← All travel signals
Trust signal

The Ownerless Itinerary

In a synthetic test, the best of 15 AI trip planners built a fully workable plan on only 46.2% of solvable tasks. EU package-travel law never names AI.

What is it?

AI agents now chain flights, rooms and transfers into one trip, and independent benchmarks find most of those plans cannot be travelled. Europe’s new package-travel law names organisers by booking conduct. It never says whether the AI counts, or the seller behind it.

Why now?

Directive (EU) 2026/1024 entered into force on 28 May 2026, abolishing linked travel arrangements. Within weeks, two benchmarks put the best agent at 46.2% fully feasible and group plans below 12% valid. The law and the tools arrived together, without meeting.

Who’s building it?

AI assistant builders, online travel agencies, EU lawmakers, national transposers and the researchers testing trip planners.

AI assistant buildersOnline travel agenciesEU lawmakersBenchmark researchersNational transposers
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — assistant makers chaining trips across separate sellers.
Buyers — travellers who accept an AI plan without knowing if it is a package.
Gatekeepers — EU lawmakers and the member states transposing the directive.
What’s at stake
Whether a trip assembled by an AI assistant carries full organiser liability and insolvency protection, or none at all.
Who answers to the traveller when a linked itinerary fails.
Whether data handovers between sellers alone make a package.
What’s unsettled
Whether an assistant’s builder can be an organiser under the text.
How member states transpose the conduct tests by 2028.
Whether benchmark failure rates carry over to the live planners travellers actually use.
02 · How the demand matureda progression, not a timeline
Novelty
Chat assistants suggest destinations, then start linking flights and rooms into one plan.
Retention
Travellers keep asking agents to assemble whole trips, not single bookings.
Willingness to pay
Independent tests show most AI-built plans fail, so trust in the assembly carries the value.
Category
Trip assembly becomes a regulated act, whether a person or a model does it.
03 · Where the market goesthe forks, not a prediction
→Conduct captures the assistant: regulators read linked bookings and data handovers by an AI as organiser conduct, and liability follows.
→The seller behind the assistant becomes organiser, and the model’s builder stays outside the text.
→Assistants stop at suggestion, and travellers book each service alone, outside any package protection at all.
→A national court rules first, and its reading spreads across member states.
The forks turn on whether an AI that links bookings and passes data is treated as a trader’s booking system or as no one.
04 · The Agent Law deskadjacent market, moving now
The Admitted Agent
Courts decide who may book online.
The directing user, not the agent, accessed the site.
Cooley
Display Neutrality
Europe proposes one ranking rule for booking.
Paid prominence cannot be a default ranking criterion.
European Commission
The Liability Question
Blame for AI output still lacks settled owners.
Proposed AI LEAD Act would split liability between developers, deployers.
Al Jazeera
05 · The Wire
01
The best AI agent reached 46.2% of tasks.
arXiv ↗
02
Frontier agents produced valid group plans on under 12%.
arXiv ↗
03
The adopted package-travel text defines organiser status by booking conduct and never mentions AI.
Council of the EU ↗
04
Directive (EU) 2026/1024 entered into force on 28 May 2026 and does not directly address AI travel assistants.
CMS ↗
The panel on this signal
They agreed on
–The directive assigns organiser duties by booking conduct, and an AI assistant can perform that conduct without being named.
–Benchmarks, not live products, carry the failure figures.
–No court has ruled on who organises a trip that an AI assembled.
The travel-insolvency underwriter disagreed on
–Liability follows whoever takes the traveller’s money, so the question is the checkout, not the model behind it.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.