A text pops up: your parcel’s address is incomplete, tap the link to fix it today or it goes back to the sender. You happen to be waiting for an online order. The link looks like the courier’s website, the wording is smooth with no typos, and the tone matches the notices you usually get. You’re between meetings and just want to get this small chore out of the way.

This is exactly the moment scammers wait for. The usual advice for spotting scam texts was to look for flaws: typos, odd phrasing, strange web addresses. Those clues still help, but they can no longer carry the decision, because AI has made a flawless fake text fast and cheap to produce. The more reliable question is this: can I do what this text asks without tapping its link? If yes, do it through an official channel you find yourself. If not, don’t do it.

Looking closely no longer tells real from fake

In June 2026, Google filed a civil lawsuit against a cybercrime group it calls the “Outsider Enterprise.” According to Google, the group sold “phishing kits” to other scammers. A phishing kit is a complete toolbox for cloning websites, sending texts in bulk, and harvesting passwords and credit card details. Google says the group was linked to about 9,000 fake websites and more than one million scam URLs, and in just two weeks of May 2026 it sent 2.5 million texts with scam links to Android users. TechCrunch reported that the kit, starting at $88 a week, let users build fake sites with AI platforms including Gemini, Google’s own AI model.

These numbers are about scale and cost. Building a convincing copy of an official website used to take someone with technical skills and time. Now you buy a kit, hand it to AI, and produce them in bulk. When every fake text can be fluent and polished, “look closely for flaws” stops being enough: a text with nothing wrong in it can still be fake.

Even the “official number” can be fake

Taiwan offers a more direct example. Its Ministry of Digital Affairs launched “111,” a short code reserved for government text messages. The idea was that only government agencies could send texts from 111, so people could trust anything that came from it. But in June 2025 the ministry announced that scam groups had illegally used 2G base stations to override the signal of legitimate ones, a technique known as signal hijacking, and send fake 111 texts in the government’s name. The ministry upgraded the system in response: government texts now begin with the last three digits of the recipient’s phone number and the name of the sending agency, so people can check three things together: the sender number, their own last three digits, and the named agency.

The upgrade is useful, and it also shows something: any feature you can look at can be imitated. Numbers can be spoofed, web addresses made to look close, and tone copied. What scammers want was never for you to fully believe a story. They want you, in the few seconds when you’re busiest and least patient, to follow the path they laid out: tap the link, fill in the form, enter the one-time code.

So instead of trying to out-look the scammers, step off the path they laid out.

Advertisement

One question, three outcomes

When a text asks you to take care of something, first identify the action it wants: complete an address, pay a fee, log in to confirm an account, update a card. Then ask: can this be done without the link in this text?

  • Yes, and the official channel shows the same thing. Open the courier’s app or your card issuer’s app yourself, or type a web address you already know into your browser, and check whether something is really waiting for you. If it is, handle it there. The text itself can simply be deleted.
  • Yes, but the official channel shows nothing. If the app has no such parcel and the bill shows no such charge, the thing the text wants you to handle doesn’t exist. Don’t act on it, and don’t reply.
  • No, only this link can do it. Legitimate organizations don’t make something important reachable only through a text message. That alone is the clearest warning sign.

The limit of this rule is that it assumes you know where the official channel is. For older relatives who aren’t comfortable with phones, opening an app to check may be harder than tapping the link. A workable alternative is to replace the official channel with a person: agree that texts like this get forwarded to a family member first, or call an anti-fraud hotline such as Taiwan’s 165. The point is the same: don’t start from the link in the text.

Some people paste the text into an AI tool and ask whether it’s a scam. That can help as a second opinion and catch common tricks, but AI can’t confirm whether the parcel or bill actually exists, so you still have to check through the official channel. Before pasting, remove names, phone numbers, addresses, and any verification codes.

Already tapped or entered details? Stop, keep evidence, then deal with it

If you or a family member already went in, don’t start with self-blame. The sooner you stop the damage, the smaller it is. Order matters:

  1. Stop. Don’t enter anything else, and don’t follow further instructions to download an app, allow remote access, share a verification code, or transfer money.
  2. Keep evidence. Take screenshots of the text, the web address, and any page where you entered details, and note what information you gave.
  3. Then deal with it. If you entered a card, call the number on the back of the card right away to block it. If you entered a password, change it through the official app or website and sign out other devices. If you’re unsure what to do, call your local anti-fraud hotline (165 in Taiwan).

If it involves a company account, customer data, or a company card, tell the people who can suspend the account and contact the bank directly, rather than only discussing it in a private group chat.

Next time a text asks you to “tap to take care of it,” pause for a second and ask that question: can this be done without tapping the link? You can also agree on one sentence with your family today: “We don’t tap links in texts. If something needs doing, we open the app ourselves.”

If you’re responsible for bringing AI tools into your company, the same thinking applies to another kind of scam: urgent calls where a familiar voice is imitated by AI; see When a Familiar Voice Calls With an Emergency, Don’t Judge by the Voice Alone. Source links attached to AI research reports can’t be trusted on looks either; see A Source Cited in an AI Research Report Shows Where It Read Something, Not That It’s True

AI handoff card

Ask AI to walk you through handling a suspicious text This prompt doesn’t ask AI to rule on whether the text is real. It asks AI to identify the action the text wants and the official channel you can use to do it without the link.

I received a suspicious text and want to handle it the way this article suggests: https://boosterminiclass.com/en/posts/ai-scam-text-message-triage-checklist/
The article's method: don't compare how real the text looks; ask "can I do what it wants without tapping this link?"

The text is attached at the end, with names, phone numbers, addresses, and codes already removed. Before you start, ask me 2 questions:
1. Do I actually have recent dealings with this organization (for example, am I really waiting for a parcel, do I really have this card)?
2. Do I have this organization's official app on my phone, or do I know its official website and customer service number?

Based on my answers, give me:
- The specific action this text is asking me to take;
- Steps to do the same thing through the official channel without using the link in the text;
- If the official channel shows nothing, a clear instruction not to act on this text;
- If I have already tapped the link or entered details, what to do now following "stop, keep evidence, then deal with it," including a reminder to call an anti-fraud hotline.
Don't stop at "this is probably a scam," and don't ask me for complete personal details.

AI can only help you sort out the steps. Whether the parcel or bill really exists must be confirmed through the official channel.

Advertisement

Share

Share this mini class

If this lesson helps untangle a work bottleneck, share it with someone deciding how to use AI.

References