💡 New script: gmail-travel-itinerary (Gemini-powered travel booking extraction to Sheets & Calendar)

[github-actions[bot]]

Summary

Extract flight, hotel, and car rental confirmations from Gmail using Gemini structured output, consolidating multi-provider travel bookings into a unified Google Sheet (with dates, confirmation numbers, costs) and optionally creating Calendar events for each trip leg. Follows the project's label-scoped processing pattern.

Market Signal

Travel itinerary extraction is explicitly identified as an underserved category in the personal productivity AI market. TripIt (SAP Concur) is enterprise-focused and requires email forwarding. Google's native travel cards in Gmail are basic and non-exportable. No Workspace-native tool consolidates multi-provider bookings into a structured Sheet or Calendar. UK Making Tax Digital (April 2026) drives automated travel expense tracking. Competitors (SaneBox, Superhuman, etc.) charge $7-36/month and don't offer travel-specific extraction. The Vertex AI Advanced Service reaching GA in Apps Script (January 2026) makes Gemini-powered parsing a first-class capability.

User Signal

The gmail-receipt-tracker idea (#534) already tracks financial transactions from email — travel itinerary extraction is the natural companion for a different data type (logistics vs. finances). No existing idea covers travel-specific extraction. The project's gmail-ai-classifier (merged as PR #492) demonstrates the Gemini-powered email parsing pattern this script would extend. The label-scoped processing architecture from gmail-to-drive-by-labels provides the proven input pipeline.

Technical Opportunity

Leverages the project's proven architecture: label-scoped Gmail processing (from gmail-to-drive-by-labels), Gemini structured output (from gmail-ai-classifier via Vertex AI), and Sheet output (from calendar-to-sheets). The dual-file pattern (code.gs + src/index.js) with dependency injection enables comprehensive testing of parsing logic. Start with major providers (United, Delta, American, Marriott, Hilton, Hertz) and expand. A validation layer (similar to validateClassification in the classifier) catches hallucinated dates or confirmation numbers.

Assessment

Dimension	Score	Rationale
Feasibility	med	Diverse email formats require robust Gemini prompting; validation layer needed to prevent hallucinations; start with top 10 providers for MVP
Impact	high	Genuine market gap with no free Workspace-native alternative; daily utility for business travelers; pairs naturally with receipt-tracker
Urgency	high	UK Making Tax Digital (April 2026) creates compliance driver; Vertex AI GA makes implementation timing ideal; no competitor filling this gap

Adversarial Review

Strongest objection: Email format diversity across airlines, hotels, and rental companies makes reliable extraction very difficult. Gemini could hallucinate dates or confirmation numbers, and non-English confirmations add complexity.

Rebuttal: Gemini 3.5 Flash's structured output handles diverse formats well — the gmail-ai-classifier already proves this pattern works in this codebase. The script starts with a focused set of major providers and uses a validation layer to catch hallucinations. Label-scoping means users explicitly mark travel emails, reducing the search space. Even imperfect extraction (90% accuracy) saves significant manual data entry versus the alternative of zero automation.

Suggested Next Step

Create a design spike: catalog the email formats of the top 5 airlines and top 5 hotel chains, define the Gemini extraction prompt and JSON schema (departure city, arrival city, date, time, confirmation number, cost), and validate extraction accuracy on sample emails before full implementation.