How this was built — bookings, transactions, photos, and AI

The premise

A journal is a reconstruction

Every source you have on hand is partial. Stitched together, they cross-validate each other and surface things any single source would miss.

• Bookings tell you where you slept.
• Transactions tell you where you ate, what you bought, and how much you spent.
• Photos tell you what you saw and exactly when (and, with EXIF GPS, where).
• AI planning chats tell you what the plan was — useful as a contrast against what actually happened.
• Workout-route GPX (Apple Health, Strava, etc.) tells you which trails you actually walked — with timestamps, distances, and elevation.
• Verbal memory tells you the why — the beats no source captures: the canyon-drive family tension, the wedding-venue twenty-minute stop, the kids playing chess at the hotel.

The method, in one sentence: look at two sources side by side and notice when they don’t agree. The disagreements are where the trip actually happened — concrete examples in the worked example below.

The materials

Eight source types — what each gives you

Almost every traveler has seven of these lying around already. The eighth comes out in conversation.

1. Booking confirmations — Airbnb, hotels

   Tells you the dates, the address, the price, the host. Search your inbox for “booking confirmed” or “reservation” and skim the last 6–12 months. Save as markdown; redact the confirmation code (it’s a credential).

2. Rental car details + host chat (Turo / traditional rental)

   Vehicle make/model, mileage cap, overage rate, pickup & drop-off logistics. For Turo specifically, the in-app host chat is editorial gold — the wrong-hotel pickup snafu, the toll question, the lockbox-code timing. Save the chat verbatim; redact phone numbers and lockbox codes.

3. Flight itineraries

   Carrier, flight numbers, dates, airports, prices. The least editorial source, but anchors the trip’s start and end. Save as markdown.

4. Credit-card transaction history

   The richest single source. Every meal, every fill-up, every gift-shop souvenir, with date, merchant, location, amount. Screenshot the trip-window from each card you used (multiple cards in a household? include all of them — transactions attribute to whoever swiped). Transcribe to markdown tables. Redact the card last 4 digits.

5. AI planning chat (ChatGPT, Claude, etc.)

   If you used AI to plan the trip, that chat is a record of the original intent. Export to markdown. Compare what was suggested to what actually happened — the diff is editorial.

6. Photo EXIF metadata

   Every iPhone photo carries a creation timestamp and GPS coordinates. Run exiftool across the trip-window photos and you have a precise where-and-when timeline. The iPhone “Keep Originals” gotcha: if your phone is set to “Most Compatible” on export, you lose EXIF and HEIC fidelity. Toggle to “Keep Originals” before exporting.

7. Apple Health workout-route GPX (or Strava / Google Fitness)

   Anyone wearing a watch on the trip likely has a stack of timestamped GPS traces sitting in Apple Health. Each outdoor walk, hike, or run gets recorded as a workout, with a .gpx file capturing every point on the route plus elevation and (sometimes) heart rate. Export from the iPhone Health app: profile picture → Export All Health Data → the resulting zip contains a workout-routes/ directory of date-stamped GPX files. Caveat: car drives don’t auto-record, so you only get walks/hikes — but those are exactly the beats your day-page narrative wants (the trail at the tower, the loop around the visitor center, the Door Trail at the Badlands). Strava and Google Fitness expose similar data via their own exports.

8. Verbal / memory additions

   The beats no source captures — the family-tension moment, the personal-history detour, the inside joke at the hotel, the “wait, did you actually do the cave tour?” question that fills in a missing piece. These come out in conversation, often in iteration with an AI partner asking narrative-shaped questions.

Pro tip: turn on Google Maps Timeline before your next trip

The order

The build sequence

Each step de-risks the next. Skipping ahead means revisiting later.

1. Source-materials gathering. Inbox sweep, card screenshots, photo export with EXIF preserved, AI-chat export.
2. Trip outline. One row per day: route, lodging, miles, where you ate. Built from bookings + transactions + photo GPS.
3. Cost roll-up. Categories (lodging / transport / food / groceries / fuel / activities) summed across cards. Catches missing days and orphan charges.
4. Photo cull. Our ratio was ~10:1 — 778 imports culled to 90 keepers across 7 days. Pick a daily hero and 8–14 supporting shots.
5. PRD. What is this journal for? What does “done” look like? Audience, sections, success criteria.
6. Tech spec. File layout, build pipeline, library choices, hosting target. Short — just enough to start building.
7. Build. Templates, copy, photos, search, ship.

The trap to avoid: starting with the build (because that’s the fun part) before you have the source materials organized. Half the drafting work is just deciding what actually happened, and you can’t do that without the data laid out.

The collaborator

AI as a build partner

AI is excellent at the structural work and bad at the specific work. Plan around that.

What AI does well

• Timeline reconciliation. Matching photo timestamps to card transactions to bookings — the kind of detail-tracking that’s tedious for a human and instant for an LLM.
• Cross-source validation. “The Redwater charge is on Apr 21 but the photo is on Apr 20” — AI catches these mismatches automatically when you give it both sources.
• First-pass narrative drafting. Given the trip outline + the editorial brief, AI can produce a 70%-there draft that you then edit for voice and detail.
• Generating Google Maps URLs from waypoint lists. The format is maps/dir/A/B/C/... — trivial once you know it, but tedious to build by hand.
• Iterating on factual corrections without losing context. Each correction propagates to every place it’s mentioned — the day page, the trip notes, the cost breakdown, the recommendations — in one round-trip.

What to keep human

• Family-context details. The kids’ personalities, the inside jokes, the spouse’s scenic-route preference. AI can’t generate these — only retell them after you supply them.
• The why behind each stop. Stone Mountain Lodge isn’t just a 20-minute parking-lot stop — it’s where Mat and Jacqui got married. The journal’s emotional spine comes from telling that out loud.
• Final factual review. Read the journal end-to-end against your sources. AI may confidently produce a wrong day or a mis-attributed restaurant; only you can spot it.

The working rule

The defaults

Privacy & redaction

A public journal earns trust by being explicit about what stays private. Here’s the default-redact list we used.

What stays out of saved markdown

• Confirmation codes — treat as credentials
• Credit-card last-4 digits — not useful for the journal, signal-rich for fraud
• Full home addresses — reduce to nearest town
• Kids’ last names, ages, schools — first name + role only (e.g., “Edrik — park-pass holder”)
• Lockbox codes — the literal code, not just the existence of one
• Host phone numbers — redact even from chat transcripts
• License plates & rental vehicle VINs — safer to omit

What we do keep

• Restaurant names, prices, dates — that’s the journal
• City- and town-level location for everything
• First names + role for family members
• Hotel and Airbnb listing names + URLs (publicly searchable)
• Trip-aggregate cost numbers and per-category breakdowns

The principle: nothing in the saved markdown should be a credential, a precise home location, or a detail about minor children that wouldn’t be in a school yearbook. Everything else is fair game for the journal.

In practice

How this journal got built

A tight retrospective for the curious. Same method as the abstract above — just with the actual numbers from this trip.

Source materials on hand

• ~778 raw photo imports across 7 days, exported with EXIF preserved
• 3 lodging confirmations (Avid Hotel Denver, Lodge at Moorcroft, Mary’s Place Airbnb in Rapid City) plus 1 hotel folio (Little America Cheyenne)
• 2 cards’ worth of transaction history (Apple Card + Citi Mastercard) covering the trip window
• 1 ChatGPT planning chat from before the trip, exported to markdown
• 1 Turo reservation with full host-chat history
• 1 flight itinerary (United outbound + JetBlue redeye home)
• 13 Apple Health workout-route GPX files covering the trail walks at Devils Tower, Mt Rushmore, the Badlands Door Trail, the Wind Cave tour, and the RMNP visit
• Verbal trip notes added during conversation with the AI partner over the course of the build

Corrections that surfaced from cross-referencing

• Redwater Kitchen day moved Apr 21 → Apr 20. Card charge posted late; photo EXIF was the source of truth.
• Day 3 route corrected to go US-85 south to Four Corners first, then north back through the canyon. The earlier reconstruction had the family taking I-90 east into the canyon and south — geographically possible, but not what happened.
• C&J Bus airport math. Earlier draft had “Uber from Logan home”; the family actually parked at the Seabrook station for $4.50/day ($31.50 total).
• RMNP visit confirmed. Was an open question; the $28 Beaver Meadows visitor-center charge plus the Pho-Thai-in-Estes-Park lunch settled it.

Looking forward

What the future Traveling Journals platform will automate

Most of the work above is mechanical — the kind of thing a platform can do for you. Here’s where this is headed.

• Bookings vault. Forward your reservation emails to a single address; they’re parsed, deduplicated, and slotted into the right trip automatically.
• Expense ingestion. Connect a card; trip-window transactions are pulled into the journal’s cost breakdown with merchant + location + category, ready to publish or keep private per row.
• EXIF processing. Drop a photo folder; we extract timestamps + GPS, cluster by day, propose hero shots, and write a first-pass per-day timeline you can edit.
• AI authoring assist. The same partner that drafted this journal’s narratives, but built into the editor — with the redaction rule installed by default and the “ask before saving sensitive” behavior baked in.
• Publish / keep-private toggles per section. Public route + photos, private cost breakdown, hidden booking details — your call, per section, per journal.