Treat an unexpected urgent message as something to verify through a trusted route. AI can explain warning signs, but it cannot certify that a message or link is safe.
Messages that request money, passwords or verification codes deserve careful checking. You can ask AI to explain the wording without clicking the link or sharing private information.
A practical workflow.
Remove sensitive information
Leave out account numbers, passwords, verification codes and personal identifiers. You can describe a request without exposing the data it asks for.
Look for the requested action
Ask the AI to identify urgency, threats, unexpected attachments and requests to sign in or pay. These are warning signs to investigate, not a complete verdict.
Verify independently
Use a website address or contact method you already know, rather than the link or number in the message. Check the request directly with the organisation or person.
Respond through the trusted route
If you have already shared information, use the relevant service’s official recovery guidance. Do not ask the suspicious sender how to fix the problem.
READY TO MAKE YOURS
Start with this prompt.
Help me assess this message with personal details removed: [message]. Point out possible warning signs, but do not declare it safe from appearance alone. Suggest how to verify the claim through the organisation's official website or a known contact method. Tell me what information is missing. Do not ask me to click links, share codes or provide credentials.
Replace the brackets. Leave out private details. Check the response.
ILLUSTRATIVE EXAMPLE
A clearer request.
A VAGUE STARTIs this link definitely safe?
A CLEARER BRIEFExplain the warning signs in this message, with personal details removed. Do not open the link or declare it safe. Suggest how I can verify the request through an independent official contact route.
The prompt supports checking without creating false reassurance.
Before you use the result.
- No secret codes or passwords are shared.
- The sender is checked independently.
- Unexpected urgency does not replace verification.
- Official recovery instructions are used when needed.
A few useful answers.
Can AI detect every scam?
No. It may miss a convincing scam or misread a legitimate message. Use it to understand wording and questions to ask, not as the final security check.
Should I reply to ask whether the message is real?
Verify through contact details you already trust. A reply to the original sender does not independently confirm their identity.
PUT THE GUIDE TO WORK
Use AI to organise a suspicious-message check
An unfamiliar message can make a routine decision feel urgent. Slow the request down before asking a model to explain it. AI can help organise the wording and the questions you need to ask; it cannot prove who sent a message, inspect an attachment for you or certify that a destination is safe. Keep the original message available privately and work with a redacted description.
Separate what the message says from what you know
Make two short lists. The first records the sender’s claims: a delivery failed, an account needs attention or someone needs an answer. The second records facts you already know: whether you expected a delivery, whether an order exists and whether the named person has contacted you before. Do not let a summary merge those lists. A confident-sounding message is still a claim. The useful output is a specific question to verify, such as “Is there an issue with my existing order?” rather than a general verdict about whether the message looks convincing.
Verify outside the message
The FTC recommends contacting the organisation through a website or phone number you already know is genuine, rather than using the unexpected message’s contact details. Avoid opening its links or attachments while investigating. This is the important practical step; a model’s opinion about the writing is not a substitute. Record the result of your independent check, including any reference the genuine provider gives you. Use your email or messaging service’s reporting feature where appropriate. The official FTC guidance below also explains what to do if you have already shared information.
Prepare a useful record for a real support conversation
A support record needs the time received, the claimed organisation, the kind of action requested and the checks you have already made. It does not need to reproduce private account details in a public chat. Replace names and identifiers with labels and describe a link as “an unfamiliar link” when the exact address is unnecessary. Keep the original privately if the provider needs it through its official reporting process. Ask AI to format the record and highlight unanswered questions. Do not ask it to contact the sender, run instructions from the message or fill in facts you have not confirmed.
Open a card, copy the prompt and replace the brackets with your own details. Each card includes a follow-up and a check before you use the answer.
SITUATION 01 ↗
Separate claims from confirmed facts
Turn an urgent message into questions you can verify.
Open the prompt +
Use this redacted message [message]. These are facts I independently know [facts], and this is what I expected [context]. List the message’s claims, the requested action and the unanswered questions. Keep claims separate from confirmed facts. Do not label the message authentic or safe, open a link or recommend replying to the sender.
Refine the next draft
Which question would resolve the main uncertainty? Rewrite it as a short question for the organisation’s genuine support channel.
Before you use it
Keep account numbers, passwords and verification codes out of the prompt. The model’s wording review does not establish the sender’s identity.
SITUATION 02 ↗
Write a question for genuine support
Ask about the real issue through a route you already trust.
Open the prompt +
I will contact the genuine provider through [route I independently know]. Help me ask about this claim [claim] using this non-private context [context]. Write a concise question and a list of facts I should have ready. Do not include a link or phone number from the unexpected message, and do not invent an account status.
Refine the next draft
The genuine provider answered [answer]. Update the record to distinguish that confirmed answer from anything still unresolved.
Before you use it
Use the provider’s actual support process. Do not give a chat model credentials or one-time codes to handle the conversation.
SITUATION 03 ↗
Prepare a redacted incident summary
Record what happened without copying every private detail.
Open the prompt +
Format this timeline [timeline] into a factual incident summary. I did [actions], and these are the observations I can confirm [observations]. Use placeholders for personal identifiers. Separate what happened from what I suspect. List missing information as questions. Do not infer that an account is compromised or secure from the timeline alone.
Refine the next draft
Shorten the summary for an official reporting form. Preserve the sequence of events and remove speculation. Leave a clear place for the provider’s reference number.
Before you use it
Follow the official reporting service’s requirements. If private evidence is requested, provide it through that genuine service rather than a general AI chat.
SITUATION 04 ↗
Explain the check to a family member
Make a calm next step clearer than the message’s pressure.
Open the prompt +
Help me explain this situation [redacted situation] to [reader]. The next step we chose is [independent verification step]. Write a calm explanation in ordinary language, with one action at a time. Explain what remains unknown. Do not shame the reader, amplify the message’s urgency or claim that spelling and logos prove authenticity.
Refine the next draft
Make it shorter and friendlier. Keep the chosen verification step and the distinction between a claim and a confirmed fact.
Before you use it
Give the person room to ask questions. A clear explanation should help them make the check, not pressure them into another rushed decision.
The examples below are original, illustrative edits using fictional situations. They show the checks to make, rather than promising what an AI tool will produce.
Your message verification record
Message received: [date / channel] Claimed sender: [organisation or role] Claim made: [what it says happened] Action requested: [what it asks me to do] What I independently know: [facts] What remains unverified: [questions] Private details removed from this summary: [categories] Genuine support route used: [known app / site / phone] Result of independent check: [answer / reference] Reporting action, if needed: [official route] Next action I chose: [action]
WORKED EXAMPLE
A parcel claim becomes a checkable question
An original fictional example: Jo receives an unexpected text saying a parcel needs an address update.
FIRST DRAFTThe logo looks right, so the delivery message is genuine. Follow its address-update link.
AFTER CHECKINGThe text claims a parcel needs an address update. Jo has not confirmed that claim. Check the existing order through the retailer or delivery service’s independently known site or app, without using the text’s link.
What changed
The edit keeps the message’s claim separate from a confirmed delivery issue. Visual resemblance does not answer the practical question. The next action is independent verification, not following the message.
WORKED EXAMPLE
A useful report states the sequence
An original fictional example: Lee wants to record an unfamiliar invoice email for workplace support.
FIRST DRAFTA criminal hacked our invoices. They must have accessed our whole system.
AFTER CHECKINGAt 09:10 I received an invoice email I did not expect. It named [supplier] and requested [action]. I have not opened the attachment. I checked our order record and did not find the stated reference. Please confirm the correct reporting process.
What changed
The report contains observations and actions the writer can support. It does not make claims about an attacker, system access or a breach without evidence.
Try it with your own task.
Take one unfamiliar message and make the claim-and-fact list. Use a genuine support route to verify the key claim. Add the result to the record; do not keep asking models to vote on whether the message is trustworthy.
Check the official instructions.
For settings and safety guidance, use the original provider's current help. Menus and account options can change.
Original sources & further reading.
Check current product guidance when a feature or setting matters to your task.
GO DEEPER / 1 FREE TEMPLATES
Prompts for this part of the task.
Use a template for the step you are working on. Add the required inputs, then check the result against your source material.
Browse the task-guide directory ↗
Use a small workflow that you can repeat.
- Describe the task and the limits that make it difficult.
- Give only the personal details needed; use placeholders for names and addresses.
- Ask for options and a small first step, not an overwhelming plan.
- Verify current facts, costs and safety information before making a commitment.
Keep in mind: Never share passwords, payment details or one-time codes. Treat help with a device or suspicious message as guidance to check, not proof that a link is safe.
Review a suspicious message safely
Identify questions to verify independently.You get: Message review and safe next steps+
COPY. PERSONALISE. CHECK.
Review this message with names, account details and links removed: [message]. Identify urgency, payment requests, impersonation or unusual instructions. Explain what each signal suggests without declaring the message safe or definitely fraudulent. Give a safe verification plan using a known official app or contact details I already trust. Do not ask me to click a link, download an attachment or share a code.
Example input
A text claims a delivery fee must be paid today; its link has been removed.
Illustrative scenario, not a generated or verified AI result.
Then ask
Give me a three-step verification checklist that never uses contact details in the message.
Before using the answer
Verify through a trusted channel; polished wording is not evidence of legitimacy.
Keep learning from original sources.
These sources inform the workflow and checks; the templates above are original LifestyleGlide prompts.
