Same five rules, one day in Amsterdam: ChatGPT and Gemini, checked

The test

The prompt, identical in both apps: Plan my Saturday in Amsterdam. Hard rules: 1. I arrive at Amsterdam Centraal station at 10:00. My train home leaves from there at 18:30. 2. I have a Rijksmuseum ticket for 13:00. 3. Lunch before the museum. 4. Walking and public transport only. 5. Spend no more than 50 euros in total, not counting the museum ticket. Give me a timed schedule with the cost of each item and the total.

Conditions: ChatGPT Pro in high reasoning mode and Gemini Pro, on the operator's own accounts, one run each, 23 September 2026. The trip and the ticket are made up; nothing personal was typed. Before running, we fixed seven checks: the museum stays at 13:00, lunch happens before it, the plan returns to Centraal before 18:30, only walking and public transport are used, the costs add up, the total is 50 euros or less, and the plan says which prices are estimates. A boat tour was defined in advance as not public transport; the free GVB ferry would count.

What ChatGPT planned

ChatGPT stated its assumption first: this Saturday, 26 September 2026. Its schedule, as it wrote it (dashes shown as hyphens): 10:00-10:10 arrive at Amsterdam Centraal, 0 euros. 10:10-10:40 walk down Damrak to Dam Square, 0. 10:40-11:00 visit the Begijnhof, 0. 11:00-11:35 walk south through the canal belt toward lunch, 0. 11:35-12:10 lunch at Stefs Bakery Amsterdam, a classic sandwich (8.50) plus a croissant (3.00), 11.50. 12:10-12:30 walk to the Rijksmuseum, 0. 12:30-13:00 buffer for the entrance and a look around Museumplein, 0. 13:00-15:30 Rijksmuseum, ticket already paid. 15:30-16:15 a loop in Vondelpark, 0. 16:15-17:05 walk back toward Leidseplein, 0. 17:05-17:35 tram 2 from Leidseplein to Centraal with a GVB 1-hour ticket, 3.40. 17:35-18:30 back at Amsterdam Centraal with roughly 55 minutes of buffer, 0.

Its total: 14.90 euros, leaving 35.10 under the limit. It added that GVB works have altered several tram routes and that GVB recommends checking its journey planner shortly before travelling.

What Gemini planned

Gemini opened with: Here is a complete, timed itinerary for your Saturday in Amsterdam that keeps you strictly under your 50 euro budget (excluding the museum ticket). Its schedule: 10:00 - 12:00 arrival and historic centre on foot, via the Begijnhof and the Flower Market, 0 euros. 12:00 - 12:45 lunch near the museum, with an estimated 20 euros budgeted and five suggested places: The Pantry (listed 20 to 30 euros), Cafe Mankind (10 to 20), Brasserie Zuiderbad (10 to 20), Bakers and Roasters (20 to 30) and Stoop and Stoop eetcafe (20 to 30). 13:00 - 15:30 Rijksmuseum. 15:30 - 16:30 Vondelpark, then a tram back to the centre, 3.40. 17:00 - 18:00 canal cruise near the station: With your remaining budget (approx. 26 euros), a boat tour is the perfect finale, with five operators listed. 18:30 train home.

Its budget breakdown: morning walking tour 0.00, lunch about 20.00, Rijksmuseum 0.00 (pre-paid), tram from Leidseplein to Centraal 3.40, canal cruise about 20.00, total estimated cost 43.40 euros. The copied answer also carried live place cards with ratings and opening status, which we left out here.

How each plan scored

ChatGPT passed six checks outright and one partly: museum 13:00 to 15:30, lunch 11:35 to 12:10, back at Centraal from 17:35, walking plus one tram, 11.50 plus 3.40 equals 14.90, under 50 euros. It flagged the tram disruption risk and stated its date assumption, but gave the menu prices as fact rather than estimates.

Gemini kept the fixed times: lunch by 12:45, the museum at 13:00 and the station by 18:00. Its arithmetic adds up, and it labelled its amounts as estimates. Two things did not hold. First, the 17:00 canal cruise is a boat tour under a walking and public transport only rule. Second, three of its five lunch picks are listed at 20 to 30 euros against its own 20 euro lunch line. Pick one of those at its top price and the day comes to 53.40 euros, over the limit the plan called strictly under.

Is a canal cruise public transport?

We decided before running that it is not: a canal cruise is a paid sightseeing tour, even when it ends near the station. You can argue the other way, because Gemini presents it as a way back to Centraal. We count it as a miss by the rule we set in advance, and we would rather show that call openly than hide it. If you disagree, tell us under the video.

The one-minute check for any AI plan

Check every fixed time: find each one you gave it and confirm it did not move. Add up the costs yourself: a total is only as good as the lines above it. Check every option, not just the total: if the plan offers choices, the promise has to hold for all of them, or it should say which ones break it.

What this test can and cannot tell you

This is one run per app, on one day, for one kind of task. It shows what happened in this checked test, not which assistant is better at planning in general. Prices, opening hours and tram fares in both answers are the apps' claims; we did not verify them, so check them before you travel. For a second reading we also ran the same prompt in ChatGPT without an account, twice; both runs walked the whole day and came to 20 euros, and also kept every rule.

Run the same test on your AI

Use the fictional example first. Do not paste confidential work into an unapproved tool.

Plan my Saturday in Amsterdam. Hard rules:
1. I arrive at Amsterdam Centraal station at 10:00. My train home leaves from there at 18:30.
2. I have a Rijksmuseum ticket for 13:00.
3. Lunch before the museum.
4. Walking and public transport only.
5. Spend no more than 50 euros in total, not counting the museum ticket.
Give me a timed schedule with the cost of each item and the total.

ChatGPT Pro, 23 September 2026: walking plus one tram 2 ride from Leidseplein, lunch 11:35 to 12:10 at a bakery, back at Centraal at 17:35. Total 14.90 euros.