The short answer
Use AI for the first draft, never the final send. Give it the full thread, forbid it from inventing facts (unknowns should stay as visible [placeholders]), make sure the provider does not store or train on your mail, and require a person to approve every message. Tutti is built around these four rules.
On this page
Why this matters now
AI drafting is already normal. In Muck Rack's State of AI in PR 2025, 75% of PR professionals said they use generative AI and 72% use it to draft content, yet over half said their company has no formal AI policy. On the receiving end, 72% of journalists in Cision's 2025 State of the Media said they worry about factual errors in AI-generated PR content. The gap between those numbers is where reputations get hurt.
These four rules close it.
Rule 1: Give the AI the whole thread
A reply written from the last message alone misses what was already agreed, promised, or asked. Make sure your tool reads the entire conversation, including your own earlier replies, before drafting.
Rule 2: Never let it invent facts
Language models fill gaps with plausible guesses: a meeting time, a job title, a statistic. In work email, a plausible guess is a mistake waiting to be sent. Instruct the model to use only the thread and the details you provide, and to leave anything unknown as a visible placeholder, like [launch date] or [spokesperson]. A person fills those in before sending.
Rule 3: Keep your mail out of training data
Email contains personal data, deals, and unreleased news. Use AI providers that process your content to produce the draft and do not store it afterwards, often called zero data retention. Check the setting, not just the marketing page.
Rule 4: A person approves every send
AI should prepare, not decide. Every message should be reviewed and sent by a person who is accountable for it. For sensitive mail, such as press, investors, and partners, name who approves.
A checklist before you roll out AI drafting
- The tool reads the full thread
- Unknown facts appear as placeholders, never guesses
- The AI provider does not store or train on your mail
- Nothing sends without a person
- Roles define who drafts and who approves
- The team knows the policy in writing
How Tutti applies these rules
Tutti was built around them. It drafts every reply from the full thread; its model is instructed never to claim facts that were not provided and to leave unknowns as [placeholders]; every AI request is routed to zero-data-retention providers; and nothing is ever sent automatically: members submit drafts and owners or admins approve. See the best AI email assistants for teams for how other tools compare.
Frequently asked questions
Is it safe to let AI write work emails?
It is safe to let AI draft them if a person reviews every message, the AI cannot invent facts, and your mail is not stored or used for training.
What is zero data retention for AI?
A provider commitment that prompts and outputs are processed to produce the answer and not stored afterwards. Tutti routes every AI request to zero-data-retention providers.
How do you stop AI from making things up in emails?
Instruct it to use only the thread and the details you supply, and to leave anything unknown as a bracketed placeholder. Then have a person fill in and approve.