Can ChatGPT Tell If a Message Is a Scam? What It Can and Can't Check

A lot of people now paste suspicious messages into ChatGPT and ask, “is this a scam?” It’s a reasonable instinct, and sometimes it works. But it’s worth understanding exactly what a general chatbot can and cannot verify, because the gap between the two is where people get hurt.

What ChatGPT genuinely can do

ChatGPT has read descriptions of thousands of scam patterns. If you paste the text of a message, it can often recognize the shape of a scam:

• Urgency plus a payment demand (“your electricity will be disconnected tonight”)
• The classic KYC SMS structure of threat, link, and deadline
• Advance-fee logic, lottery logic, and job offers that ask you to pay first

For pattern recognition on plain text, a chatbot is better than guessing. If the message is a well-known template, it will usually say so.

What ChatGPT cannot do

The problem is that modern scams are rarely decided by the text alone. The danger lives in the parts a chatbot can’t reach:

It can’t open the link. The message says sbi-kyc-verify.info. Is that page a credential-harvesting clone? Was the domain registered last week? ChatGPT can’t visit it, sandbox it, or inspect where it redirects. It can only speculate from the URL’s spelling.

It can’t verify the sender. Whether that number has been reported before, whether the sender ID matches the real bank’s, whether the WhatsApp account was created yesterday: all invisible to a chatbot.

It can’t check breach databases. “Has my email leaked?” needs live lookups against breach corpora, not general knowledge.

It can’t examine images properly. Fake payment screenshots and AI-generated photos need forensic checks, not a plausible-sounding opinion.

It doesn’t know today’s scams. A chatbot’s knowledge has a cutoff. The scam wave that started circulating in your city this month, the fresh domain, the new script: fact-checkers document these daily, and that stream is not in the model.

It won’t follow up. You close the chat and it forgets you. If the same scam resends in three days, or your dad gets the same message, the chatbot doesn’t know and doesn’t care.

It can be confidently wrong. Chatbots aim to be helpful and fluent. Asked about a borderline message, they can produce a reassuring answer with no evidence behind it. In scam checking, unfounded reassurance is the most expensive possible output.

The test that matters: a real forward

Take a typical dangerous message: “India Post: parcel held, pay ₹49 to release: indiapost-track.top”.

A chatbot will correctly note this looks like the parcel fee scam. What it can’t tell you is what’s actually behind that specific link. A verification tool opens it in a sandbox, sees the card-detail harvesting page, checks the domain age, and can say: this exact link is a scam, here is why, delete it, and here’s what to do if you already paid.

The difference is between “this resembles a scam” and “I checked, it is one.”

What to use instead (or alongside)

Use a focused review contact alongside the original source. Kaval lives on WhatsApp: forward the actual message, link, screenshot, or claim and ask what the available evidence supports, what remains uncertain, and what to do next. Depending on the item and available checks, Kaval can help examine links, claims, images, and suspicious wording without moving the content to another app. It is useful context, not proof or a security scan.

We’ve written a fuller side-by-side in Kaval vs ChatGPT for fact-checking, and if you’re choosing tools more broadly, see the best free fact-checking tools.

The bottom line

ChatGPT is a good second opinion on the wording of a message. It is not automatically a scam checker: what it can inspect depends on the tools and context available in that product. For a message that involves your money, compare any AI answer with the official organisation and primary evidence. You can forward it to Kaval on WhatsApp at +91 7200218310 and ask for a review with relevant evidence and next steps where available.