AI can imitate confidence and familiar voices. Build a habit of checking the source.

Start with the claim you need to verify

A polished answer, an official-looking document or a familiar voice can make information feel settled before it has been checked. Generative artificial intelligence adds new ways to produce those signals.

The practical response is to decide what evidence would establish the particular claim in front of you.

The National Institute of Standards and Technology describes confabulation as AI generating and confidently presenting false or erroneous content. Its generative-AI profile recommends reviewing sources and citations in system outputs.

That is a useful principle for everyday research as well as organizational use.

For a family using a chatbot to understand an unfamiliar subject, the answer can suggest questions to investigate. A deadline, quotation, eligibility rule or claim about a real person needs a source that actually supports it.

Fluent wording and a neatly formatted reference list do not finish that work.

Open the reference and test what it says

The Library of Congress’s primary-source analysis guide encourages readers to examine origins, context and unanswered questions. Our suggested method applies that habit to everyday digital claims: identify the sentence that would change your decision, then find the relevant passage in the original document.

Check its issuer, date and scope, and whether a later version replaced it.

Imagine a community-group draft announcing that registration closes Friday. A useful check finds the organizer’s current registration page and confirms the event year.

A search result about last year’s event cannot establish this year’s deadline, even if the organization and weekday happen to match.

If a source is missing, inaccessible or does not contain the claimed fact, keep that question open. Asking the same system to sound more certain does not supply the missing evidence.

Record the verified link beside the draft so another person can review it before the information is distributed.

For a frightening call, verify through a known number

The Federal Trade Commission warns that scammers can clone a loved one’s voice from a short audio sample. In a family-emergency scheme, the caller may claim to be in serious trouble and ask for money immediately.

The FTC’s advice is to contact the person using a phone number you already know. If that person cannot be reached, try another family member or friend who can help verify the story.

The familiar sound of the caller is not sufficient authentication.

For households with relatives away at college, in military service or traveling, discuss this response while everyone is calm. Keep useful contact information accessible.

A pause to make an independent call is a concrete action someone can remember under pressure. Requests for gift-card numbers, cryptocurrency or a money wire deserve particular scrutiny under the FTC’s warning.

Apply the same check to an invoice or account message

Phishing messages often claim that an account has a problem or that a payment requires attention. The FTC explains that these messages try to obtain passwords, account numbers or other sensitive information, sometimes through a convincing imitation of a known business.

If you have an account with the organization, reach it through a website or phone number you already know is genuine. A link or contact number supplied in the questionable message keeps you inside the sender’s version of the story.

For a small business or volunteer organization, make independent confirmation part of any unexpected payment-change request. A named person, familiar logo or professional tone can be copied.

Review the actual invoice and contact the established person through existing records before changing where money goes.

Account protections still matter. The FTC recommends software updates, backups and multifactor authentication as additional defenses against phishing.

Treat image labels and detection tools as evidence with limits

NIST’s report on synthetic content examines approaches such as provenance records, watermarks and detection tools. Provenance concerns where content came from and how it was handled.

Detection tools try to determine whether material was generated or manipulated.

These approaches answer narrower questions than whether a story is accurate. Knowing a photograph’s origin does not establish that every caption attached to it is true.

An image can be authentic while a repost gives it the wrong place or date.

For a dramatic image shared in a neighborhood group, look for the original publisher and contemporaneous reporting about the claimed event. Preserve the distinction between what the image shows and what the accompanying text asserts.

A detector score can inform a check, but it should not become the sole basis for accusing a person of deception.

If something has already been sent, move to recovery

If you paid a suspected scammer, the FTC says to contact the bank, card issuer, payment app or other company used to send the money immediately and ask whether the transaction can be reversed. Recovery is not guaranteed, but a prompt report gives the responsible company an opportunity to act.

Keep receipts, account notices and the original messages. If a password was disclosed, change it through the genuine account service; also change it on other accounts where it was reused.

The FTC directs people dealing with stolen personal information to IdentityTheft.gov for appropriate next steps.

Report the scam at ReportFraud.ftc.gov. You do not need to prove that AI was involved before reporting what happened.

The useful facts are the request, the contact details, the payment or information provided and the sequence of events.