of B2B buyers say they are more likely to meet misleading information from GenAI than from a sales rep.
Gartner, 645 buyers, 2025of B2B buyers turn to sales reps to check what AI tools told them.
Gartner, 645 buyers, 2025of people say they ignore marketing emails they suspect were written by AI.
Exclaimer, 1,000 consumers, 2024A verified job title or a reported headcount. The email can state it plainly.
"Pipeline might be a challenge" becomes a question to the reader, not a statement about them.
If a fact is not there, it is not mentioned. If a required fact is missing, no email is written.
"7 of 10 ready" tells you what to fix. "68%" does not.
All four or nothing gets written.
Each one present is a fact the email is allowed to use.
A draft that fails any of these is never saved, so it can never be sent, whether you approve by hand or use autopilot.
wrong_recipient_nameWrong recipientThe greeting names someone other than the lead.unresolved_placeholderUnfilled placeholdersTemplate variables that never got filled in.markdown_or_htmlLeaked formattingMarkdown or HTML in a plain-text email.wrong_stepWrong sequence stepFollow-up text saved as the first email.duplicate_contentDuplicate contentThe same text repeated across steps.empty_subjectEmpty subjectNo subject line at all.empty_bodyEmpty bodyNothing in the message.subject_too_longSubject too longA subject line that would be cut off.Instead of a vague warning not to make things up, the AI gets a concrete list: no funding stage, no headcount, no technologies.
A reported headcount is a fact. A suspected challenge is a hypothesis, and hypotheses only ever become questions.
The facts behind each email are saved as a fixed, fingerprinted snapshot, so you can see exactly what it was written from.
Yes, AI can be wrong, which is why every draft waits for your review by default. The guardrails cut the most damaging errors: invented facts, wrong names and broken templates.
Because a required fact was missing: the person's name, job title, company or a verified work email. The lead stays visible with the reason, so you can fix it or move on.
From the lead database and enrichment, plus any details you upload or add. The agent only uses what came back, and says nothing about what did not.
No. Autopilot changes who approves an email, never whether it was checked. Every draft passes the same eight checks either way.
It stays general and truthful, and refuses to invent customers, prices or results. That exact situation is one of the ten pre-launch test scenarios.