SPAM Act covers commercial email, including B2B cold email. The basics: don’t fake who you are, don’t use misleading subject lines, include a valid physical postal address, and give a clear way to opt out and honor it promptly.
• Other places are stricter. The EU/UK (GDPR / PECR) and Canada (CASL) can require consent or a legitimate basis before cold outreach, depending on the recipient. Check before you email people there.
• AI doesn’t change any of this. You are responsible for every email you send.

3. Text messages: consent is the rule, not the exception

• In the US, marketing texts generally need the person’s prior consent (the TCPA and carrier rules are strict, and penalties can be steep). A person asking for a quote by text is different from a person on a purchased list. When in doubt, don’t text.
• Honor STOP replies right away.
• Don’t use AI to send bulk “personalized” texts without a consent record.

4. Platform rules (LinkedIn and others): read the Terms of Service

• LinkedIn’s terms prohibit scraping and unapproved automation tools (bots, auto-connect, auto-message). Accounts get restricted. Selling With SI plays use AI to draft; you copy, paste, and send by hand.
• Same for any site or tool you use: no scraping where the terms forbid it, no fake accounts, no bypassing limits.
• Check the AI tool’s own usage policy too.

5. Privacy: don’t paste what you shouldn’t

• Don’t paste customer data, contracts, pricing you’ve been told to keep confidential, or personal info into a consumer AI tool without checking your company policy and the tool’s data settings.
• Learn with fake data first. Every Selling With SI sample uses fake data.
• Strip names, emails, phone numbers when you can.

6. Facts: AI makes things up

• Never send a number, quote, customer name, or claim you didn’t verify yourself.
• “Personalization” that is invented is worse than no personalization. Don’t pretend you read something you didn’t, and don’t use flattery you don’t mean.
• Don’t let AI promise discounts, deadlines, or terms on your behalf.

7. Things we say no to

• Auto-sending any message without a human OK.
• Fake reviews, fake testimonials, fake scarcity, fake “we spoke last week” openers.
• Impersonating a person or implying a human wrote something personally when no one did, when that would mislead the reader.
• Scraping personal data from sites that forbid it.

8. AI agents that act on their own: keep approval on

An AI agent is different from a chat window. It can be connected to your email, calendar, CRM, and the web, and it can keep working after you close the tab. That means it can send, post, record, edit, and spend, not just draft. Our rules still apply to it.

• Start with approval on. Let it draft and propose. You read it, and you send. If an agent offers a “run on its own” or autopilot mode, leave it off until you’ve read every message it would send and your company has said yes.
• Give it the least access that does the job. Connect one inbox or one CRM view, not everything. Know how to disconnect it and where it keeps what it read.
• Agents are not exempt from consent, email, text, or platform rules. An agent that emails, texts, records a call, or drives LinkedIn for you is you doing those things. Sections 1 to 4 above apply.
• Check the facts it brings back. Agents research on their own and can be wrong. Don’t send an agent’s number, quote, or claim you haven’t checked yourself.
• Don’t let it pretend to be you. If an agent writes or talks to a person on your behalf, make sure the person isn’t misled about who or what they’re dealing with, and follow your company’s disclosure rules.
• Check where your data goes. Ask your company before you connect customer data, and read the vendor’s data and training settings.
• We list agents; we don’t endorse them. Our agent entries are information pages built from vendor sources with a last-verified date. They are not tested plays.