Learn to GPT

Learn to GPT

For Analysts & Researchers

Analyze Data, Write SQL, and Extract Insights Without Code

CSV analysis, SQL generation, research synthesis, structured output

Claude's 200K context window, reliable JSON output, and strong reasoning make it the most versatile AI for data work. Paste a CSV, describe your database schema, dump a research paper — Claude analyzes, extracts, queries, and reports without you writing a line of code.

Start Learning Free Prompt Engineering

Why Claude

What makes Claude good at data work

200K context — your entire dataset fits

A 200,000-token context window holds roughly 150,000 words or ~500 pages of text. In data terms: a 5,000-row CSV with 10 columns, 20 research papers, an entire annual report, or a year of customer support tickets. Claude reads it all in one session — no chunking, no summaries, no information loss at the edges.

Reliable structured output — JSON schemas that hold

Claude follows JSON schemas reliably — more consistently than most alternatives. For data extraction pipelines where every record needs to conform to a schema, this matters enormously. Define your output structure once; Claude applies it faithfully across thousands of documents.

Strong quantitative reasoning

Claude scores well on quantitative benchmarks — MATH, GSM8K, MMLU. For data work this means: it catches calculation errors, reasons correctly about statistical concepts (correlation vs. causation, sampling bias, p-values), and flags when your framing of a question is analytically flawed.

Workflows

Six data analysis workflows — with prompts

CSV and spreadsheet analysis

Paste raw CSV data (or describe the columns) and ask Claude to analyze it. Descriptive stats, trend identification, anomaly detection, cohort comparisons — in natural language.

Here's a CSV of monthly sales data for 3 products over 24 months. Identify: (1) which product has the strongest growth trend, (2) any seasonal patterns, (3) months where all 3 products underperformed, (4) the top 5 individual anomalies.

SQL generation and query writing

Describe your database schema and the question you want answered — Claude writes the SQL. Works for Postgres, MySQL, BigQuery, SQLite. Include your schema DDL for best results.

Schema: users (id, email, created_at), orders (id, user_id, amount, status, created_at). Write a query that returns monthly revenue by cohort (month of user signup), for the past 12 months, with MoM growth rate.

Research synthesis

Paste multiple papers, reports, or sources. Claude synthesizes key findings, surfaces contradictions, maps the evidence, and highlights gaps — across the full 200K context window.

Read these 5 research papers on [topic]. Synthesize: (1) findings that appear in 3+ studies, (2) directly contradictory findings and likely explanations, (3) methodological limitations across the literature, (4) the most important open questions.

Structured data extraction

Feed Claude unstructured text — PDFs, emails, web pages, news articles — and ask it to extract structured data into JSON, CSV, or a schema you define. Reliable, fast, and auditable.

Extract from this job posting: {role_title, company, location, salary_range (null if not listed), required_years_experience, tech_stack (array), remote_policy}. Return as JSON. If a field is ambiguous, add an 'extraction_notes' field.

Report generation

Give Claude raw data and a report template (or describe the audience). It generates the narrative, pulls the key metrics, writes the executive summary, and flags what needs human judgment.

Here's our Q2 marketing data. Write a board-ready executive summary: lead the headline finding, show the 3 most important metrics with context, flag 1 risk, end with 2 recommended actions. Max 350 words. No jargon.

Exploratory data conversations

Use Claude as a thought partner on messy data. "I have this dataset and I'm not sure what questions to ask." Claude helps you frame the analysis, choose the right approach, and interpret ambiguous results.

I have 18 months of customer support ticket data: ticket_id, created_at, category, resolution_time_hours, CSAT_score (1-5), agent_id, product_area. What are the 5 most valuable analyses I could run? For each, describe what I'd learn and why it matters.

Code Examples

Working with the Claude API for data tasks

Python: CSV analysis with pandas prompt

# Prompt Claude with your data
import anthropic
import pandas as pd

df = pd.read_csv("sales_data.csv")
csv_preview = df.head(20).to_csv()
schema = df.dtypes.to_string()

client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=2000,
    messages=[{
        "role": "user",
        "content": f"""Schema:
{schema}

Sample data (first 20 rows):
{csv_preview}

Analyze: top 3 trends, any anomalies,
and the single most actionable insight."""
    }]
)
print(response.content[0].text)

Structured output: extract to JSON

# Extract structured data reliably
response = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1000,
    messages=[{
        "role": "user",
        "content": """Extract from this job posting as JSON:
{
  "role_title": string,
  "company": string,
  "salary_min": number | null,
  "salary_max": number | null,
  "remote_policy": "remote"|"hybrid"|"onsite",
  "tech_stack": string[],
  "years_experience": number | null
}

Job posting:
""" + job_posting_text
    }]
)
import json
data = json.loads(response.content[0].text)

By Role