Last Updated: September 4, 2026
The short version, in one breath.
Three parts of the product can use AI: the optional trip-helper chat, the signed-in AI itinerary action, and review-derived evidence prepared in advance. Search filters, maps, saved lists, packing, budgets, and deterministic catalog picks are regular software. The itinerary action remains unavailable until its destination passes EaseAway’s launch-certification gate. When available, AI may arrange and explain only grounded catalog places; software rejects unsupported place names and falls back to a catalog-only plan.
The trip-helper chat (real-time AI).
When you open the chat and type a message, that conversation is sent to Anthropic's Claude. Requests normally travel through Cloudflare AI Gateway for reliability and cost control; a narrow server-side direct route can be used if Gateway authorization fails. A live Gemini backup is not currently enabled. If Claude cannot return a valid grounded reply, EaseAway surfaces a clear failure or a deterministic catalog fallback rather than inventing an answer.
What the model sees: the messages you type in that conversation, and the trip preferences you state in it. What the model does not see: your email address, your account identity, or your saved trips. (We keep an internal log tying chats to accounts for abuse prevention and cost accounting — that log stays with us and is never part of what the AI reads.)
A practical ask: please don't type sensitive personal information into the chat — no government ID numbers, financial account numbers, or health details. It doesn't need them to help you plan a trip.
The AI itinerary action (real-time AI).
When the action is available for a launch-certified destination and a signed-in adult chooses “AI + live data,” Claude receives the selected destination, exact dates, party counts or age bands, pace and budget constraints, saved preference signals, and a server-built list of real catalog candidates. It does not receive your email address. Claude may select, order, and explain those candidates; it may not introduce a place outside the candidate list. If its response fails that validator, EaseAway discards it and shows a grounded deterministic plan.
Review summaries (prepared in advance).
The category verdicts you see on places (food / scenery / kid-friendly / and so on) are summarized from real Google reviews using Google Gemini — in advance, on our side, never triggered by your browsing. Every quoted snippet must pass a strict validator proving it is a verbatim quote from a real review. If the AI paraphrases, embellishes, or invents, the output is rejected and regenerated. Your usage of EaseAway never sends any of your data to this pipeline.
How we keep AI honest.
- No fabrication. AI-surfaced quotes must be verbatim from real source reviews, enforced by software, not policy.
- Honest empty states. Where we don't have enough real data, we show "not enough data yet" — never an AI-generated guess dressed up as fact.
- Cost and generation record. Paid generation records provider, model, prompt version, request identifiers, token counts, and cost fields needed for auditing. Provider access fails closed if the spend ledger cannot be read.
- AI makes mistakes. Hours, prices, and availability change. Treat AI suggestions as a starting point and verify directly with the hotel, attraction, or restaurant before you book or drive.
What AI never does here.
- It never books, reserves, or pays for anything. EaseAway has no booking or payment capability, AI or otherwise.
- It never writes reviews or ratings. Every review snippet traces to a real Google review.
- It does not decide where data is sent. EaseAway’s server sends only the request context described above to the named AI processor under that provider’s API terms.
- We don't use your conversations to train AI models.
If a future AI feature is not described on this page, it is not approved for public use.
Found a wrong answer?
If the AI told you something that turned out to be wrong, we want to know — it helps us fix the system for everyone. Use the private AI-error form with what you asked and what it said.
Changes to this disclosure.
We may update this page as features change. Changes are posted here with an updated date.