Meet Triply AI: A Travel Assistant That Can't Invent Restaurants

You have probably had this happen. You ask a chatbot for the best little seafood place in Lisbon, it names one with total confidence, describes the tiles on the wall, and recommends the grilled sardines. You walk twenty minutes to get there. It is a phone shop. It has been a phone shop for six years.

This is not a bug you can prompt your way out of. A general-purpose language model produces text that sounds like a restaurant recommendation, because that is what it was built to do. It has no way to check whether the restaurant is there. So it invents an address, invents opening hours, and hands you a plan with the shape of a good trip and none of the substance.

Triply AI is our answer to that, and it ships in Triply 2.0.

The idea: an assistant that has to look things up

Triply AI is a travel assistant built into the app, but it does not answer from memory. Before it can suggest anything it has to query real data — the same place database that powers Triply's itineraries, your own saved spots, and a library of travel videos other travellers have imported. If it cannot find a place, it says so rather than filling the silence with something plausible.

In practice it can do these things, and only these things:

• Search your saved spots, lists and trips. "What did I save in Rome?" is answered from your account, not guessed.
• Find places near you. Real coordinates, real categories, with photos.
• Check opening hours live. This one actually goes and looks, at the moment you ask — the single most useless thing an AI can get wrong on a trip.
• Search community imports. Thousands of travel videos that other Triply users have already turned into places. Ask for "food spots creators keep posting about in Osaka" and you get places that came out of real videos.
• Plan a route. Order stops so you are not zig-zagging across the city.
• Propose a trip. It builds the plan and shows it to you. You tap Create, or you dismiss it.

It proposes. You decide.

That last point matters more than it sounds. Triply AI cannot silently create, change or delete anything in your account. When it wants to build a trip it puts the proposal on screen — days, stops, route — and waits. Nothing exists until you tap Create.

We built it that way on purpose. An assistant that quietly writes to your data is one wrong interpretation away from wrecking a plan you spent an evening on, and you would have no idea which turn it took. A proposal you can read and reject is slower by one tap and considerably more honest.

Why grounding beats a better prompt

There is a common assumption that hallucinated recommendations are a quality problem — that a bigger model, or a cleverer system prompt, will eventually stop making things up. It will not, because the model is not consulting anything. Asking it to "only mention real places" is asking it to know something it has no mechanism to know.

The fix is architectural, not stylistic. You give the model tools, you make it use them, and you let its answer be limited by what those tools return. That is less impressive in a demo — a grounded assistant sometimes has to say "I could not find anything matching that" — and it is the difference between a plan you can walk and a plan you can only read.

What it is good at, and what it is not

• Good at: turning a vague intention into a real, mapped plan. "Three days in Porto, mostly food, nothing that needs booking" is exactly its shape.
• Good at: working with what you already saved. It can see your spots and lists, so it plans around them instead of ignoring them.
• Good at: the boring, high-value checks — is it open, how far apart are these, what order makes sense.
• Not good at: places that are not in any catalogue and have never appeared in a travel video. A brand-new bar with no online footprint is invisible to it — and it will tell you it found nothing rather than invent one.
• Not instant. When it checks live opening hours it is genuinely going and looking, which takes a moment.
• Premium feature. Triply AI is part of the paid tier. Trip creation and the free monthly allowance work without it.

How to try it

Update to Triply 2.0, tap the add button, and pick Triply AI. Then ask it something you would actually ask a friend who lives there — "where should I eat near my hotel on Thursday" works far better than "plan my trip", because the specifics are what it can go and check.

It pairs naturally with the import side of the app: pull places out of a YouTube vlog, TikTok saves or Instagram reels, then ask Triply AI to turn the pile into days.

Common questions

Why do AI trip planners recommend places that do not exist?

Because most of them generate text without consulting any source of truth about places. The model predicts a plausible-sounding restaurant name, address and set of opening hours. Nothing in that process checks whether the business is real, still open, or in the right city. The only reliable fix is to make the assistant look places up in a real database before it answers.

How is Triply AI different from ChatGPT for trip planning?

A general chatbot answers from training data. Triply AI has to query real sources — a place database with coordinates and photos, your own saved spots and trips, live opening hours, and thousands of imported travel videos. It also cannot change anything in your account on its own: it proposes a trip and you approve it before it is created. There is a full side-by-side comparison here.

Can Triply AI check if a place is open?

Yes. Opening hours are fetched live at the moment you ask, rather than recalled from a model's training data, which is the reason AI-generated itineraries so often send people to closed businesses.

Does Triply AI create trips automatically?

No. It builds a proposal and shows it to you — the days, the stops and the route — and nothing is saved to your account until you tap Create. It cannot edit or delete existing trips on its own.

Is Triply AI free?

Triply AI is part of the premium tier. Creating trips and the free monthly allowance do not require it.