Holmes’s Methods as Prompts(1)

Sherlockian Way of Thinking

1. Introduction: Question Is New Form of Reasoning

A prompt is a question—but not a mere inquiry. When we converse with a generative AI, we are not only asking what we want to know; we are also, at once, designing the form in which the answer should return and the logic by which that answer should be reached. A prompt, then, is not a casual string of words but an instruction that encodes a structure of thought—a deliberate design. And once we peer into this design process, we discover how strikingly it resembles the logical methods Sherlock Holmes employed.

Holmes was hailed as a genius of detection, yet his true gift lay neither in razor-edged instincts nor in superhuman memory. He began by questioning. “Why is one shoe more worn than the other?” “If the window was fastened, how did the intruder enter?” “Why does the letter mix manuscript and type?” Such questions were not themselves clues; they were starting points for interpreting clues. Rather than first piling up facts, Holmes asked why a given fact existed at all—and in doing so, moved closer to the heart of the matter.

Working with generative AI is no different in essence. AI does not think autonomously or define problems for itself; it predicts the most plausible output from a given input. The “intelligence” you get is therefore bounded by the quality of the prompt you give. Vague prompts yield blurry replies; simple, under-specified prompts produce ordinary results. Conversely, a prompt with crisp logic and structure elicits responses that are more precise and persuasive—just as Holmes pierced the thicket of a case by framing the simplest, most penetrating question.

What is now called “prompt engineering” is not merely a technical trick; it is a way of thinking. In a time when the quality of our questions, rather than the quantity of data or speed of computation, often determines AI performance, we must return “how to ask” to the center of our practice—precisely as Holmes did. Change the question and the answer changes; design the frame of thought and the result changes course.

In this chapter we will examine six methods of reasoning used by Sherlock Holmes and show how each can be translated into a modern strategy for prompt design: deduction, induction, the hypothetico-deductive method, abduction, retroduction, and the method of elimination. These are not merely tools from classic detective fiction; they can be revived as finely tuned strategies for questioning a generative AI. A question is another name for reasoning—and today, that reasoning is reborn in the shape of the prompt.

2. Deduction: When Condition Given, Narrow Answer

Deductive reasoning draws a specific conclusion by applying general principles or rules to a particular situation. Holmes was a master of this method. From scattered scraps at a crime scene, he swiftly recombined conditions and drove the conclusion toward a narrow target.

In A Study in Scarlet, for instance, Holmes infers that the culprit must be over six feet tall from the spacing of footprints and the height of blood-written letters on a wall. The logic is straightforward: longer strides usually imply greater height, and people typically write at roughly eye level. From general premises—“wide stride, tall figure” and “writing at eye level”—he deduces a concrete conclusion for the case at hand.

In prompts, deductive thinking means stating the operative conditions clearly so the model must produce a conclusion that fits only those constraints. Rather than asking for information in the abstract, you give the AI a bounded space in which only certain answers can survive. The more definite the conditions, the narrower the candidate set—and the clearer the response.

Prompt strategy: Lay down the logical ground rules

State the general principle or background.

Specify concrete conditions.

Ask for the conclusion that holds only if all conditions are satisfied.

Use this when:
• multiple constraints interact;

• you need to filter for qualifying cases;

• policy/rule-based judgments are required.

Examples

Career fit: “Recommend five jobs that satisfy all of these: (A) minimal face-to-face interaction, (B) high creative workload, (C) stable salaried position. Explain specifically why each fits.”

→ Prevents scattershot listing; drives the model to check each option against all constraints.

Historical figure: “Find one 20th-century figure who (A) sparked innovation in technology and (B) was also controversial for political speech or action. Explain in three paragraphs.”

→ Forces an intersection of conditions.

Constrained sentence: “Write one sentence that (A) is interrogative, (B) contains the word ‘time,’ and (C) has no more than 15 words.”

→ Even creative generation benefits from clear bounds.

Break-in reconstruction: “Evidence: (A) barefoot prints, (B) broken glass, (C) window locked from inside. Given all three, what is the most plausible entry route? Justify step by step.”

→ The model must propose a scenario that simultaneously satisfies all constraints.

Bottom line

Deductive prompts are the art of narrowing. Holmes liked to say that facts are plentiful but truth is rare; truth is what remains after passing through rigorously set conditions. So too with prompts: the more logical the question, the closer the answer cleaves to truth.