AI Prompts for Data Analysts gives US data analysts, BI analysts, and analytics engineers copy-paste prompts for the work that surrounds the SQL — the intake questions, the dashboard narratives, the KPI definitions, and the executive summaries that turn a query result into a business decision. These are for the parts of the job where clear writing matters as much as clean joins.
Every prompt is written for how analytics actually works in US product and BI teams: stakeholders who ask fuzzy questions, dashboards that need a narrative, A/B test writeups that survive review, and exec one-pagers that lead with the answer instead of the methodology. Fill in the brackets with your actual data model, KPI names, and stakeholder audience, then check the output against what your warehouse actually contains.
Do not paste raw customer data, unreleased revenue figures, PII from your warehouse, or the contents of tables under a data-sharing agreement into a public AI tool. For any prompt that involves live data, use your BI tool's built-in AI, an enterprise AI approved by your data governance team, or write schemas and column names as generic placeholders.
AI Prompts for Data Analysts gives US data analysts, BI analysts, and analytics engineers copy-paste prompts for the work that surrounds the SQL — the intake questions, the dashboard narratives, the KPI definitions, and the executive summaries that turn a query result into a business decision. These are for the parts of the job where clear writing matters as much as clean joins.
Every prompt is written for how analytics actually works in US product and BI teams: stakeholders who ask fuzzy questions, dashboards that need a narrative, A/B test writeups that survive review, and exec one-pagers that lead with the answer instead of the methodology. Fill in the brackets with your actual data model, KPI names, and stakeholder audience, then check the output against what your warehouse actually contains.
Do not paste raw customer data, unreleased revenue figures, PII from your warehouse, or the contents of tables under a data-sharing agreement into a public AI tool. For any prompt that involves live data, use your BI tool's built-in AI, an enterprise AI approved by your data governance team, or write schemas and column names as generic placeholders.
Guides, tips, and deep dives for this prompt category
Create stunning Studio Ghibli-style AI art with 50 free prompts for ChatGPT. Magical landscapes, characters, food scenes, and cozy interiors in Miyazaki style.
Read moreCollectionCreate stunning Studio Ghibli-inspired images using ChatGPT GPT-4o. 50 free prompts for Ghibli art, landscapes, characters, and scenes.
Read moreCopy any prompt below, paste into ChatGPT, Claude, Gemini, or Copilot, and fill in the placeholders in [brackets].
Act as a US senior data analyst. Write a SQL query to answer: [business question, e.g., "what's the 90-day retention rate of users who signed up in Q2, broken down by acquisition channel?"]. My warehouse is [PostgreSQL / BigQuery / Snowflake]. Relevant tables: [table_name (grain: one row per __, key columns: __)] and [table_name (grain: __, key columns: __)]. Use CTEs, not nested subqueries. Add inline comments explaining any non-obvious logic. Flag any assumption I need to verify.
Act as a US BI analyst. Write a dashboard narrative for a [KPI name] dashboard viewed by [audience — e.g., the growth team, the CFO, the sales VP]. Cover: what this dashboard answers in one sentence, the top KPI and how to read it, 2 to 3 supporting metrics and what they add, common questions this dashboard should preempt, and what this dashboard does not measure. Written to sit above the dashboard as context, not to replace it.
Act as a US data analyst handling a stakeholder request. A [role — e.g., product manager, marketing lead] just asked me [paste the vague question they sent]. Write a stakeholder question intake response that: acknowledges the request, restates my understanding of what they actually need to decide, asks 3 to 5 clarifying questions (audience, timeframe, cut of data, decision this informs, deadline), and proposes a scoped deliverable. Friendly, not gatekeeping.
Act as a US data quality engineer. Write a data quality check summary for [dataset / pipeline name]. Cover: freshness (last update, expected cadence), completeness (row counts vs. expected, null rates on key columns), validity (type mismatches, out-of-range values, referential integrity failures), consistency (row counts across joined tables), and anomalies (unexpected spikes or drops). Format: table of check name, result, threshold, and status. Written for both an analyst and the data engineering owner.
Act as a US senior data analyst. Write an insight writeup for [finding — e.g., checkout completion dropped 8% week-over-week]. Structure: The finding in one sentence at the top, the supporting number with its baseline, the segment or driver most responsible for the change, hypothesis for why (labeled as hypothesis, not fact), what would confirm or refute the hypothesis, and what this analysis does NOT tell us. Under 300 words. No throat-clearing intro.
Act as a US analytics engineer. Write a KPI definition document for [metric name — e.g., Weekly Active Users, MRR, NPS]. Sections: Business definition (what this measures and why it matters, plain English), Formula (exact calculation), Data source (tables and columns), Filters and exclusions (test accounts, internal users, refunded orders, whatever applies), Refresh cadence, Known limitations, Owner. Written so an analyst joining the team next week can compute this without asking anyone.
Act as a US data analyst. Write a cohort analysis explanation for [cohort type — e.g., new signups by month, first-purchase customers by acquisition channel]. Cover: what a cohort is (2 sentences for a non-analyst audience), how this cohort is defined (start event, cohort assignment rule), what metric we're tracking across the cohort's lifetime, how to read the resulting chart or table, and one insight this view surfaces that a rolled-up metric hides.
Act as a US data analyst diagnosing a funnel drop-off. The funnel is [funnel steps in order]. The step with the largest drop is [step name] at [conversion rate]. Write a funnel drop-off diagnosis covering: baseline vs. current conversion, segment cuts to investigate (device, channel, cohort, geography, product variant), 3 hypotheses ranked by likelihood, the data needed to test each hypothesis, and a recommended next step. Analytical, no jumping to conclusions.
Act as a US data analyst. Write an A/B test result summary for an experiment on [surface / feature]. Structure: The test (control vs. variant, one sentence each), Primary metric result (delta, confidence interval, p-value, sample size), Guardrail metric results, Segment breakdowns (if any showed materially different behavior), Interpretation (labeled as interpretation, not fact), Recommendation (ship / kill / iterate / re-test), and What we'd want to learn next. Skip the "we ran an experiment" preamble.
Act as a US senior data analyst. Write an executive one-pager for [analysis or initiative] to be read by a VP or C-level in under 3 minutes. Structure: The answer / finding in one sentence at the top (bold), Key numbers (3 max, with baseline for context), What's changed and why, Business implication, Recommendation with the one reason it wins, Risks or caveats, What we need to decide. One page. No methodology in the body — footnote it.
Act as a US analytics engineer. Write technical documentation for [data pipeline name]. Sections: Purpose (what business question this pipeline supports), Sources (upstream tables and their owners), Transformations (key logic, in plain English before code snippets), Output (destination tables, grain, refresh schedule), Dependencies, Known issues and workarounds, On-call runbook (what to do if it fails at 3 AM), Owner and backup owner. Written for both the current team and whoever inherits it in 2 years.
Act as a US BI analyst. Write a metric change alert email for [KPI] which moved [direction, magnitude] over [time window]. Cover: what changed (the metric, the delta, the window), what's contributing (segment or driver breakdown), whether this is likely real or a data artifact (freshness check, anomaly history), what we're doing next (investigating, monitoring, action), and who to contact. Under 200 words. Direct tone — this goes to a Slack channel, not a formal report.
Act as a US senior data analyst preparing a quarterly business review. Write a QBR deck outline for [business area — e.g., Growth, Product, Revenue]. Slides: Executive summary (1 slide, the 3 things that matter), KPI scorecard vs. target, What worked (1 to 2 examples with impact numbers), What didn't (honest — 1 example), Key learnings from experiments, Focus areas for next quarter, Metrics we're watching, Open questions for leadership. Each slide gets a headline that states the takeaway, not just the topic.
Act as a US data analyst triaging an ad-hoc analysis request. I received [paste request]. Write an ad-hoc analysis triage response covering: my read on what they're actually trying to decide, whether an existing dashboard or query can answer this (name it if so), estimated effort if new analysis is needed (S/M/L), other requests this would push back if I prioritize it, and a recommended next step (existing resource / scoped analysis / needs a meeting to define). Direct but collaborative.
Understanding the building blocks lets you adapt any prompt to your own creative direction.
Tell the AI who the output is for and what real workplace situation it should support.
Act as a federal program analyst preparing a plain-language memo for agency leadership.Name the exact deliverable: email, memo, checklist, SOP, meeting recap, training note, or status update.
Format the answer as a one-page briefing with bullets, risks, and next actions.Specify whether the output should sound official, executive-ready, plain-language, or employee-friendly.
Use a professional, neutral, public-sector tone suitable for a US agency audience.For government, HR, finance, healthcare, legal, and compliance workflows, accuracy guardrails matter more than clever wording.
Use only the facts below, flag assumptions, and include a section for items that need verification.Ask the model to surface uncertainty so the user can verify sensitive or official information before using it.
Before finalizing, list compliance risks, missing details, and any claims that need human review.Tested on this prompt category as of mid-2026. Ratings reflect quality for AI Prompts for Data Analysts specifically.
| Model | Best for | Rating |
|---|---|---|
| ChatGPT (GPT-4o / GPT-5) | Everyday drafting and summaries | |
| Claude Sonnet 4.5 | Long documents and policy | |
| Gemini 2.5 Pro | Grounded in Google workspace | |
| Copilot (M365) | Office 365 integration | |
| Perplexity | Answers with citations |
Ratings reflect suitability for this category. Free tiers available on all listed models. Last tested May 2026 by PromptSpace editors.
Not without review. AI-generated SQL frequently references columns that sound right but don't exist in your schema, and often gets joins wrong for anything beyond a two-table aggregation. Use it to draft the structure, then check every table and column against your data catalog and run the query on a limit before trusting the output.
All of them, but you have to specify. Date functions, window function syntax, JSON handling, and string functions vary between PostgreSQL, BigQuery, Snowflake, Redshift, and DuckDB enough that a generic SQL prompt often produces something you have to translate. Add "my warehouse is BigQuery" (or your dialect) to every SQL prompt.
Describe your schema with generic placeholders (fact_orders, dim_customer) rather than pasting real tables. For actual data, use your BI tool's built-in AI (Hex, Mode, Looker have these) or an enterprise AI approved by your data governance team. Never paste PII, unreleased financials, or data under a customer contract into a public AI tool.
For the structure, yes. For the interpretation, edit carefully. AI drafts default to overclaiming — "the variant caused a 5% lift" — from correlational evidence. Rewrite every causal statement into what the data shows (a change happened) vs. what caused it (a hypothesis). Add sample size, confidence interval, and guardrail results explicitly.
Anything requiring judgment about your business — which KPI to pick, which segment matters, whether a result is real or noise, whether to ship an experiment. Also: any task that needs live production data in the prompt without an enterprise AI in place. Use AI for the writing around the analysis, not to replace the analyst's read of what the numbers mean.
Learn the basics of creating stunning AI-generated images using prompts from our library.
GuideDiscover the secrets to crafting prompts that produce consistent, high-quality results.
CollectionCopy-paste 100 tested Midjourney v6 prompts: portraits, cinematic, fantasy, product shots & more. Free, updated for 2026 - instant results.
Social MediaCreate scroll-stopping Instagram content with these AI image prompts designed for Reels, Stories, and posts.
Browse our full library of ai prompts for data analysts — all free, copy-paste ready, no signup.
Or use our AI Prompt Generator to create custom prompts for your exact style in seconds.
Start with whatever is blocking your day — a stakeholder question that isn't specific enough to answer, a dashboard nobody reads because the narrative is missing, or a KPI whose definition changes depending on who you ask. Fill in the placeholders with your specific tables, columns, KPIs, and the actual audience. For SQL prompts, describe your schema (fact table, key dimensions, grain) explicitly — AI cannot guess your data model.
For any output that involves numbers, verify against the warehouse before sharing. AI will produce plausible SQL that references columns that don't exist in your schema, or draft an insight paragraph that assumes a metric direction that turned out to be reversed. Run the query, sanity-check the numbers against a known baseline, and only then attach the narrative.
SQL from AI has one common failure mode: it references columns and tables that sound right but don't exist in your warehouse. Before running any AI-generated query in production, check every table name and column reference against your data catalog. For anything more complex than a single-table aggregation, expect to rewrite the joins.
For insight writeups and executive summaries, AI defaults to strong claims from weak evidence. "Signups increased 12% due to the new onboarding flow" is a causal claim from correlational data. Always separate what the data shows (a change happened) from what caused it (a hypothesis, not a fact) — and rewrite any AI paragraph that conflates the two.
For prompts involving SQL, specify your dialect: PostgreSQL, BigQuery, Snowflake, Redshift, DuckDB. Syntax for date functions, window functions, and JSON handling varies enough that a generic SQL prompt often produces a query you have to translate. For dbt or SQL model prompts, add "this needs to run in dbt — reference models with ref() and include a config block."
For BI tool-specific prompts (Looker, Tableau, Power BI, Metabase, Mode, Hex), add the tool to the prompt. Looker's LookML syntax, Tableau's calculated field grammar, and Power BI's DAX are different enough that generic "dashboard narrative" prompts miss the mark. The more specific the tool and audience, the more usable the output.
The single move that raises the quality of any AI-drafted analysis: lead with the answer, not the methodology. Executives and PMs don't want to read the join logic — they want the number and what it means. Rewrite every AI draft so the first sentence contains the actual finding, the second sentence contains one supporting number, and the methodology moves to a footnote or an appendix.
For any insight paragraph, add a "what this doesn't tell us" line. AI drafts default to confident conclusions; the mark of a senior analyst is knowing where the data runs out. Adding one sentence of honest scope — "this uses signup data only, not activated users" or "the change coincides with a pricing update we can't isolate from" — turns a good draft into a defensible one.
Not without review. AI-generated SQL frequently references columns that sound right but don't exist in your schema, and often gets joins wrong for anything beyond a two-table aggregation. Use it to draft the structure, then check every table and column against your data catalog and run the query on a limit before trusting the output.
All of them, but you have to specify. Date functions, window function syntax, JSON handling, and string functions vary between PostgreSQL, BigQuery, Snowflake, Redshift, and DuckDB enough that a generic SQL prompt often produces something you have to translate. Add "my warehouse is BigQuery" (or your dialect) to every SQL prompt.
Describe your schema with generic placeholders (fact_orders, dim_customer) rather than pasting real tables. For actual data, use your BI tool's built-in AI (Hex, Mode, Looker have these) or an enterprise AI approved by your data governance team. Never paste PII, unreleased financials, or data under a customer contract into a public AI tool.
For the structure, yes. For the interpretation, edit carefully. AI drafts default to overclaiming — "the variant caused a 5% lift" — from correlational evidence. Rewrite every causal statement into what the data shows (a change happened) vs. what caused it (a hypothesis). Add sample size, confidence interval, and guardrail results explicitly.
Anything requiring judgment about your business — which KPI to pick, which segment matters, whether a result is real or noise, whether to ship an experiment. Also: any task that needs live production data in the prompt without an enterprise AI in place. Use AI for the writing around the analysis, not to replace the analyst's read of what the numbers mean.