# How I Use AI as My "Junior Analyst" 5 Prompt Templates That Actually Work > Source: > Published: 2026-06-21 21:42:41+00:00 Most data analysts I know are using AI wrong. They type things like "analyze this data" or "write me a SQL query" — and get back generic, surface-level responses. That's like telling a junior analyst "do analysis" with no brief, no context, no expectations. The difference between a useless AI response and one that saves you 3 hours? **The prompt.** I've spent months engineering prompts specifically for data workflows. Here are 5 that consistently deliver production-ready output — and the principles behind why they work. Data cleaning is 60-80% of an analyst's job, but most prompts skip edge cases entirely. This one doesn't: ``` I have a dataset with mixed data quality issues — missing values, inconsistent formats, duplicates, and outliers. Here's a sample: [PASTE 5-10 ROWS] Give me a complete cleaning pipeline in Python (pandas) that: - Detects and reports all data quality issues - Handles missing values intelligently (not just dropna) - Standardizes date/text/number formats - Flags outliers with justification - Outputs a clean dataset + a cleaning report ``` **Why this works:** Instead of "clean my data" (which gets you `df.dropna()` at best), you're telling the AI exactly what quality dimensions to check and what kind of output you expect. "Not just dropna" is the key phrase — it forces the model to think about imputation strategies. Writing SQL is easy. Writing SQL that runs fast on 10M+ rows is an art. This prompt turns AI into your DBA: ``` This SQL query is running too slow against a large table. Optimize it: [PASTE YOUR SLOW QUERY] Explain: - What's causing the slowness (scan type, joins, subqueries) - 3-5 specific optimizations in order of impact - Which indexes I should create - The rewritten optimized query - Estimated improvement for each change ``` **Why this works:** You're not just asking for a rewritten query — you're demanding a diagnosis. The structured output format forces the model to think in terms of query execution plans, not just syntax. "Estimated improvement" makes it quantify its claims. Something broke. A KPI tanked. Your boss wants answers by EOD. This prompt is your investigation partner: ``` [KPI] dropped by [X%] starting on [DATE]. Here's the available data: [DESCRIBE DATA SOURCES] Walk me through a structured root cause analysis: - What to check first, second, third (prioritized) - Segmentation cuts that could reveal the driver - Common pitfalls in this type of investigation - How to present findings to stakeholders - Python/SQL code templates for each diagnostic step ``` **Why this works:** The prioritization is the secret sauce. Most analysts go down rabbit holes. This prompt forces the AI to give you a triage framework, not just "look at the data." Dashboards that impress engineers are useless. Dashboards that impress executives get you promoted. Same data, different framing: ``` Design an executive dashboard for monitoring [BUSINESS AREA]. Key stakeholders care about: [LIST 5-7 KEY METRICS] Provide: - Dashboard layout wireframe (top-level, detail, drill-down views) - Which charts for which metrics (and why) - Color scheme and visual hierarchy - What should be real-time vs daily vs weekly refresh - Alert thresholds for each metric - Plotly/Dash or Streamlit starter code ``` **Why this works:** Most dashboard prompts produce a list of charts. This one produces a design system — visual hierarchy, refresh cadence, alert thresholds. That's what separates a monitoring tool from a decision-making tool. Your analysis is solid. Your presentation is… technical. This prompt bridges the gap: ``` I need to present [TECHNICAL FINDING] to [EXECUTIVE/MARKETING/ENGINEERING]. Translate my technical findings into their language: - The 30-second version (elevator pitch) - The 3-minute version (key insights + recommendation) - What they'll care about (tailored to their priorities) - What they'll push back on (and how to handle it) - One visual that makes the point better than words ``` **Why this works:** Different audiences have different "so what"s. This prompt forces the AI to reframe the same finding for different mental models. The "what they'll push back on" section is pure gold — it prepares you for the actual conversation, not just the presentation. All of these follow the same pattern. I call it the **Genie Formula:** ``` [CONTEXT] I am a [ROLE] working on [PROBLEM]. [INPUT] Here is my data/code/situation: [DETAILS] [TASK] I need you to: [SPECIFIC REQUEST] [FORMAT] Please provide the output as: [STRUCTURE] [CONSTRAINTS] Important constraints: [TECH STACK, PRIVACY, FORMAT] [QUALITY] Double-check for: [EDGE CASES, PERFORMANCE, ACCURACY] ``` The more structure you give, the better the output. AI isn't a magic wand — it's a junior analyst who needs clear instructions. I've compiled my full prompt library — 55 prompts across 8 categories (data cleaning, SQL, Python, visualization, business intelligence, automation, career acceleration) — into a PDF. Each prompt follows the Genie Formula and is engineered for production use, not demo purposes. [Get the full pack → $9.99 on Gumroad](https://geniehq.gumroad.com/l/data-analyst-ai-command-center) What's inside: I'm curious — what's the one prompt structure you've found that consistently produces great results for data work? Drop it in the comments. And if you've never structured your prompts beyond "analyze this" — try the Genie Formula above. The difference will surprise you.