References

An off-the-record reference is still on the record.

"Between us, I would not rehire him, he was difficult and probably dishonest." An informal reference can be defamation or a negligent-reference claim. VerbaPulse flags it before the email sends.

Book a demo →See pricing

The problem

‘Off the record’ does not exist in an email

Giving a reference is a minefield: say too little and it is unhelpful, say too much and it is actionable. The risk is the candid, unsubstantiated line: I would not rehire him, she was let go for stealing, he was probably dishonest. Written in an email, that can be defamation or a negligent misstatement, and "off the record" does not protect it.

The writer is being honest with a peer. But an opinion stated as fact, about a named person, in writing, is exactly what a reference claim needs.

What it catches

The disparaging line, marked for removal

VerbaPulse reads the draft as it forms and flags the span that creates the exposure, with a plain reason. Real output from the product:

VerbaPulseHigh

"Off the record, I would not rehire him, he was difficult and probably dishonest."

Disparaging comment with no legitimate content

Remove before sendRemove

For a disclosure the safe move is removal, not a reword. VerbaPulse marks the exact span while the message is still a draft.

The same check covers the related patterns:

Opinions stated as fact: "he was dishonest", "she was a problem".
Unsubstantiated allegations: "probably", "I heard that".
Disclosing a prior termination reason about a named person.
"Off the record" or "between us" framing that gives no real protection.
Your own rules: upload your reference policy and VerbaPulse enforces it on every draft.

How it works

In the inbox your team already uses

Install in minutes

The Outlook add-in deploys org-wide through the Microsoft 365 admin center, and the Chrome extension covers Gmail. No new tab, no change to how the team works.

Write as usual

As the draft forms, a risky span is flagged with severity and a plain reason, so the writer sees the exposure in the line they just wrote.

Fix before it sends

Apply the safer wording, or remove the line, while the message is still a draft. Anonymized events feed an audit trail your team can show.

Where this fits

A writing-time check, not a reference policy

VerbaPulse does not set your reference policy or give legal advice. It catches the disparaging, unsubstantiated language in the draft, while the wording can still change, so a candid line does not become a defamation or negligent-reference claim. It is one control inside email compliance for HR and recruitment, and it complements the systems you already run.

Email content is never stored: drafts are analyzed in memory and discarded immediately.
Nothing is used to train AI models.
The audit trail is anonymized by design: risk types, severities, and outcomes, never message text or named individuals.
It tells the difference between a factual, policy-compliant reference and a disparaging opinion stated as fact.

For the evidence behind this, our language risk benchmark runs real, anonymized cases through the product and reports what it flags.

FAQ

Common questions

Why are reference emails a legal risk?

An opinion stated as fact about a named person, especially an unsubstantiated one, can be defamation, and an inaccurate or careless reference can be a negligent misstatement. "Off the record" in an email does not protect it. This page describes language risk and is not legal advice.

How does VerbaPulse help?

It flags the disparaging, unsubstantiated line as it is written and marks it for removal. For example, "I would not rehire him, he was difficult and probably dishonest" is flagged as a disparaging comment with no legitimate content.

Does it write the reference for us?

No. It does not draft references or decide what to say. It only checks the outbound email for disparaging or unsubstantiated language before it leaves.

Is email content stored anywhere?

No. Drafts are analyzed in memory and discarded immediately. They are never stored and never used to train AI models. The audit trail keeps anonymized risk events (type, severity, action taken), never message text and never named individuals.