AI Interrogation Strategies

The strongest AI Interrogation strategy is a funnel: collect a bounded account, lock down a few concrete facts, find one contradiction, and use that contradiction to request the complete truth. You have ten tries in the standard loop, so every prompt should either create a testable claim or test one that already exists.

The official description establishes the two tone hazards. Being too mean can make the suspect shut down; being too nice can let the suspect keep lying. It does not publish a numeric tone meter, safe-word list, or universal threshold. Use pressure from facts, not insults, and use respect without treating the story as true.

Pick a question by its job

A context question defines the event you are discussing. A timeline question pins down order and time. An access question asks how the suspect reached a room, object, person, or location. An evidence question connects the suspect to something visible in the case. A contradiction question compares two claims. A confession question asks for the action, method, and motive after the chain is ready.

Open Best Questions when you need wording for one of those jobs. The examples are structures, not secret commands. Replace generic words such as “incident” or “object” with facts from the current case.

Divide the ten tries

Use tries one through three to establish the suspect’s version. Ask for the event sequence, one exact location or time, and a person or object that can confirm the story. Avoid opening with guilt because the suspect has not yet committed to enough details.

Use tries four through six to test access, evidence, or an explanation that could conflict with the opening. Press one branch at a time. If a response introduces an object, ask when it was handled. If it introduces a witness, ask exactly what the witness can confirm. If it introduces money or motive, connect it to the incident window.

Use tries seven and eight to state the clearest mismatch. Quote the two facts without adding a weak guess. Use tries nine and ten for the confession request and, if needed, one narrower follow-up. The 10-Try Strategy gives recovery branches for runs that reach the middle without a useful contradiction.

Plan the route before opening the terminal

Use the Question Planner when you need a writable view of the whole run. Give each slot a job—gather, test, press, recover, or close—then mark it planned, asked, or skipped as the conversation changes. The ten-row sheet also keeps the main claim, strongest conflict, and missing confession part visible without turning the plan into a fixed script.

Do not fill every row with case details from another player. Prepare the first timeline question, two evidence roles, and one complete close, then leave room for facts you have not seen. If a reply exposes the key contradiction early, skip the weaker prepared questions and move toward responsibility. After a failure, replace the first row that neither added nor tested a useful fact.

Build pressure without hostility

“You said you were home, but the timestamp places your vehicle at the scene. Explain the difference” is firm because it requires one conflict to be resolved. “Stop lying and confess” is hostile but vague. The first prompt produces a fact you can use; the second risks shutdown while giving the suspect no clear evidence to address.

Use a neutral verb such as explain, clarify, account for, or walk me through. Name the exact time, place, object, or statement that conflicts. Keep the prompt short enough that the suspect cannot answer only the least important part.

Move toward a full confession

A confession request works best after the suspect has changed a statement, admitted access, or failed to explain evidence. Summarize the chain in order: original claim, conflicting fact, hidden action. Then ask the suspect to state what they did, how they did it, and why.

Do not add every clue from the case. Weak or unconfirmed details give the response an escape route. The Get a Confession guide shows how to build a three-part final prompt and how to respond when the suspect admits only part of the action.

Recover when the tone goes wrong

If the suspect gives short refusals, remove emotional wording and ask for one concrete detail. You do not need to apologize repeatedly or agree with the story. A useful reset is: “Let’s stay with the timeline. What happened immediately after the time you gave me?”

If the suspect becomes agreeable but keeps avoiding personal action, stop offering reassurance. Ask which action they took, what object they handled, or why their account changed. The Tone Recovery guide separates shutdown, evasion, vague agreement, and a genuinely changed story.

Recovery is not guaranteed. If several replies contain no usable facts and only a few tries remain, restarting may be more efficient than spending the whole budget on a damaged conversation. Record the prompt that caused the change so the next attempt can avoid it.

Adapt the strategy to each mode

Base cases let you reuse a known evidence order, but AI wording can still vary. Daily cases can rotate, so begin with premise mapping rather than a memorized answer list. Custom cases depend on how another player built the evidence chain, so test whether the suspect has enough coherent facts to support a fair confession.

Specific Case 1–5 guides include community-reported objects and contradictions. Use those details only when the current run matches. Case 6 does not have a publicly confirmed name in the checked badge data, so do not force another case’s script onto it.

A compact note format

Write four lines: claim, support, conflict, final ask. “Claim” is the suspect’s central alibi. “Support” is the witness or object they use to defend it. “Conflict” is the timestamp, access path, object, money source, or changed statement that does not fit. “Final ask” is one sentence connecting the conflict to the action.

This format reduces repetition. Before submitting a question, look at the four lines and ask whether the prompt fills a blank or tests an existing line. If it does neither, rewrite it.

Strategy mistakes to avoid

• Asking two or three unrelated questions in one terminal submission.
• Repeating the same accusation with stronger insults instead of adding evidence.
• Treating a friendly answer as true without testing time, place, or access.
• Collecting background through all ten tries and leaving no confession attempt.
• Copying an old-universe transcript without matching the current Place and case.
• Treating a community-reported sequence as an official guaranteed answer.

Strategy FAQ

What is the best first question?

Ask for a bounded timeline around the incident. It produces several facts that can be checked later while keeping the tone neutral.

When should I accuse the suspect?

After you can name a concrete contradiction. Pressure is more useful when the suspect must choose between two incompatible claims.

What if I have no contradiction by try seven?

Stop broad exploration. Choose the strongest time, location, access, object, or witness detail and ask the suspect to explain it precisely.

Is there a guaranteed confession script?

No public source establishes one. Current walkthroughs can provide useful pressure routes, but AI responses and case details can vary.

Continue with Best Questions for prompt structures, fill the Question Planner before a difficult run, or open Interrogation System for the rules behind the strategy.