SQL, Python and Tableau show up in most data analyst job postings, which makes listing them a weak way to tell your resume apart. What separates one analyst from another is downstream: who read the analysis, what they did differently afterward, and whether the dashboard is still open a year later. These prompts push an AI to trace each project to the decision it changed, keep the tooling in a trailing clause, and mark where a figure belongs instead of producing one.
Want a finished document to work backwards from? There is a full data analyst resume example in Markdown you can copy into the editor.
Before you paste any of these
Give it the query, the dashboard and the audience — Name who consumed the work — a finance lead, an ops manager, a weekly exec review — and ask the AI to write from that reader's point of view. An analysis with no named consumer usually turns out to be a bullet with no ending.
Put the decision first and the method last — Ask for a sentence structured as: the action taken (a price change, reallocated spend, a retired report, a fixed metric definition), then how you got there. The technique is the trailing clause, which is where SQL and Python read as evidence rather than as a list.
Separate one-off analyses from what you own on a schedule — Recurring ownership reads differently from ad hoc work: a nightly job you maintain, the canonical definition of a core metric, a data quality check that catches bad loads. Ask the prompt for 2 groups so both make the page.
Refuse the offered percentage — Asked for a strong analytics bullet, a model will hand you a lift figure it has no way to know. Instruct it to write [X%] and to name where you would look the real number up — the experiment readout, the finance model, the ticket where the change was approved.
3 prompts to copy
Every prompt ends with the same rules: use only what you give it, and mark a missing number with a placeholder instead of guessing one.
Your bullets read like a stack list — SQL, Python, dbt, Tableau — and a hiring manager still cannot tell what anyone did differently because of your work
What to have ready
•3 to 8 bullets from your current resume, exactly as written
•One line per bullet on who read the analysis and what they decided afterward — write "not sure" where you cannot remember
You are reviewing resume bullets written by a data analyst. Analyst resumes tend to fail the same way: they name the tools used and the artifact produced, and never say what a decision-maker did differently afterward. Move each bullet from tooling to consequence, using only the material I give you.
My bullets:
[PASTE 3 TO 8 BULLETS]
What I remember about who used each analysis and what happened next:
[PASTE ONE LINE PER BULLET — WRITE "not sure" WHERE YOU DO NOT KNOW]
For each bullet, work through 4 steps and show all 4:
1. Diagnose. Name what the bullet currently proves: tool familiarity, volume of work, or a decision that changed. Quote the words carrying the claim.
2. Locate the decision. Using my notes only — never what is usual for analytics work — state who acted on the analysis and what changed: a threshold moved, a launch was held, a segment was dropped, a weekly report was retired. Where my note says "not sure", write "decision unknown".
3. Rewrite. One line, under 30 words, ordered as: what the business needed to decide, what I built to answer it, what changed. Keep a tool name only where the sentence stops making sense without it — a migration, the first pipeline of its kind. Use [add detail] wherever the sentence needs a fact I did not give you.
4. Ask. One specific question that would strengthen the rewrite, aimed at the missing fact rather than at general resume advice.
Rules:
- Where the decision is unknown, offer the bullet as a scope statement instead — what I owned, over what data, for whom — and say in one line what would upgrade it.
- Treat "built a dashboard", "automated a report" and "analyzed data" as unfinished sentences. The finished version names the reader of the analysis and the change.
- Never supply the reader or the change yourself. If my notes do not name them, the bullet stays a scope statement.
- Do not rank the bullets or add a closing summary. Return the 4 steps per bullet and stop.
Rules — these override everything above:
- Use only the experience I give you. Do not invent employers, job titles, dates, degrees, certifications, skills, or metrics.
- If a number would strengthen a line and I have not given you one, write a placeholder like [X%] or [$X] for me to fill in. Do not estimate it.
- Do not add a tool, skill, or credential I did not list, however standard it is in my field.
- Rewrite, reorder, tighten and re-emphasize what is already there. Where I have given you too little to work with, ask me for it or leave [add detail].
Send this next
Answer only the questions you genuinely know the answer to, paste those answers back, and ask for a second pass. Leave the rest as scope statements rather than reaching for a number.
Open inCopies the prompt, then opens the assistant
You know the work mattered, but the only figures you can name are row counts and dashboard views — and those measure effort, not outcome
What to have ready
•1 to 3 projects described in your own words — the question, what you built, who used it
•Any figure you can actually verify: query runtimes before and after, refresh cadence, recurring reports replaced, headcount using the dashboard, the business metric the team tracked
You are helping a data analyst find the measurable part of their work.
My projects, in my own words:
[PASTE 1 TO 3 PROJECT DESCRIPTIONS]
Figures I can verify:
[PASTE REAL NUMBERS — RUNTIMES, REFRESH CADENCE, REPORT COUNTS, AUDIENCE SIZE, THE METRIC THE BUSINESS TRACKED. WRITE "none" IF YOU HAVE NONE]
For each project, build a 4-row table and fill only what my text supports:
- Scale — what the analysis covered: sources joined, entities, time span, business units served.
- Speed — how long something took before and after: manual assembly of a report, query runtime, time from question asked to answer delivered.
- Reach — who consumed the output: named teams, functions, the recurring meeting it fed.
- Consequence — the business metric someone moved after reading it, and which direction it moved.
Then write 2 candidate bullets per project:
Bullet A uses only the figures I verified.
Bullet B is the stronger shape with a bracketed placeholder where a number belongs — [X hours per week], [$X], [X%] — plus one line telling me exactly where to go looking for it: the ticket that requested the work, the dashboard's own view count, a scheduler log, the monthly close pack, the person who asked.
Rules:
- Never supply a number yourself, not even an approximate one, and never round or extend a figure I gave you. A placeholder is always the correct answer when I did not provide the figure.
- Where a row has no support in my text, write "no evidence yet" and move on.
- Do not convert a count into a percentage, or a percentage into a dollar amount, unless I gave you both sides of the arithmetic.
- Flag any figure I supplied that reads like an outcome but is really a volume measure — rows processed, dashboards built, queries written — and say what would turn it into one.
- Rank the placeholders at the end by how findable they are, so I know which to chase first.
Rules — these override everything above:
- Use only the experience I give you. Do not invent employers, job titles, dates, degrees, certifications, skills, or metrics.
- If a number would strengthen a line and I have not given you one, write a placeholder like [X%] or [$X] for me to fill in. Do not estimate it.
- Do not add a tool, skill, or credential I did not list, however standard it is in my field.
- Rewrite, reorder, tighten and re-emphasize what is already there. Where I have given you too little to work with, ask me for it or leave [add detail].
Send this next
Chase the 3 most findable placeholders and fill them in yourself. A number you can defend in an interview beats a rounder one you cannot.
Open inCopies the prompt, then opens the assistant
Analyst postings tend to list the same tools, so stack overlap tells you little. You need to know whether this team wants a product analyst, a reporting owner, an analytics engineer or a risk modeler — and which of your work speaks to that
What to have ready
•The full job posting, pasted as text
•Your current resume
You are comparing a data analyst's resume against one job posting. Analytics postings tend to list the same tools, so stack overlap tells you little. Read for the job underneath the posting.
Job posting:
[PASTE THE FULL POSTING]
My resume:
[PASTE YOUR RESUME]
Work through 4 sections:
1. What kind of analyst this is. Decide which of these the posting actually describes, and quote the 3 lines that decided it: product and experimentation analytics, business intelligence and reporting ownership, analytics engineering and modeling, or risk, credit and finance analytics. If it is a mix, name the dominant half. Then say whether the posting shows who owns the pipelines, quoting the line that shows it, or write "cannot tell from the posting".
2. Their standing question. Infer the recurring question this team is asked — why churn moved, which channel to fund, whether a limit should change, where margin leaked. Quote the responsibilities and metrics that support your reading, and say how confident you are. If the posting does not support a reading, say so instead of choosing one.
3. Evidence map. 2 columns. Left: each requirement in the posting's own words. Right: the single strongest line from my resume that answers it, quoted exactly. Where my resume uses a different word for the same work than the posting does — "reporting" against "business intelligence", "experiment" against "A/B test" — note both words and say which one to use. Where nothing answers a requirement, write "gap" and label it a stack gap (learnable — name the substitute I already have) or a domain gap (regulated reporting, experiment design, financial close) that I should not paper over.
4. Rewrite plan. List the resume lines to reorder or reword, and give each replacement sentence built only from words already in my resume.
Rules:
- Reordering and rewording my experience is allowed. Adding experience is not — where the posting asks for something my resume does not show, it goes in the gap list, not into a new bullet.
- Quote my resume rather than paraphrasing it, so I can see which line you matched.
- Return the 4 sections and nothing else.
Rules — these override everything above:
- Use only the experience I give you. Do not invent employers, job titles, dates, degrees, certifications, skills, or metrics.
- If a number would strengthen a line and I have not given you one, write a placeholder like [X%] or [$X] for me to fill in. Do not estimate it.
- Do not add a tool, skill, or credential I did not list, however standard it is in my field.
- Rewrite, reorder, tighten and re-emphasize what is already there. Where I have given you too little to work with, ask me for it or leave [add detail].
Send this next
Apply the rewrite plan one line at a time, and keep the gap list — those are the questions the first interview will open with.
Gemini: Keep the posting and the resume under separate headings, and ask Gemini to quote from one source at a time — otherwise the evidence map blends the two.
Open inCopies the prompt, then opens the assistant
General prompts that also work for a data analyst resume
A chat window returns text. Paste it here and it opens in the editor, where 32 templates turn it into an interview-ready PDF. Free, no signup, and your draft stays in your browser.
