A post shows up in a group chat or a feed: a screenshot, a quote, a photo with a caption, a claim about something that just happened. It has a specific detail that makes it feel real - a named place, a number, a face. Forwarding it takes one tap. Checking it takes longer, so most people do not.
That asymmetry is not an accident of human laziness. A large-scale study of roughly 126,000 news cascades on Twitter, tracked from 2006 to 2017, found that false stories reached far more people, and reached them faster, than true ones - the effect was strongest for political claims, and the researchers traced it to novelty and emotional charge, not bots (Vosoughi, Roy & Aral, “The spread of true and false news online,” Science, 2018). A claim built to spread and a claim built to be accurate are frequently different objects, and the platform gives no visual cue telling you which one you are holding.
This is a short method for tracing a viral claim to where it actually started, using AI to find and organize what exists - not to decide whether the claim is true. That boundary matters because a chatbot asked “is this real” will often answer with the same fluent confidence whether it actually knows or is filling a gap plausibly.
Why a screenshot is not a source
A screenshot proves that someone, somewhere, typed or posted those words. It proves nothing about whether the underlying event happened, whether the quote is real, or whether the image was captured where and when the caption says. Screenshots also strip away the one thing that would let you check: the original post, with its account, timestamp, and replies. By the time a claim reaches you, it may be three or four screenshots removed from anything checkable.
Do not ask a chatbot “is this true” as your first move. That framing invites a plausible-sounding guess. Ask it to help you find the original post, the original account, or the original source instead - a search-and-locate task the model is actually built to help with.
Step 1: Name the checkable claim inside the viral post
Strip the caption, the outrage, and the call to action down to one checkable sentence before you do anything else.
Here is a post I received: [paste it]. Extract the single specific,
checkable factual claim being made, in one plain sentence. Ignore the
framing, the emotional language, and any call to action - I only want
the underlying claim and, if visible, the name of the original poster
or outlet.
A post captioned “You won’t believe what they just did!” usually reduces to something narrower and checkable, like “a specific official made a specific statement on a specific date.” That sentence is what you actually verify.
Step 2: Find the original post, not a description of it
Ask the model to help locate where the claim originated, and be explicit that you want the source, not a summary of what the source probably says.
For this claim: "[claim from step 1]" - help me identify what the
original source is likely to be (a specific news outlet, an official
account, a press release, a court filing) and how I would search for
the original post or article directly. Do not tell me whether the
claim is true or what the source says - I will check that myself.
Chat models without a live search connection cannot browse the current post for you; treat their answer as a research plan, not a result. If your tool has a web-search mode, turn it on for this step so it returns links you can check rather than a source it describes from memory - then open every one yourself, because a returned link can still be dead or point to something that does not carry the claim.
Step 3: Check the original directly
Open the actual original post, article, or account named in step 2. Look for three things: does the original account match the one the screenshot implies, does the date match the event being described, and does the original wording match the version that reached you. A claim can travel through several accurate-sounding intermediaries and still drift from what was actually said - a screenshot of a screenshot is exactly where that drift accumulates.
If you cannot find the original post at all - the account does not exist, the outlet never published it, the quoted event is not covered anywhere else - treat that absence as a strong signal, not a gap to fill in with another AI query. A real, significant event is almost never reported by exactly one unverifiable screenshot.
Step 4: Check whether the claim is being used honestly
A real photo or a real quote can still be presented dishonestly - real but old, real but from a different event, real but missing context that changes its meaning. This is a distinct check from whether the claim is fabricated outright.
Here is what the original source actually shows or says: [paste it].
Help me identify: when this was originally published, what event it
actually documents, and whether the version I received adds any
claim, date, or context that is not present in the original. Do not
tell me whether the overall framing is fair - just show me what
matches and what does not.
Step 5: Decide, and write it down
With the claim, the original source, and the honesty check in hand, make one of three calls: share it because it holds up, share it with the missing context attached, or do not share it because the source could not be verified or the framing has drifted from the original. Writing down “could not verify - held” is usually more useful to the next person in the thread than saying nothing, and it is the habit that actually interrupts a false claim’s cascade rather than joining it.
The full viral claim source-trace worksheet walks through all five steps with space to record the original post, the source chain, and your decision.
Where this fits with the rest of your verification toolkit
This method is deliberately general - it works for a political claim, a natural-disaster photo, or a workplace rumor equally. Two related methods handle situations this one does not cover well: checking a health claim walks through the added step of checking whether a real study still applies to the general population, and checking an AI-generated news summary covers the specific case where the “source” your summary came from is itself a chatbot’s compression of a real article. If the claim arrived inside a live group conversation rather than as a single forward, interrupting a rumor in a group chat covers how to raise the question without starting a fight. And if you are deciding search versus chat as your first move on any factual question, search vs chat for facts covers that decision directly.
Common pitfalls
- Asking “is this true” instead of “where did this originate.” The first invites a guess dressed as an answer; the second keeps the model on a task - locating and organizing - it is actually reliable at.
- Trusting a screenshot as if it were the original. Every layer of screenshotting removes the account, timestamp, and context that would let you check anything.
- Stopping at “I found a similar claim elsewhere.” Other people repeating an unverified claim is not independent confirmation; it is the same cascade the Science study describes.
- Forwarding “just in case people should know.” An unverified claim is not a neutral public service to pass along - it costs everyone downstream the same checking work, and if it is false, actual harm.
- Treating hallucination risk as a footnote. Why AI sometimes gives confident wrong answers explains why a model can describe a plausible-sounding “original source” that does not exist - always open the link yourself.
Try it today
The next time something in a chat or a feed makes you want to hit forward, run it through the five steps first: extract the claim, locate the original, check it directly, test whether the framing has drifted, and write your decision down. Use the viral claim source-trace worksheet to make the habit stick past the first time you try it.
