How to spot AI phishing emails: 5 checks AI can't fake

AI makes phishing emails hard to spot by their writing, because the grammar is clean, the tone fits and the details can be about you. So stop judging the words and check the five things AI cannot change: the real sender domain, where replies go, where the links lead, what the email asks for, and whether the person confirms it when you contact them yourself.

What AI changed about phishing emails

The FBI spelled it out in a December 2024 public service announcement on criminals using generative AI. AI tools "can correct for human errors that might otherwise serve as warning signs of fraud." Criminals use them to translate scams so foreign actors make fewer grammar and spelling mistakes, to write messages faster and reach a wider audience, and for spear phishing, which the FBI defines as a directed attempt to trick a specific person using information "intended to increase the chance of success."

Two old tells are gone:

Bad grammar. CISA now says a common sign "used to be poor grammar or misspellings although in the era of artificial intelligence (AI) some emails will now have perfect grammar and spelling" on its page about how to recognize and report phishing.
Generic wording. A model can write a note in your boss's style, mention your team, or refer to an order you really placed.

The numbers are early but real. The FBI's 2025 Internet Crime Report counted 22,364 complaints that mentioned AI, with more than $893 million in reported losses. For business email compromise, where a fake executive or supplier asks for a payment, the FBI notes that "chat generators can quickly create official-sounding emails mimicking a company's CEO," and businesses reported over $30 million lost to BEC scams involving AI in 2025.

What AI did not change is the plumbing. The scammer still has to send from an address they control, receive your reply somewhere, host a page somewhere, and get you to do something. That is where you look.

Check 1: the sender's domain, not the name

The display name is whatever the sender typed. AI can make it perfect. The domain after the @ is the part they have to own.

Open the full sender address and read the domain from right to left. In billing@yourbank.com.secure-notice.net, the domain is secure-notice.net. CISA's example of a lookalike is "amaz a n.com", one swapped letter that is easy to miss when the rest of the email reads well.

The FBI's 2025 alert on AI voice and text impersonation gives the same instruction for every channel: "Carefully examine the email address; messaging contact information, including phone numbers; URLs; and spelling used in any correspondence." Our main guide on how to spot a phishing email shows where to find the full address in Gmail, Outlook and on a phone, and how to run a 60-second header check that proves whether the From domain is forged.

Check 2: where your reply would go

This one catches well-written fakes that pass a quick look at the From line. The email can show a real-looking sender and still route your reply to a different address.

Press Reply, look at the address that fills in, and then close the draft without sending. If the From line says your supplier and the reply goes to a free mailbox or an unrelated domain, stop. That is how invoice scams keep the conversation going with the scammer while it looks like the company.

Check 3: where the links actually lead

AI writes the button text. It does not change where the button points.

On a computer, rest the pointer on a link and read the address that appears. On a phone, press and hold the link until the preview shows the full address. CISA lists "untrusted shortened URLs" and incorrect links as signs of phishing.

The FBI adds an AI-specific twist: criminals "embed AI-powered chatbots in fraudulent websites to prompt victims to click on malicious links." So a helpful chat window on the page you landed on is not reassurance. If the link does not go to the company's real domain, nothing on the page matters.

The safest habit is to not use email links for anything that involves signing in or paying. Type the address yourself or use a bookmark. If the email was real, the same notice will be in your account.

Check 4: what the email wants, and how fast

A flawless email still has to ask for something. The ask is where AI phishing looks exactly like the old kind.

Be suspicious of any email that wants one of these:

a sign-in through its own link, or a "confirm your account" step
a change to bank details, a new payment, or an invoice paid to a new account
a verification or two-factor code
gift cards, cryptocurrency or a wire transfer
a move to a different app or phone number "to continue the conversation"

CISA flags "urgent or emotionally appealing language, especially messages that claim dire consequences for not responding immediately." The FTC is plain that "legitimate companies won't email or text with a link to update your payment information," on its page about recognizing phishing scams.

The last bullet comes from the FBI's December 2025 alert on impersonation campaigns that use AI-generated voice messages. The actors make contact, briefly talk about a topic the target knows well, and ask to move to an encrypted messaging app "almost immediately." Once there, they have asked for an authentication code, copies of documents such as a passport, and wire transfers. The FBI's advice: never give a two-factor code to anyone over email, text or a messaging app.

Check 5: confirm it through a channel you choose

This is the check that still works when everything else looks perfect. If an email asks for money, a code or a sign-in and you are not sure, contact the sender a different way, using contact details you already had or looked up yourself.

A company: go to its website by typing the address, and use the phone number or support page listed there. CISA's advice is to "go to the company's website and capture their contact information from the verified website."
A colleague or boss: call or message them on the number or chat you already use with them. Do not reply to the email and do not use a phone number from the email's signature.
A friend or relative: text or call the number in your contacts. CISA gives the example of a strange Facebook message from a friend: text or call them to ask if they sent it.

The FBI's generative AI alert puts it the same way for calls: hang up, research the contact of the organization yourself, and call that number directly. A phishing email can fake every detail on the screen. It cannot answer a phone number it does not control.

Why personal details are not proof

AI-written spear phishing often opens with something true: your job title, your manager's name, a recent purchase, a project you posted about. That feels like proof the sender knows you. It is not.

Those details are usually public or leaked. Profiles, company pages and old data breaches give a scammer plenty to feed a model. The FBI recommends limiting what you share, making social media accounts private and limiting followers to people you know, so there is less to build a convincing message from. If you want to see what has already leaked with your address, check our guide to finding out whether your email was in a data breach.

Treat accuracy as neutral, the same way you now treat good grammar. A real detail tells you the sender did research. It says nothing about who they are.

If you already clicked or replied

Do not panic, and act in this order: change the password for the account the email was about (from the real site, not the link), sign out other sessions, and turn on two-factor authentication if it is not on yet. If you entered card or bank details, call your bank using the number on your card. Our step-by-step guide on what to do if you clicked a phishing link covers each case, including attachments.

Then report it. In the US, forward the email to reportphishing@apwg.org and report it at ReportFraud.ftc.gov. If you lost money, also file at the FBI's ic3.gov. If someone may be using your personal information, IdentityTheft.gov gives a recovery plan.

Frequently asked questions

Can you tell if an email was written by AI?

Not reliably, and it does not matter much. The FBI says AI-generated content "has advanced to the point that it is often difficult to identify." A real email can be AI-assisted and a scam can be hand-written. Judge the sender domain, the reply address, the links and the request instead.

Are AI phishing emails more dangerous than regular ones?

They are more convincing, especially when they are personal, but they are stopped by the same checks. The sending domain, the reply address and the link destination are not something a language model can change.

What does an AI phishing email look like?

Usually like a normal work or account email. Common forms are a message from your "CEO" about an urgent payment, a supplier asking you to update their bank details, an account alert with a sign-in link, or a short friendly note that tries to move you to a messaging app. The writing will often be clean and specific to you.

Does two-factor authentication stop AI phishing?

It helps a lot, as long as you never hand over the code. Phishers ask for codes because codes are what stand between your password and your account. Passkeys and security keys go further because there is nothing to type into a fake page. See how safe two-factor authentication is for how each method holds up.

Make a targeted phish easier to catch

One habit makes spear phishing much easier to spot: give each site or service its own email address. With Secria Mail you can use a separate alias for your bank, your shopping and your work tools. If a "bank" email shows up at the alias you only gave to a shop, you know it is fake before you read a word, and if one alias starts getting phished you delete it and the rest of your inbox is untouched. Tracking pixels are removed automatically and senders never see your IP address. The free plan includes 2 aliases, and sign-up is a name, the address you want and a password, with no phone number. Create a free account.