How-to
Use AI to check a suspicious email safely
Chatbots are good at spotting scam emails and texts, and misjudge some real ones. Describe the message in this prompt and get the red flags, what it cannot know, and the checks that settle it.
In a test published on 15 November 2026, Android Police gave the free versions of Gemini, ChatGPT and Claude 20 emails and texts, 15 of them modelled on scams the FTC and Amazon have warned about, and all three caught every scam.
Two Amazon emails the tester treated as genuine, sent one at a time, split them: Gemini called the sign-in alert “almost certainly a phishing scam” on a second run, Claude grew more suspicious, and only ChatGPT called it legitimate every time. Use the prompt for the red flags, then open the company’s app yourself to settle it.
The prompt: check this message
Use a fresh chat in any assistant, and fill in the set-up without opening the link.
You are helping me decide whether a message I received is a scam. I have not clicked any link, opened any attachment or replied. If you call a scam genuine, I could lose money or an account. If you call a genuine message a scam, I could ignore a real warning from my bank or miss a delivery I am waiting for. So be careful in both directions, and tell me plainly what you cannot know from the text alone. ABOUT THE MESSAGE (I fill this in once; any line can say "not sure") - How it arrived: [email / text message / WhatsApp / social media message / other] - Who it claims to be from: [e.g. "my bank", "a parcel delivery company", "my streaming service", "my boss", "a tech company inviting me to a free event"] - The sender exactly as shown: [the display name and the full email address or phone number, e.g. "Account Team, account-update@secure-mail-centre.net", "+1 555 0100", "a short code: 72000"] - Do I have an account with them, or know this person: [yes / no / not sure] - Was I expecting anything from them: [e.g. "yes, I ordered a lamp on Monday", "no"] - The links, described without clicking: [the address as written in the message, or what shows when I rest my mouse pointer on the link, e.g. "button says Verify now, address shows yourbank.com.login-help.net"; "a short link, bit-style, cannot see where it goes"; "link hidden behind a button, address not visible"; "no links"] - Attachments, not opened: [none / the file name exactly as shown, e.g. "Invoice_0921.pdf.html"] - What it asks me to do: [e.g. "pay a fee", "confirm my card details", "sign in", "call this number", "enter a code", "buy gift cards", "sign in with my work account to claim a reward"] - Deadline or threat: [e.g. "account closes tomorrow", "none"] - The country I live in: [your country] - The message text, with my details swapped for [MY NAME], [MY ACCOUNT NUMBER], [MY ADDRESS], [ORDER NUMBER]: [paste here, or attach a screenshot with my name and numbers cropped out] SAFETY RULES FOR THIS CHAT - If I have pasted a password, a one-time code, a full card or account number, or an ID number, tell me in your first line to delete this chat and start a new one without it, call it [REMOVED] from then on, and never repeat it. - Never tell me to click a link, scan a QR code, open an attachment, reply, or call a number that appears in the message, even to test it. - Do not open or visit any link from the message yourself, even if you can browse the web. - Never invent a company's phone number, email address, website or reporting address. If you are not certain of one, write [FIND ON OFFICIAL SITE] and tell me where on the company's own site or app to look. HOW TO WORK (in this order) Check the set-up. If the message text, the sender or the links line is blank, ask for it and wait. If I said a link is hidden or shortened, do not guess where it goes; call it "not visible". If I sent a screenshot, read the text from it and remind me that a picture does not show the real address behind a button. Red flags. List every warning sign you find. For each one, quote the exact words or address from my message that show it, and say in one line why it matters. Check at least these: - The sender address against the company it claims to be. The name just before the ending (.com, .net and so on) is the real owner, so yourbank.com.login-help.net belongs to login-help.net. A display name proves nothing. - Pressure: a deadline, a threat to close an account, a fine, an arrest, or a family emergency. - A request for payment, card details, a password, a one-time code, gift cards, or personal details. - A link to update payment details or sign in. - A generic greeting where the real company would use my name. - An invoice, order or delivery I do not recognise. - An attachment with an odd or double file ending. - An offer that is too good: a free subscription, a refund, a prize, a free premium tool for registering. - A request to copy a code and enter it on a sign-in page, or to "continue with" my work or email account to claim something. A real sign-in page can still be part of a scam if the code was not my own request. Then list any signs that point the other way, such as the message matching something I was expecting, the sender address being the company's usual one, or the message asking me only to check my account through the app I already use. If you find no red flags, write "No red flags found in the text" and still give the checks below. What you cannot verify. Say plainly what text alone cannot prove: a sender address can be faked or look right; you cannot see where a hidden or shortened link goes; a familiar link service or a genuine sign-in page can still be part of a scam; and you cannot see my account, my orders or my bank. Verdict. Give exactly one of these labels and the main reason in one sentence: "Looks like a scam", "Suspicious: check before you do anything", "Looks genuine: confirm in your account", or "Cannot tell from this". Never say "safe" or "definitely real". - If I have no account with them and the message asks for anything, lean towards "Looks like a scam". - If it matches something I was expecting and asks me to do nothing beyond checking my account, do not call it a scam on tone or formatting alone. - If the clues conflict, say which ones and choose "Suspicious" or "Cannot tell from this". The checks that settle it. Give me the independent checks, in the order I should do them, using only routes I reach myself: - Open the company's app, or type its website address myself, and look for the same alert, order, bill or message there. - If I still need to contact them, use a number from my card, a statement, a past bill, or the company's own site. If you name a number or address, follow the never-invent rule above. - If it claims to be from someone I know, contact them another way, using details I already had before this message arrived. - If it is about my work account, ask my IT or security team through the usual channel. For each check, tell me what I should see if the message is genuine. If it is a scam. Tell me how to report it where I live. Name the type of place (my country's consumer protection or fraud reporting agency, my email app's report phishing or report junk option, my phone's report junk option for texts, and the company's own report address from its website). Name a specific agency or address only if you are certain it is right for my country; otherwise write [FIND ON OFFICIAL SITE]. Then tell me to delete the message. If I tell you I already clicked, entered details or paid. Stop the analysis and ask exactly what I entered or did. Then give calm steps in order of urgency: - Card or bank details: call my bank on the number on my card or the bank's own app, and ask them to block the card. - A password: change it by going to the site or app directly, then change it anywhere else I used the same one, starting with my email. - Sign out of all other devices on that account, and turn on two-step verification. - Check my email settings for forwarding rules I did not set. - A work account or a code I entered on a sign-in page: tell my IT or security team now, because changing a password may not end a session the scammer already has. - A one-time code for a personal account: go to that account now through its own app, change the password and sign out of all other devices, because the code may have let someone in. - If I only opened the link and entered nothing: say so plainly, tell me to close the page and check whether anything downloaded, and do not alarm me. - A file I opened or downloaded: run my security software and update it first. - If money left my account, ask my bank about disputing the payment and report it as above. SELF-CHECK BEFORE YOU ANSWER - Every red flag quotes something I actually gave you; you invented none. - You did not tell me to click, open, scan, reply to or call anything from the message. - Every phone number, email address or website you gave is one you are certain of, or marked [FIND ON OFFICIAL SITE]. - The verdict is one of the four labels, and you said what you could not verify. - You did not repeat any personal detail or secret I pasted.
Fill in the set-up safely
- Leave the link, the attachment and the reply button alone. The FTC’s phishing page warns that links and attachments can install harmful software.
- Read the link address without opening it. On a computer, most email apps and browsers show the address when you rest your pointer on the link; on a phone, write down only the address you can see in the message. If it is hidden or a short link, say so, since the AI cannot see where it goes either.
- Copy the sender exactly. Tap or click the sender’s name to show the full email address or number, then answer the FTC’s question: do you have an account with this company, or know this person?
- Take your details out before you paste. Swap your name, account and order numbers, address and phone number for
[MY NAME]and the like, because a genuine message carries real details that do not belong in a chat log. - Never put a password or a code into the chat. If the message contains a one-time code, delete it from what you paste; nobody checking a message needs it.
What the text cannot show
In late October 2026, fake event invitations copying OpenAI, SpaceX and Anthropic, one offering three months of free ChatGPT Pro, sent people through shortened links from Google and Zapier so the real destination stayed hidden, as Fox News reported on 15 November 2026 from Push Security’s research.
Some then asked people to copy a code and enter it on Microsoft’s real sign-in page, so a genuine sign-in page proves nothing; a code you did not ask for is the red flag the prompt looks for.
Settle it outside the message
- Open the company’s app or type its address yourself. A real alert, bill or delivery problem shows up in your account, which is the quickest answer either way.
- Call a number you already trust. Use the one on your card, a statement or the company’s own site; the FTC says to use a number or website you know is real, never the details in the message.
- Check with the person another way. If it claims to be a friend, your boss or a family member, call or message them on details you had before this arrived.
- Report it, then delete it. Use your email app’s report phishing or junk option and your country’s fraud reporting agency, so the next person gets a warning.
If you already clicked
Do the urgent things first. Call your bank on the number on your card if you gave card details, then change the password by going to the site or app yourself, and anywhere you reused it, starting with your email account. Check your email settings for forwarding rules you did not set, which in the test only ChatGPT suggested.
Then paste the message into the prompt and tell the AI exactly what you entered; it switches to recovery steps for the rest.
The honest bit
- The AI can invent the fix: in the same test, Claude said to forward the email to reportphishing@amazon.com, an address the tester could not find on Amazon’s pages; Amazon lists reportascam@amazon.com for people without an Amazon account.
- Your check: before you use any number, email address or website the AI gives you, find it on the company’s own site or app. If the verdict says the message looks genuine, still open the company’s app or type its address yourself before you act on it.
The facts here come from Android Police, the FTC’s phishing page and Fox News, read 16 November 2026.
Do it now
Find the last email or text that made you hesitate, fill in the set-up with the link described and your details swapped out, and run the prompt. Then save it somewhere you can reach in a hurry, such as a note on your phone.
A few quick questions
Can I send a screenshot of the message?
Yes, but type the link address separately if you can see it, or say it is hidden, because a screenshot of a button does not show where the link goes.
Where do I report a scam message?
Use your email app’s report phishing option, your phone’s report junk option for texts, and your country’s consumer or fraud reporting agency. In the US, the FTC asks for phishing emails to be forwarded to reportphishing@apwg.org, texts to 7726, and reports at ReportFraud.ftc.gov.