💡 New script: gmail-receipt-tracker (Gemini-powered receipt & transaction extraction to Sheets)

[github-actions[bot]]

Summary

A GAS script that uses Gemini Flash to identify transactional emails (receipts, order confirmations, shipping notifications, subscription charges) and extract structured data (vendor, amount, date, category, order number) into a Google Sheet. Turns your inbox into an automatic expense/purchase ledger without manual data entry.

Market Signal

AI email automation moved from novelty to necessity in 2026. Knowledge workers spend ~3 hours/day on email. Paid tools like Expensify ($5–20/mo) and Dext extract receipts but require third-party data access and paid subscriptions. No free, privacy-respecting, Workspace-native alternative exists. Gemini 3.5 Flash (launched at I/O 2026) makes per-email classification cheap and fast. The Gemini structured output mode (JSON schema enforcement) enables reliable field extraction without fragile regex parsing.

User Signal

The gmail-ai-classifier (#492) validates that Gemini + Gmail integration works reliably in this codebase. Users already deploy gmail-to-drive-by-labels for email archiving — receipt tracking is a natural adjacent need. Self-employed users and freelancers managing expenses for tax purposes are an underserved segment in the personal GAS ecosystem. No existing idea discussion covers structured data extraction from transactional emails.

Technical Opportunity

Builds directly on the proven gmail-ai-classifier pattern: Gmail API message scanning → Gemini classification → structured output. The dual-layer architecture (code.gs → src/index.js with dependency injection) and test-utils/ mock infrastructure are already established. Deploy page Step 4 can provide a Sheets picker for the destination spreadsheet plus category configuration. Gemini structured output mode (JSON schema) eliminates parsing fragility — the model returns typed fields (amount as number, date as ISO string) instead of free text.

Assessment

Dimension	Score	Rationale
Feasibility	high	Proven pattern from gmail-ai-classifier; Gemini structured output handles extraction; Sheets API is well-documented
Impact	high	Eliminates manual expense tracking for freelancers/self-employed; saves 30–60 min/week for active users
Urgency	med	No competitor threat — the free/open-source niche is unoccupied — but Gemini capabilities are mature enough to build now

Adversarial Review

Strongest objection: Receipt formats vary wildly across vendors and languages. Gemini may miss fields or hallucinate amounts, creating unreliable financial data that users trust for tax purposes.

Rebuttal: Gemini excels at unstructured text extraction, and the structured output mode (JSON schema enforcement) constrains hallucination to defined fields. The design should start with high-confidence fields (vendor, amount, date) and flag low-confidence extractions with a confidence_score column for human review. Partial extraction is still more useful than zero automation. The existing gmail-ai-classifier proves Gemini classification reliability at inbox scale with gemini-2.0-flash.

Suggested Next Step

Prototype the Gemini prompt for receipt extraction using 10 sample transactional emails from common vendors (Amazon, Google, airline confirmations, subscription services). Validate extraction accuracy, define the Sheets output schema (date, vendor, amount, currency, category, order_number, confidence), and assess Gemini API cost per email at gemini-2.0-flash pricing.