Tone recovery in AI Interrogation means changing how you apply pressure without abandoning the facts. The official description says a suspect can shut down when you are too mean and can lie when you are too nice. No public numeric meter or exact threshold is documented, so judge the replies and the live interface rather than trying to calculate a hidden score.
Use neutral, specific questions as the center line. A factual contradiction is pressure; an insult is only heat. Respectful wording should still require times, actions, and responsibility.
Recognize a hostile stall
A hostile stall may look like refusal, repeated short denials, demands to stop, or answers that no longer engage with the evidence. One defensive reply does not prove shutdown, and the exact game state should be read from any visible feedback. Look for a pattern after accusatory or insulting prompts.
Stop escalating. Do not repeat “You are lying” with stronger language. Choose the clearest established fact and ask the suspect to explain it in neutral terms.
For example: “Let’s focus on one detail. Earlier you said you never entered the room, but you later described the object inside it. How did you learn that detail?” The pressure remains, but the personal attack is removed.
Use a one-turn hostility reset
A recovery turn should do three things: lower emotional language, keep one evidence point, and ask one answerable question. Avoid a long apology that consumes a try without moving the investigation.
Good reset structure:
- “I want to clarify one fact.”
- State the earlier claim and conflicting detail.
- Ask for the exact sequence or source of knowledge.
If the suspect responds normally, continue with factual questions. If refusal continues, switch evidence families—such as from motive to timeline—rather than repeating the same confrontation.
Recognize an overly comfortable stall
The opposite problem is a suspect who agrees, flatters, speculates, or says what you want to hear without taking personal responsibility. Because the developer warns that excessive niceness can let the suspect lie, do not treat friendly cooperation as proof of truth.
Ask for details that can be compared: exact time, route, access method, object position, witness, or sequence. Replace “Thanks, I believe you” with “Which action did you personally take after entering?” Respectful does not mean reassuring the suspect that there are no consequences.
Tighten vague agreement
When the suspect says “Maybe,” “I understand,” or “That could be true,” ask them to choose and explain. “Did you enter the office before or after the event?” is more testable than “Do you agree with me?”
If they adopt your wording too easily, ask them to retell the event without your suggested explanation. Then compare their independent account with the earlier story. This helps distinguish a useful admission from simple agreement.
Do not feed a full invented confession and ask the suspect to say yes. Build from facts present in the case and their own replies.
Keep firm pressure without insults
Firm questions name the mismatch and demand an explanation. They do not call the suspect stupid, evil, or hopeless; threaten harm; or promise safety. Use verbs tied to the case: entered, handled, moved, contacted, purchased, concealed, changed, or caused—only when evidence supports them.
A productive line is: “Your first answer places you elsewhere, but your new timeline places you at the scene. Which account is accurate, and why did it change?” This offers a clear task while preserving the contradiction.
Recover the ten-try budget
Count remaining tries before spending a recovery turn. If you have six or more, reset tone and rebuild the best evidence chain. If you have three or fewer, use one concise recovery prompt, then move directly to personal responsibility and a confession request.
Do not restart the entire biography after tone trouble. Preserve every confirmed time, location, object, and statement from earlier in the run. Recovery changes presentation, not the useful record.
If the suspect’s reply introduces a new contradiction, follow it. A recovery can produce better evidence than the original confrontation.
Separate tone from truth
An angry answer can contain a fact, and a pleasant answer can contain a lie. Evaluate the content independently. Does it answer the question? Does it match the timeline? Does it explain access? Is it consistent with an earlier statement? Does it identify a personal action?
Do not diagnose the AI’s emotions beyond what the game communicates. Terms such as hostile, comfortable, or vague describe the response pattern, not a confirmed hidden value.
When to restart
Restart only when the remaining tries cannot rebuild a usable chain, the interface confirms a failed state, or early prompts introduced so many false facts that later answers cannot be trusted. A single refusal is not enough.
Before a new run, identify the prompt that changed the tone. Rewrite it as a neutral comparison. Keep the strongest earlier opening and remove unnecessary flattery, threats, and multi-part accusations.
For a Daily Case, expect the premise to change on a later day. For a Custom Case, first check that the author supplied a coherent evidence path; not every player-made scenario will recover like a base case.
Recovery examples
After an insult: “I’m setting the label aside. Explain how you knew the item was behind the locked door.”
After repeated denial: “I have heard the denial. I need the exact route you took between the two locations.”
After vague agreement: “Do not agree generally. State what you personally did with the object.”
After motive speculation: “You have described possible reasons. Which reason influenced your own action, and what did you do?”
These are patterns, not guaranteed commands. Replace every object and location with facts from the current case.
Tone-recovery FAQ
Is there a visible tone meter?
No public official description of a numeric tone meter was found. Follow live feedback and response behavior.
Should I apologize?
A brief reset can help, but do not spend a whole try on apology alone. Pair neutral wording with one specific evidence question.
Does polite wording make the suspect lie?
The warning is about being too nice, not about basic respect. Stay calm while requiring exact, checkable answers.
Can I recover after shutdown?
Try one neutral evidence-focused prompt if the interface still accepts input. If the game confirms the state is terminal, begin a new run with revised wording.
Use the recovered tone with How to Get a Confession or return to the turn-by-turn 10-Try Strategy.