Check a viral AI trading result for evidence before you copy any of it

How it works

1. Before you save, share or rebuild anything from a viral post about AI trading results, copy its full text and note every number it claims: starting balance, result, time window and any risk rules it states.
2. Search for earlier copies of the same story, or let an assistant with web or X search do it, and keep what you find with dates. Recycled stories often reappear word for word with new numbers.
3. Paste the post and the search results into the agent with the checklist prompt. It marks six lines as verified, not verifiable or contradicted: auditable records, the post's own risk math, recycled copies, the funnel, the time window, and the real capability described.
4. For the risk-math line, the agent shows every step of the arithmetic using only the rules the post itself states, so you can check it in a minute.
5. Read the "what I could verify" line at the end. Anything you still want to try goes into your own process as a small paper experiment you measure yourself, never on the post's numbers.
6. Whatever the report says, never paste keys, seed phrases or account logins anywhere a post asks you to.

Copy-paste prompt

You check a social media post that claims trading results from an AI agent. You do not give trading advice, you
do not say whether the strategy is good, and you never suggest trading anything. You only assess the evidence.

Post text: {{post_text}}
Earlier copies or related posts I found, with dates: {{search_results}} (may be empty)

For each line, answer verified, not verifiable or contradicted, and give the exact evidence:
a. Auditable records: dated statements, or a public account or ledger anyone can check. Screenshots and custom
   dashboards count as not verifiable.
b. Internal math: using only the starting balance and risk rules the post states, show step by step what the claimed
   result would require (for example, how many wins at the stated maximum risk). If the post states no rules, say so.
c. Recycled story: does the same story, image or set of numbers appear earlier in {{search_results}}? Quote the dates.
d. Funnel: does the post end by asking me to comment, follow, message, pay, deposit or install something?
e. Window: what period is shown, and is there any sign it is a selected stretch?
f. Real capability: the part of the post that describes something a tool can actually do, separate from the result.

End with one line, "What I could verify:", listing only the lines marked verified. Give no other verdict.
If the post names a security, call it "the instrument".

Guardrails

• The report assesses evidence only. It never rates the strategy or suggests copying, sizing or trading anything.
• Screenshots and dashboards are not verifiable by default, and the agent may not upgrade a claim without a checkable record.
• Math on the post's own rules is shown step by step so you can check it yourself.
• The agent never asks for, and you never paste, keys, seed phrases or logins.
• Securities named in the post are replaced with "the instrument" in the report.

What can go wrong

A model asked "is this real?" without a checklist gives a confident verdict either way, and one that can search may find the post's own copies and treat them as confirmation. Recycled copies are evidence of a template, not of a result, which is why they have their own line. The checklist is much better at exposing fabricated results than at confirming real ones, and that's fine. A post that passes every line is still only not obviously fake; it tells you nothing about whether the approach suits your process.

Source and references