AI Troubleshooting Prompt Builder

What it does. Turn your field readings into a structured AI troubleshooting prompt. Six inputs, one copy-ready prompt, plus a strength meter that grades it.

Detailed Overview

The AI Troubleshooting Prompt Builder takes the six facts you already gather on every service call and assembles them into a clean prompt you can paste into ChatGPT, Gemini, or a manufacturer app. It exists because the tool is rarely the problem when AI gives a useless answer. The prompt is. A tech who types “furnace won’t heat” gets a textbook. A tech who hands over the model, the codes, the readings, and what was already checked gets a ranked, testable answer. This builder makes the second kind of prompt the default.

Purpose

Most field techs try AI once, type a vague question, get a generic ten-item list, and conclude the technology is useless for real work. The miss is that an AI model predicts an answer from what you give it. Thin input gives a thin answer. The builder solves that by enforcing a structure: the same structure you would use briefing a senior tech on the phone. It also pushes two habits that separate a second opinion from a guess. It asks the model for a ranked list with a confirming test for each cause, which marches you back to the meter instead of a parts swap, and it tells the model to cite and weight its sources so a manufacturer spec is not treated the same as a five year old forum post.

When and Where to Use It

• Stuck on a callback. Build a prompt from your readings before you escalate, then take the ranked causes to the meter.
• Apprentice or first-year tech. Use it to check yourself before calling the boss, while still confirming everything with your own measurements.
• Office or training. Owners and trainers can show new hires what a useful AI question looks like versus a lazy one.
• Any system type. Furnace, AC, heat pump, mini-split, RTU, boiler, or refrigeration. The structure holds across all of them.

How to Use It

1. Pick the system type and enter the make and model number.
2. Describe the exact symptom, then enter any codes, lights, or alarms verbatim.
3. Add your measured readings, what you already checked or ruled out, and the operating conditions.
4. Leave the two toggles on to request a ranked list with confirming tests and to make the model cite its sources.
5. Watch the prompt-strength meter. Thin means you are in the “furnace won’t heat” trap. Strong means it reads like a clean service handoff.
6. Press Copy prompt and paste it into your AI tool of choice. Load example fills a realistic no-cool call so you can see what a strong prompt looks like.

Outputs

• A copy-ready prompt. A structured, plain-language prompt assembled from your inputs, editable in the box before you copy it.
• A prompt-strength meter. A Thin, Decent, or Strong grade based on how many of the six fields you filled, reinforcing that better input drives better output.
• A verify reminder. A standing note that AI points you at which sense and meter to use but never confirms the fix.

Context: Where This Tool Lives in HKIA’s Content

The tool was built to accompany the following HKIA content. Specifically:

• “Prompting a Service Call: How to Ask AI a Troubleshooting Question That Actually Helps”. argues that a good AI prompt is just a structured service report, and lays out the six-part field prompt this tool generates.
• “The Limit Circuit Defective Trap: Why an AI Answer Isn’t a Diagnosis”. argues that an AI answer is a starting point, not a verdict, which is why this tool builds in the ranked-list-plus-confirming-test request.

The tool pairs naturally with a future Furnace Temperature Rise and Airflow Calculator (which produces the kind of readings this prompt asks for) and an AI Answer Trust Scorecard (which grades the answer the prompt returns). All three reinforce the same loop: measure, prompt, verify.

Math and Logic

• Prompt assembly: the builder concatenates eight labeled lines (equipment, symptom, codes, readings, already checked, conditions, the ranked-list request, the source-citation request) into one block. There is no numeric math. The logic is templating, with empty optional fields replaced by bracketed reminders rather than dropped, so the user sees what is missing.
• Strength score: counts how many of the six substantive fields (model, symptom, codes, readings, checked, conditions) are filled. Zero to two filled, or both model and symptom empty, reads Thin. Three to four reads Decent. Five to six reads Strong.
• No persistence: all state lives in browser memory. Nothing is stored or transmitted. The copy action uses the clipboard API with a select-and-copy fallback.

Limitations

• It writes the question, not the answer. The quality of the AI response still depends on the model you paste it into and the data you entered.
• It does not validate your readings. If you enter an impossible pressure or a wrong model number, the prompt carries that error forward. The tool assumes competent field input.
• It is not a substitute for fundamentals. The builder is a structuring aid. Confirming the diagnosis with your meter and senses is still the job.

Sources Used

• Anthropic Economic Index. Anthropic, 2026. Basis for the principle that prompt sophistication tracks response sophistication, which the six-part structure is designed to exploit.
• Omar et al., adversarial hallucination in clinical decision support. Communications Medicine, 2025. Basis for the ranked-list-and-confirm requirement, since models will run with a planted wrong premise in a majority of cases.
• Retrieval-augmented generation accuracy study. Journal of Orthopaedic Research, 2025. Basis for the source-citation toggle, since grounding a model in real source documents materially improves accuracy.