Practical Task
Something doesn’t match — document the discrepancy
Turn screenshots, notices, receipts, and provider rules into a calm evidence packet before you contact support — without guessing the hidden cause.
What this helps you do
Anyone dealing with a provider, platform, order, account, score, claim, or record that does not appear to match the evidence they can see.
What you need
- The result, decision, score, status, or record that seems wrong or incomplete
- The smallest set of screenshots, notices, receipts, messages, or timestamps that show what you can actually prove
- Any provider rule or help text you already have
- The official review or support route, if it is already shown in your material
What this Task does
Turn scattered evidence into a review packet you can actually inspect.- The notice, score, rejection, or result
- Screenshots, files, receipts, or timestamps
- Relevant provider rules or support replies
- Known facts tied to evidence
- The narrow mismatch
- What is still unknown or missing
- A neutral review request
- Correct anything the AI overstated
- Remove sensitive details you do not need
- Send or escalate through an official route
The AI organizes your side of the record. It does not gain access to the provider’s hidden system.
Do it step by step
- 1
Remove account numbers, IDs, credentials, medical details, and other personal information the review does not need.
- 2
Give AI the result you are questioning plus the smallest useful evidence set. Ask it to separate what happened, what your evidence supports, what the provider explicitly says, and what is still unknown.
- 3
Build an evidence index so every important factual statement points back to a screenshot, document, receipt, message, or timestamp instead of becoming an uncited story.
- 4
Describe the mismatch neutrally. Do not treat 'not shown' as 'did not happen'. Do not let AI decide that the provider acted in bad faith, infer a hidden algorithm, or turn timing into proof of causation.
- 5
Have AI draft a short review request you can read before sending. If the material contains an official review route, include it; otherwise mark the route as still needing verification instead of inventing one.
Use this in your AI assistant
Starter instructions
Copy the starter below, then paste it into ChatGPT, Copilot, Gemini, Claude, or whichever capable AI assistant you use. If this Task calls for photos or files, attach them in that AI assistant too — not on this page.
Help me document a mismatch between what a company or system says and the evidence I have. Use only what I provide. Do not guess the hidden reason, intent, algorithm, policy violation, fraud, discrimination, negligence, or whether I am definitely right. Do not promise a reversal or correction. Give me a simple result first with: 1) what happened, 2) what my evidence actually shows, 3) what does not appear to match, 4) what I still cannot know from this evidence, 5) a calm message I can review and send, and 6) the official review route only if it appears in the material I gave you. Then add a "Go deeper" section with an evidence index, the provider-stated rule, useful missing evidence to request, and exact source/date references. End by stating what the evidence does not prove. Evidence: [PASTE OR UPLOAD]
Helpful checkWhat a good result looks like
You can see what the records actually prove, what only appears inconsistent, what remains unknown, and exactly what to ask the provider to review — without making the accusation stronger than the evidence.
Tips and watch-outs
- A screenshot may show only part of the provider’s internal record. Preserve that limitation when you summarize what happened.
- Do not let AI invent a support path. If the official route is not in your material, verify it from the provider’s current official site or app before acting.
- For legal rights, employment actions, insurance coverage, immigration, medical decisions, credit, government eligibility, or safety-critical outcomes, use the packet to organize facts and questions — not as a definitive ruling.
OptionalGo deeper when you want to
Build a chronology when the mismatch spans several records
A screenshot can establish one state without explaining how the system got there. When the problem spans messages, transactions, status changes, or repeated support contacts, put the evidence in time order before deciding what appears inconsistent.
Keep each row tied to its source. The chronology should make it easier to see the transition that needs review without filling gaps with a plausible story about what the provider's system must have done.
- Observed state or event → source + timestamp.
- Provider-stated rule → exact supplied help/policy text.
- User-reported claim → label it as reported, not independently established.
- Unknown transition or cause → leave it unresolved until authoritative evidence exists.
Label the kind of evidence before drawing a conclusion
The strongest discrepancy packets separate observable facts from provider rules, calculations, attributed statements, and inferences. That prevents a reasonable concern from quietly turning into a factual accusation during summarization.
A useful assistant can say that two records appear inconsistent while still preserving the possibility that an unseen rule, additional record, or different system state explains the difference.
Evidence-safe wording
Observed: the notice says the full name is not visible. Observed: the supplied document visibly contains the name at the top. Unknown: how the provider's review system processed that submission. Request: please manually review the original file and confirm the reason for rejection.
Add new support replies as evidence instead of rewriting history
When support answers, preserve that response as another dated source. Do not replace the earlier record with the new explanation or pretend the explanation was known from the beginning.
If the provider corrects the record, keep the original discrepancy and the correction outcome separate. If the provider does not correct it, the packet should still show what was established, what explanation was given, and what remains unresolved.