The risks of AI written sales emails and how to avoid them

The three biggest risks with AI written sales emails are hallucinated details that destroy trust, generic tone that burns your domain, and legal mistakes around data protection and marketing law. All three can be avoided with the same principle: train the AI properly on your voice, require a verifiable source behind every claim, and keep full visibility into what gets sent.

What is the most dangerous risk: hallucinated details?

An AI that lacks facts invents facts. In sales emails it looks like this:

• "Congrats on the expansion to Norway" when the company never expanded
• "I saw you recently switched your business system" when no switch happened
• Wrong CEO name, wrong product name, wrong industry
• Made up numbers about the recipient's market

A generic email gets deleted. An email with invented details about the recipient's own company gets remembered, screenshotted and shared. The hallucination is worse than silence because it looks like personalization but reveals that no human looked.

The protection is a source requirement: every claim about the recipient in the email should be traceable to an actual source, for example a press release, a job ad, a website or your own CRM history. If the system cannot show the source, the claim should not be sent.

How does generic AI tone burn your domain?

Recipients recognize AI emails by the patterns: the same opening phrases, the same polite filler sentences, the same structure in every paragraph. The consequences are technical, not just aesthetic:

1. Low reply rates and high delete rates signal to Google and Microsoft that your emails are unwanted.
2. Spam complaints from annoyed recipients weigh heaviest of all in their filters.
3. Domain reputation drops, which hits every email from your domain, including quotes and customer emails.
4. Recovery takes months. A burned domain reputation is not fixed with one setting.

Volume makes everything worse. One bad email is a miss, a thousand identical bad emails are a pattern the filters punish you for.

What legal risks come with AI emails?

Three areas need order:

• Data protection. B2B emails to relevant decision makers can rest on legitimate interest, but you have to be able to show the relevance assessment, offer easy unsubscribe and have a data processing agreement with the AI vendor. If the AI sends to the wrong roles, the basis falls apart.
• Marketing law. Sends to sole traders are in practice treated like sends to private individuals, which requires consent. An AI that does not distinguish between company types can make mistakes in your name.
• Liability. You are the sender. "The AI wrote it" is no legal defense if an email contains misleading claims about your own product.

What is the solution: careful voice training and full visibility plus sources?

The risks do not come from AI writing, but from no one reviewing. A working process has four parts:

1. Research with sources. Every personal detail in the email links to where the fact came from.
2. Careful voice training and full visibility. Calibrate the AI against how your best salesperson actually writes, and make sure you can review every sent email afterward.
3. Stop rules. An uncertain fact means no send, not a guessed send.
4. Logging. You should be able to see afterward what was sent, to whom and on what basis.

That process is slower than letting the AI send freely. That is the point. You trade a marginal time saving for the protection of your domain and your brand. It is also how we have built Leo at Revexa: research with sources and careful voice training and full visibility before anything leaves the outbox. If you want to see what your existing leads are worth with that method, there is a calculator on the homepage.

Common questions

Can recipients tell an email is AI written?

Often, from the patterns: generic openings, filler phrases and details that are almost but not quite right. Reviewed emails with verified research are indistinguishable from handwritten ones.

Is it legal to send AI written sales emails?

Yes, with the same rules as for any sales email: a legal basis under data protection law, relevant recipients, an unsubscribe option and caution with sole traders. AI does not change the rules, only the speed at which mistakes can be made.

Is it enough to spot check the AI emails?

At the start, no. Review everything until you have seen the system's error patterns. After that you can move to spot checks on the routine and full review of anything built on new research.