Start with the sender and the request
Polished writing cannot establish who sent an email, and an AI score cannot tell you that its links or requests are safe. If an unexpected message asks you to pay, share a password or open an attachment, verify the request before spending time on how it was written.
- Read the sender's full address, including the domain after the @ sign. A familiar display name on its own does not establish identity.
- Compare the request with the conversation you expected to have. If it changes payment details or asks for sensitive information, contact the person using a number or address you already had.
- Open the relevant service using your usual bookmark or its known address. You do not need to follow the message's link to check your account.
- If it is a work message and you remain unsure, use your organisation's reporting process and keep the original email for the person reviewing it.
What writing style can tell you
A generic greeting, repeated phrasing or a closing paragraph that restates the message may prompt a closer read. None establishes AI authorship. People use templates, and someone using AI can supply details about you or a previous conversation. Specific details are worth checking against what happened, rather than taking them as proof that the sender is human.
If you need to know how a colleague prepared a message, ask about the work itself or the assistance they used. An unusual phrase is not a sound basis for an accusation.
When you can try an email scan
Cobalynx accepts English text of at least 150 words. You can paste a single longer email into the text scanner for an AI-text estimate. Passing the word minimum only makes the text eligible for a scan; it does not establish the accuracy of the result.
We have not published separate false-positive measurements for emails or business writing. The limits on the evidence page explain why the overall error rate cannot be treated as an email-specific rate. The methodology explains what the score means and why a result may be inconclusive.
Do not join unrelated messages or a conversation between several people just to reach the minimum. That produces a different text from any one message you are trying to understand, and a combined score cannot assign authorship to each reply. Keep quoted messages separate from the email you intend to assess.
If someone questions an email you wrote
Keep the original message and any drafts that already exist. Ask which part raised the concern and what policy applies to AI assistance. Explain the work behind the message, including any translation or editing help you used. A second detector score cannot settle the disagreement.
For an application, see the cover letter guide. If an accusation affects your work, the false-accusation guide covers the records to keep and the questions to ask.
Common questions
Can I check a 60-word email?
No. A 60-word email is below our minimum input length. Do not combine a thread or unrelated messages to force a score. The result would describe the combined input, not establish who wrote each message. Verify unusual requests through a contact channel you already trust.
What words give away ChatGPT in an email?
No word or stock greeting proves AI use. People and AI both use phrases such as “I hope this email finds you well.” Read the content and verify the sender rather than using a vocabulary blacklist.
Are AI-written emails a security risk?
The real change is that well written no longer means safe. For years people relied on broken English as the tell for a scam email, and phishing written with AI help took that tell away. So we would judge an email by if the sender’s address checks out, if you were expecting it, and what it is asking you to do, and we would give no weight at all to how fluent it is.
Can people tell if I use AI to write my emails?
There is no dependable guarantee either way. A detector can be wrong and its score does not prove authorship. Follow the disclosure rules that apply to your work and check the accuracy of what you send.