TripSapien
Paste a messy pile of travel research. Get a trip where every place is already checked against your dates.
- 30,093
- visitorslast 90 days · as of 2026-08-03
- 1,683
- trips plannedlast 90 days · as of 2026-08-03
- 17,664
- visitorslast 30 days · as of 2026-08-03
Source: the cross-project traffic dashboard (Amplitude, bot-filtered) · read 2026-08-03
Who it's for
Type-A travelers who over-research a trip, then lose a morning to a locked door.
The problem
People collect recommendations for weeks — friends' texts, blog lists, Reddit threads, a ChatGPT itinerary — into one messy document. Nothing in that pile knows whether a place is closed on the Tuesday they land.
What it does
You paste the pile in. A parser turns it into discrete places, each matched to a real venue in a ~365,000-row catalog, then checked against your actual trip dates: closed, limited hours, needs booking. Recommendations are shaped by your trip and your stated preferences.
What I did
ChatGPT writes broken itineraries and can't afford to verify them — checking a 30-place trip means dozens of live lookups nobody spends per answer. TripSapien precomputes the fix.
Killed my own first bet the day the evidence turned. March's bet was one-click Google Maps export; May 13 killed the premise twice in one day. The bet flipped to date-checking — the only job with a quantified failure mode and no incumbent.
Beat the Places-API cost wall with a free-data catalog: ~365k places from an academic Foursquare set matched against OpenStreetMap, and every paid lookup writes back so a place is paid for at most once. In the one $100 Maps month, real users cost about $13 — the rest was my own scripts.
Went AI-first on parsing after real pastes broke the deterministic rules, and moved the rules to where they win: matching against the catalog — exact, rare-token, phonetic, fuzzy, abstaining when unsure.
Built acquisition from assembled free public data — thousands of long-tail pages nobody defends — and precomputed recommendations in SQL with a per-city review floor, because a flat bar admits 12 places from a 1,754-row catalog in Xi'an.
Re-based every conversion rate on real devices after a July audit found 52% of raw SEO visitors were stealth crawlers converting at 0.012%.
What happened
Dated results, zeros included.
- 6,035
- Google clicks on city-month pages, at average position 9 — the family carries nearly all search traffic90d · as of 2026-07-31
- 0 → 138
- Bing AI citations, every one landing on an event pagebaseline 2026-05-21 → read 2026-07-20 · as of 2026-07-20Citations, not clicks — Bing search clicks read 198 on 2026-07-21, so the two metrics are demonstrably distinct.
- 693
- visitors referred by AI assistants — ChatGPT 618, and the only product Gemini refers90d · as of 2026-07-31
- 52%
- of raw SEO visitors were stealth crawlers — every rate re-based on real devices (5.9% → 7.8%)30d to 2026-07-07 · as of 2026-07-13
- self-overturned22 cut → 20 restored
- cities convicted by one similarity probe, cleared by three — the wrong audit kept on disk, struck through2026-05-22 → 2026-05-29 · as of 2026-05-29
- 0/15 → 15/15
- parse runs reaching the AI model, before vs after the fix — shipped knowing it cost ten secondsas of 2026-05-27
What's next
The rebrand's Change-of-Address matures around December 2026 — both legacy domains stay verified until it does. Neighborhood pages shipped July 21; their ranking verdict lands when the indexing window closes.