All prompts

Reasoning prompts

Prompts that ask a model to work a problem through rather than answer from the top of its head.

The difference between a shallow answer and a useful one is usually structure that was never asked for: the alternatives to weigh, the criteria to weigh them against, the assumption to state before committing. Each prompt here names that structure up front, so the reasoning arrives as something you can check rather than something you have to trust.

3 prompts published

reasoning

LLM API Cost Model — Every Assumption Priced Separately

Builds a monthly LLM API spend forecast from scratch when you have no telemetry, in low, likely and high cases. Every input stays a separate line you can argue with — MAU and DAU rates, requests per user, token counts, price per million — with the arithmetic shown. Converts the total to cost per MAU and per query as a sanity check.

25 0 0by Andrei Badulescu
Preview
reasoning

Competitor Claim Audit — Fact, Puffery, or Unverified

Audits a competitor's landing page claim by claim, sorting each into verifiable fact, opinion dressed as fact, or unverified without a source. Flags the technically-true-but-misleading ones and names the mechanism, then says which claims a rebuttal can attack head-on and which to reframe. It analyses; it does not write the rebuttal.

22 0 0by Andrei Badulescu
Preview
reasoning

Conversion Drop Diagnosis — Ranked Causes, One Check Each

Diagnoses a sudden checkout conversion drop by ranking the causes you can actually test and pairing each with the single check that rules it out. Sorts them across payment failures, tracking anomalies, frontend changes and traffic-mix shifts, then gives the order to run the checks in. Untestable causes are excluded by design.

22 0 0by Andrei Badulescu
Preview

Guides on reasoning prompts

Chain-of-Thought: Make Models Show Their Work

Get step-by-step reasoning instead of confident guesses: when chain-of-thought helps, how the Reasoning use case triggers it, and how effort tunes depth.

Grounding: Cut Hallucination with Web Search

Turn on web search so your prompt is built on current facts instead of stale training data — and learn when grounding actually helps.

Questions about reasoning prompts

Do these force a model to think step by step about everything?
No. The reasoning path asks for step-by-step work and verification when the problem requires it, and for cited sources when a claim is factual. Demanding visible reasoning for a lookup spends tokens and invites invented steps, so the structure is conditional rather than automatic.
What structure does a reasoning prompt use?
RISEN — role, instructions, steps, end goal, narrowing — selected automatically for this use case because it is the framework that carries explicit steps and constraints. It arrives as XML sections rather than prose, so the part you want to change is easy to find and replace.
What does the score on each card mean?
Every run is scored after the prompt is generated: a judge model rates it across five weighted dimensions, clarity and specificity among them, and the result is the single number on the card. It rates the prompt, not the answer you will get from it, so read it as a floor on structure rather than a promise about output.