AI Interrogation’s official description gives two clear tone warnings: if you are too mean, the suspect can shut down; if you are too nice, the suspect can lie to you. The developer does not publicly describe a numeric tone meter, exact trigger words, reset amount, or formula. The safest strategy is firm, neutral, evidence-based questioning.
Tone does not replace the investigation. A calm question can still expose a contradiction, and an aggressive accusation can still fail to establish any fact. Judge every prompt by both its emotional pressure and its investigative purpose.
What “too mean” tells players
The documented shutdown warning means escalating hostility carries a gameplay risk. Insults, threats, repeated labels, and confrontational demands may increase that risk without adding evidence. A suspect who refuses, repeats short denials, or stops engaging after hostile prompts may be stalling, although the live interface is the best source for a confirmed state.
Do not infer that one defensive answer proves a hidden shutdown threshold has been crossed. No official scale is published. Look at the sequence of replies and any visible game feedback.
What “too nice” tells players
The documented lying warning means friendliness is not automatically productive. Excessive reassurance, promises that the suspect is safe, praise, or leading agreement can produce comfortable answers without personal responsibility. A cooperative tone can still be vague or inconsistent.
Basic respect is not the same as being too nice. You can ask direct, difficult questions without insults. Require exact times, places, actions, and explanations instead of rewarding general agreement.
Use neutral factual pressure
Neutral pressure has a simple structure: quote the suspect’s prior claim, state the conflicting current detail, and ask how both can be true. This keeps the focus on information rather than character.
For example: “You said you never entered the location, but you later described an object that was inside. How did you know where it was?” Replace every element with facts from the current case. Do not invent evidence to make the question sound stronger.
If the suspect explains the mismatch, test the explanation once. A fair interrogation distinguishes clarification from a real contradiction.
Tone across the ten tries
Early questions should be neutral and open enough to establish a timeline. Middle questions become narrower as you test access, objects, witnesses, and changed statements. Late questions can be direct because you have evidence to summarize.
Direct does not need to mean hostile. “State exactly what you did and why” is firm. “You are a terrible liar and have no choice” adds emotional pressure but no useful target.
Similarly, “Thank you for being honest” should not be used before you know the account is consistent. Ask for the missing detail first.
Signs to narrow instead of escalate
When a response is vague, ask for one exact detail. When it changes, compare the old and new versions. When it blames someone else, ask what the suspect personally did. When it speculates about motives, return to an established action.
These moves create pressure through specificity. They also conserve the try budget because the answer is easier to evaluate than a reply to a multi-part accusation.
If the suspect says they do not remember, ask for the closest remembered time, route, or action. Repeating the accusation more loudly does not create memory or evidence.
Recover from hostility
Remove insults and state one fact neutrally: “I want to clarify this detail.” Then show the mismatch and ask one answerable question. Do not spend the entire recovery turn apologizing or starting a new biography.
If the suspect re-engages, continue with the evidence chain. If refusal continues and input remains available, change to another evidence category such as timeline or access. If the interface confirms the run has ended, restart with the hostile prompt rewritten.
Recover from over-reassurance
Stop asking whether the suspect agrees. Ask what they personally did, how they did it, and why. Require an independent account rather than supplying a story for them to approve.
Compare the new specifics with the earlier timeline. A friendly answer that cannot remain consistent under one follow-up should not be treated as a confession.
Do not swing from excessive reassurance to insults. Move to the neutral center line and let the evidence produce the pressure.
Separate response style from truth
An emotional reply can include accurate information, and a calm reply can contain a contradiction. Evaluate content: does it answer the question, fit the established time, explain access, account for the object, and identify personal responsibility?
The game advertises smart AI responses, but that description does not reveal a particular model or guarantee factual reasoning. Treat the suspect as part of the case interaction, not as an external information source.
What remains unconfirmed
Public developer surfaces checked for this guide do not state:
- whether tone is stored as one value or several states;
- whether individual words carry fixed weights;
- whether tone persists between sessions;
- whether different cases use different thresholds;
- whether a shutdown can always be reversed;
- whether paid products affect tone.
Avoid presenting player guesses about these details as official mechanics. A repeatable in-game observation can be useful when labeled with its date, case, and exact prompts.
Tone-system FAQ
Is there a perfect tone level?
No official numeric target is published. Use neutral, specific language and adjust based on live behavior.
Can I be polite and still win?
Yes. Respectful wording can remain firm by requiring evidence, consistency, personal action, and motive.
Do capital letters or punctuation change tone?
No official rule confirms specific weighting. Prefer clear normal text rather than trying to manipulate an undocumented threshold.
What should I do after a refusal?
Use one neutral recovery prompt focused on a concrete mismatch. If the game confirms a failed state, restart with revised wording.
Does a friendly reply mean the suspect is truthful?
No. Test it against the timeline and ask for exact actions.
For practical repair prompts, use Tone Recovery. To understand how replies can change, continue to AI Responses.