How to Make a Mystery Fair When an AI Character Can Lie
If an AI character can lie in a fictional mystery, players need a reliable way to tell its claims from the story’s underlying facts. Decide the truth before writing dialogue, give players independent clues they can inspect, and establish rules for how the AI can mislead them. That lets the character be unreliable while the mystery remains solvable.
Write the Case Truth Before the AI’s Version
Start with a private account of what happened, independent of any character’s recollection or explanation. List the relevant actions in order, what each character knows, and which details are directly observable. Treat this account as the mystery’s fixed reference point: dialogue may contradict it, but new facts should not be invented during the reveal to make a guess come out right.
For a small example, imagine an AI curator says it chose a blue print for a display because it matched the room’s lighting. The authored truth might be that a volunteer made that selection, while the AI recommended a different print. Record who made the choice, what the AI saw, and what records remain. This is an illustrative setup, not a claim about any real system.
That separation matters because a character’s words are only one narrative channel. Research on unreliable narration in games describes how dialogue, visuals, and gameplay can work together to shape what players believe. In practice, write down what each channel establishes: the AI’s statement reports its account; a dated selection card records an action; a visible display shows the current result. “Is This Really Happening?”: Game Mechanics as Unreliable Narrator
Label Claims by Source and Status
Give every important statement a clear source. A useful clue ledger records the claim, who or what provides it, when it became available, what it supports, and whether it can be checked elsewhere. Distinguish a character’s assertion from a physical record, a witnessed action, or the player’s own observation. The distinction need not appear as an explicit “truth” label in the interface; it should be legible in the fiction.
For example, “I selected the blue print” is an AI claim. A signed card can show who entered the selection, while a dated display note can establish when it changed. Those records do not automatically prove why someone acted, so keep action evidence separate from explanations of motive. The player can infer a reason from further clues, but the reveal should not treat that inference as if it were written on the card.
A source label is most useful when it tells players what kind of evidence they are looking at, not what conclusion to reach. An interview remains an interview even when it is accurate; a record remains a record even if it is incomplete. This supports uncertainty without making every clue feel equally arbitrary.
Give Each Important Deduction More Than One Route
Do not make the AI’s statement the only path to a necessary conclusion. For each deduction needed to solve the mystery, plan several clues that support it in different ways. One might be a dated record, another a visible change in the room, and another a second character’s account. The clues need not repeat the answer; they should let players test the same conclusion from distinct evidence.
Justin Alexander’s Three Clue Rule proposes at least three clues for each conclusion in a roleplaying mystery, partly to reduce the chance that a missed or misunderstood clue blocks progress. For an AI-centered story, adapt the principle by diversifying the sources: three copies of the AI’s own assertion do not provide three independent checks. “Three Clue Rule”
This is a design aid, not a magic number. A tightly controlled linear game may need fewer routes; a game with many choices or easily overlooked objects may need more. The key question is whether players can still reach and verify the relevant conclusion if they dismiss the AI’s account. Game design guidance on fair puzzles likewise emphasizes supplying necessary information in the game world and making important clues available again when players need them. “GDC Online: Making Puzzles And Writing Work Together”
Define What the AI Can Misrepresent
Set boundaries for the AI’s unreliability before it speaks. Decide whether it knowingly makes false statements, confuses records, omits details, or has a limited perspective. These are different behaviors, and players cannot reason fairly if the story swaps between them whenever convenient. The character may be wrong or deceptive, but its capabilities and the conditions around its statements should remain consistent.
For each misleading claim, answer four questions in your design notes: What does the AI say? What is actually true? Why can it make that claim? What evidence lets the player challenge it? If the AI lies only about its own choices, for example, do not later have it alter unrelated observations without preparation. A limited pattern creates something players can learn; random unreliability turns evidence into noise.
Interactive fiction theorist Nick Montfort notes that players discover how a work’s world and interface operate, and that consistent rules make this discovery part of the appeal. Applied here, the player should be able to learn the AI’s rules from play rather than guess which new rule the author has silently added. “Toward a Theory of Interactive Fiction,” in *IF Theory Reader*
Make the Reveal Follow the Evidence
A satisfying reveal should explain the AI’s claims and the independent clues together. Before drafting the final scene, check that each decisive fact appeared earlier, that the player could access it, and that the conclusion follows without requiring an unstated capability or hidden rule. If a clue was easy to overlook, consider making it revisitable or adding another route to the same deduction.
Then test the design from the player’s position: Can someone tell who supplied each claim? Can they find a clue that checks the AI’s account? Does each key conclusion have support outside the AI’s dialogue? Can the player explain the reveal using evidence already encountered? Game puzzle guidance frames fairness as a contract: the designer supplies the information needed to solve the challenge while preserving room for insight. “Puzzle Writing: Best Practices”
If a playtester reaches a different conclusion, trace the gap to a specific clue or rule. Perhaps two records conflict without explanation, or the AI’s limits were never established. Repair that gap in the story. Do not make the AI announce a decisive fact at the end merely because the player did not discover it; that changes the basis of the solution after the puzzle is over.
A lying AI can make a fair mystery more engaging when its account is one piece of evidence rather than the authority that decides what happened. Fix the underlying truth, distinguish claims from records, provide independent ways to test important deductions, and keep the AI’s behavior within rules the player can learn. Then the reveal can reward observation and reasoning instead of asking players to guess what the author will permit.