AI for Economist (Private Sector / Consulting)
Economists in consulting and corporate teams spend an estimated 9 hours a week assembling and documenting data and another 9 writing reports, memos and briefings, with about 4.5 more hours checking drafts, numbers and citations. The guides below show how to hand the drafting, formula explaining, code scaffolding and consistency checking to an AI tool, from copy-paste prompts to weekly release-watch digests. Your own models and spreadsheets still compute every figure, and your project lead or testifying expert still owns every opinion.
Try right now
Copy a prompt, paste into ChatGPT, Claude, or Gemini
Works with any free AI chatbot, no signup needed
A claim-by-claim list showing which table cell or citation supports each statement, followed by the claims with no support, numbers that disagree, and citation gaps.
Check this draft against its tables. List each claim with the table cell or citation that supports it. Then list every claim with no support, every number that appears more than once with different values, and every citation gap. Quote the draft text for each item. Do not fix or rewrite anything. Draft: [paste draft] Tables: [paste tables]
View full prompt →Tip: Use a public or redacted draft only. A client report, anything under a protective order or NDA, or non-public company data stays in the firm's approved systems unless the firm, the engagement and (on litigation matters) counsel approve the tool, and the tool's data settings are checked. Treat the output as a lead list. Then open the table or citation yourself to confirm each flag.
An R or Python version of your Stata lines that matches Stata's treatment of lags and gaps, with a list of anything that did not translate one-to-one.
Translate this Stata snippet to [R with dplyr / Python with pandas]. Match Stata's lag behavior exactly: the lag operator returns missing when the prior year is absent from the panel. Keep variable names, comment each line, and list anything that does not translate one-to-one. [paste snippet]
View full prompt →Tip: Compare results, not appearance. If the two programs disagree on the count of missing growth values, the gap handling is the first place to look. Paste only code and column names into the chat, never data, and keep any run on confidential data in your firm's own environment.
A clean source log with one row per series and a separate list of every field you still need to fill from the source page.
Build a source log from my notes below. Make a table with columns Series, Source, Release, Units, Seasonal adjustment, Vintage or retrieval date, Notes. Use only what I typed. Never fill a missing field and never supply a value. Write BLANK in any empty cell, then list every blank field after the table. Notes: [paste what you typed from each source page]
View full prompt →Tip: The chatbot only reformats what you typed, so the log is exactly as good as your notes. Type from the source page, not from memory, and compare each finished row against the page before it goes into the project folder. If any of the series came from a client, keep the client's name and files out of the chat and check your tool's data settings first (whether chats train the model, how long they are kept).
One paragraph per scenario that restates your assumptions and numbers as given, followed by a list of what would change the picture, built only from your stated assumptions.
Write up these scenarios from my table and assumptions. Give one paragraph per scenario that restates the assumptions and numbers exactly as given. Then add a "what would change this" list built only from my stated assumptions. Assign no probability or likelihood to any scenario. Label it a draft. Table: [table] Assumptions: [assumptions per scenario]
View full prompt →Tip: Read the "what would change this" list against your assumptions and cut any item that brings in a factor you never stated. If the table or assumptions come from a client project, code the labels ("[Client]", "Scenario 1") and check the tool's data settings and whether the firm approves it for that work.
A plain-language restatement of a results table that sticks to the reported values, plus a list of what the table cannot tell the reader.
Restate this regression output in plain language for a client reader. Use only the reported values and compute nothing new. Write "not reported" for anything missing. Then list what the table does not show, including causality, omitted variables, whether the model is right and whether the sample fits the question. [paste the results table]
View full prompt →Tip: You choose and defend the model. The chatbot only restates what is printed, so reread its text against the table, line by line, and delete any sentence that adds a number or an interpretation you did not make. Output from client or protected data stays out of the chat unless the firm and the engagement approve the tool. Retype it with the variable names changed if you are unsure.
A short structured summary of a paper (question, data, method, stated finding) and a table that ties each claim to a section or page, with an empty Checked column for you.
Using only the pasted text, summarize this paper as: question, data, method, stated finding. Then make a table with columns Claim, Section or page, Checked, and leave Checked empty. Write "not in the pasted text" for anything not given. Add no outside knowledge, no citations and no numbers I did not paste. [paste the abstract and the sections you have the right to use]
View full prompt →Tip: Paste only a public paper you have the right to use. Never cite the chatbot. Cite the paper, after you have checked each claim against the page. If the reply names an author, a year or a statistic you did not paste, discard that line.
A 300-word briefing draft in which every number is tagged to a cell in your table, ready for your project lead to mark up.
Draft a 300-word outlook briefing for [audience] using only the table below. Tag every sentence that contains a number with its table cell, such as [B2]. Cover these messages: [message 1], [message 2], [message 3]. Add no forecast beyond the table and no statement about what a central bank or government will do. End with: Draft for project lead review. [paste table]
View full prompt →Tip: Your model produced the numbers, so confirm that each tag points to the right cell and that no sentence quietly adds a number, a trend or a claim about policy. Keep the table to your own figures. If the table comes from a client engagement, use coded labels such as "[Client]" and check the tool's data settings (training use, retention) and whether the firm approves it for that work.
A short, jargon-free explainer for a client reader, anchored to a worked pair of numbers from your own series.
Explain "real versus nominal" and "what an index base year means" for a client reader with no economics training. Use my numbers: [your series, index and base year]. Stay under 200 words, explain every term you use, state no real data values and make no forecast.
View full prompt →Tip: Compute the real figure in your spreadsheet and put it into the prompt yourself, because the chatbot should explain, not calculate. Count the words of the reply. If a term appears without a plain definition, add "define that" and re-run.
Excel or Google Sheets formulas for growth, rebasing, real values, compound annual growth and a moving average, each with a one-sentence explanation.
My sheet has A Year, B [series], C Price index in rows 2 to [last row]. Write Excel formulas for year-over-year growth, rebasing B so the first year is 100, real B (B divided by the index over 100), compound annual growth and a 3-year average. Guard against blank cells where it applies. Explain each formula in one sentence. Do not calculate any results.
View full prompt →Tip: The math stays in the sheet. If any cell disagrees with the list above, fix the formula in the sheet and ask the chatbot to explain the difference, but never accept a number it types. Use the invented series for practice and for testing a formula before you point it at real data. Apply the same check on your own series by hand-computing one row.
A numbered data request list with field, period, frequency, format and purpose for each item, plus a set of questions to put to the client.
From this scoping note, write a numbered data request list with columns Field, Period, Frequency, Format, Why it is needed. Then list questions for the client. Request nothing the note does not imply, and mark every guess "to confirm". Note: [scoping note with placeholders]
View full prompt →Tip: On litigation matters, counsel reviews every request before it goes out. Keep the client's name and the matter name out of the chat and use placeholders, including in the pasted note. Check your tool's data settings, since a free consumer plan may train on chats. Cut any request that you cannot tie to the analysis plan.
Starter code in Stata, R or Python that loads a file, runs basic integrity checks and summarizes an outcome by year, with a marked gap where your own model goes.
Write [Stata / R / Python] code for a file with columns [column names]. Load the file and check for duplicate [id] and [year] pairs and for missing values, and summarize [outcome] by [year]. Then leave one comment reading MODEL GOES HERE. Do not choose or write a model. Add a one-line comment above each block.
View full prompt →Tip: Run the code on a small synthetic file first and read every line before it touches anything real. The prompt contains column names only, so no data leaves your machine. Code that runs on confidential data runs in your firm's own environment, never inside a chatbot's code tool. Confirm the duplicate check uses both columns together, not each one alone.
An empty assumptions log and a blank one-way sensitivity grid you can paste into a spreadsheet, with your assumption names in place and every value left for you.
Create an empty assumptions log and a one-way sensitivity grid layout. Log columns: ID, Assumption, Value (from your model), Source, Owner, Date set, Range for sensitivity (your choice). Grid: rows are assumptions, columns Low, Base, High, Result. Use these assumption names: [names]. Structure only. Leave every other cell blank.
View full prompt →Tip: The project lead or testifying expert owns every assumption and every opinion, and the chatbot should never suggest a value or a range. If it fills any cell, delete the entry and re-run with "leave blank" repeated. Treat the blank grid as a form. The model computes the Result column, not the chat.
Use AI in your tools
AI features built into tools you already have
No new subscriptions, just features you may not have noticed
Set up an AI assistant
Step-by-step guides for dedicated AI tools
10 to 30 minute setup, then ongoing time savings
Go further
Advanced workflows, automation, and custom AI setups
For when you’re ready to connect tools and automate
Recommended Tools
6Ranked by relevance for economist (private sector / consulting)
- 1
ChatGPT
Source Log and Series Documentation from What You Retrieved, ChatGPT Data Analysis to Check Growth Rates and Index Calculations
BeginnerVerified Oct 2026 - 2
Claude
Plain-Language Reading of Regression Output You Paste, Claude Project with Your House Style and Past Public Memos + 1 more
BeginnerVerified Oct 2026 - 3
Microsoft Excel
Formula Help for Growth Rates and Indexes in Copilot in Excel
BeginnerVerified Oct 2026 - 4
Microsoft Word
First Memo Draft in Copilot in Word from Your Notes
BeginnerVerified Oct 2026 - 5
Gemini Notebook
Gemini Notebook for a Literature Review of Papers You Upload
IntermediateVerified Oct 2026 - 6
Zapier
Weekly Data Release Watch Digest with Zapier
IntermediateVerified Oct 2026
Common questions
- What is the best AI tool for an economist (private sector / consulting)?
- 1. ChatGPT: Source Log and Series Documentation from What You Retrieved, ChatGPT Data Analysis to Check Growth Rates and Index Calculations. 2. Claude: Plain-Language Reading of Regression Output You Paste, Claude Project with Your House Style and Past Public Memos + 1 more. 3. Microsoft Excel: Formula Help for Growth Rates and Indexes in Copilot in Excel.
- How can an economist (private sector / consulting) use ChatGPT or another AI chatbot?
- Start with copy-paste prompts that work in any free chatbot. For example: An empty assumptions log and a blank one-way sensitivity grid you can paste into a spreadsheet, with your assumption names in place and every value left for you. Starter code in Stata, R or Python that loads a file, runs basic integrity checks and summarizes an outcome by year, with a marked gap where your own model goes. A numbered data request list with field, period, frequency, format and purpose for each item, plus a set of questions to put to the client.
- Do I need technical skills to start?
- No. Level 1 prompts work in any free AI chatbot with no signup beyond the chatbot itself: copy the prompt, fill in the bracketed details, and paste it in. Later levels add AI features in tools you already use, then dedicated AI tools and automation.
New to AI?
The Big Four AI Platforms
ChatGPT, Claude, Gemini, and Grok do roughly the same thing. Pick one and start.
The AI Usage Ladder
See how you use AI today, where that sits on the map, and the one next step to try.
Aim Higher with AI
Doing the same work faster is the smallest win. Three ways to ask AI for more, and why most of us under-ask.
How to Keep Up with AI
The landscape changes fast. A low-effort system to stay informed without drowning.
We update this guide when the tools change. See what's changed →