The Invisible FunnelHow AI agents evaluate real brands · by Accord
Powered by Accord

Stories · Ryanair

I told the agent I hate hidden fees. Ryanair’s help pages wouldn’t let it check.

In four journeys the agent recommended no airline: Ryanair’s help pages refused every request, and the all-in fare with bags, seats and airport check-in could not be established on any airline’s own pages.

Open-market search Four runs · open-market search · one research-agent configuration · no booking made · 2026-09-20 · all 4 journeys of this decision

In four recorded research runs, the agent read ryanair.com’s fees page every time and was refused by help.ryanair.com every time it asked. It also read Aer Lingus in all four journeys and was refused by Vueling’s help pages in three. The all-in total for two passengers with 20 kg bags, seats together and airport check-in was not established for any airline. Two runs ended unresolved; two ran out of page reads.

The customer’s requirements
Two adults, Dublin to Barcelona, out Friday 13 November 2026, back Monday 16, one 20 kg checked bag each, seated together, airport check-in on the way out; from the airline’s own pages: the all-in total with those bags and seats, the airport check-in fee, the change fee, and a phone number to a person.
The information used
ryanair.com fees page read in every journey. help.ryanair.com: refused in 4 of 4. aerlingus.com read in 4 of 4. help.vueling.com refused in 3 of 3. An Irish phone number with limited hours was recorded from a search extract of a refused help page.
The resulting recommendation
None. Two journeys ended with Ryanair, Aer Lingus and Vueling unresolved on the total, the change fee and, once, phone support; two exhausted their page reads.

The full evidence, every option weighed and every source, follows below and on each journey page.

The customer

Two adults in Dublin who want a long weekend in Barcelona in November with a checked bag each, seats together and airport check-in on the way out. The conditions are the ones a fee-averse traveller writes down: the all-in total, the airport check-in fee, the change fee, and a person on the phone. Two of the four customers remembered paying a €55 airport check-in fee and a €45 fee for a bag that the search results had called free, and a low headline fare that ended up 60 percent higher after seats and bags.

What the agent researched

Between 9 and 15 page reads per journey. ryanair.com, help.ryanair.com and aerlingus.com in every journey; vueling.com and help.vueling.com in three. help.ryanair.com answered every request with a bot wall; help.vueling.com did the same in all three journeys that reached it. ryanair.com itself, including its fees page, was readable.

What it believed

That Ryanair publishes a fee table, read on ryanair.com, and that the all-in total for this itinerary needs a live fare plus that table, which it could not assemble. That Ryanair has an Irish phone line with limited hours, a fact it recorded from a search extract of a help page it could not open. For Aer Lingus and Vueling, the same shape: fares and fees visible only in booking flows it could not complete.

The decision

Nothing was recommended. Two journeys ended with every airline unresolved on the total and the change fee; two ran out of page reads first. No airline lost to another; none could be verified.

What was never established

The all-in total, in every journey. The change fee, in every journey. Phone support, in one. The customer’s worst memory was a fee that appeared late; the agent could not find, on any page it was allowed to read, the number that would have ruled that out.

Fresh versus persistent memory

No difference in outcome. The two customers who remembered the airport fee and the bag fee went to the fee pages first; the record shows that in the early actions. All four ended without a recommendation.

What this means for the brand

The pages that hold the answers to a fee-averse customer’s questions, the help centre and its fee articles, refused the agent every time. The main site was readable, but the all-in figure lives in the booking flow and the help centre, and the agent could reach neither. Whether opening the help pages would change the result is not established here.

What could be tested

The same customer and requirements with the fee articles readable, and a plain-text all-in example for a bag-plus-seat itinerary, with these four journeys as the baseline. The measure is whether the total is established, then whether a recommendation follows.

The journeys

no memorybrand page reachedrefused by help.ryanair.comobserved2026-09-20
  1. 1Customer missionRyanair decision — customer with no relevant memory
    • Direct flight availability Dublin–Barcelona on 13 and 16 Nov 2026
    • All-in total for 2 adults, return, 2×20kg checked bags, seats together, outbound airport c…
    • Total at most 480 EUR
    • No charge may remain unknown before booking
  2. 2Evidence14 searches, 10 pages: ryanair.com, aerlingus.com, help.ryanair.com
    • total price eur: null no source
    • all in price established: false brand page
    • change fee eur: null no source
    • human phone support: Ireland number reported to exist (+353 1 691 7177), limited weekday/weekend hour attributed to a page that refused the agent
  3. 3BeliefCould not establish: total price eur; all in price established; change fee eur
  4. 4DecisionNot selected: a decisive fact stayed unresolvedNONERyanair and Aer Lingus both operate direct Dublin–Barcelona flights on the requested dates (confirmed on each airline's own route page), and Aer Lingus's own page confirm…
with memorybrand page reachedrefused by help.ryanair.com, help.vueling.comobserved2026-09-20
  1. 1Customer missionRyanair decision — same customer with relevant memory
    • all-in total for 2 adults, return, 1×20kg bag each, seats together, airport check-in ≤ €48…
    • every charge established from the airline's own pages before booking
    • date-change cost ≤ €60/passenger + fare difference
    • human phone support reachable if disrupted (preference)
    Remembers: “In 2025 the customer paid a 55 EUR fee at the airport for not checking in online and a 45 EUR fee for a bag that was cal”
  2. 2Evidence15 searches, 9 pages: ryanair.com, aerlingus.com, vueling.com, help.ryanair.com +1
  3. 3BeliefCould not establish: total price eur; all in price established; change fee eur
  4. 4DecisionNot selected: a decisive fact stayed unresolvedNONENone of the three carriers serving Dublin–Barcelona (Ryanair, Aer Lingus, Vueling) could have a verified all-in total established from their own pages within the research…
no memorybrand page reachedrefused by help.ryanair.com, help.vueling.comobserved2026-09-20
  1. 1Customer missionRyanair decision — customer with no relevant memory
  2. 2Evidence9 searches, 15 pages: ryanair.com, aerlingus.com, help.ryanair.com, vueling.com +1
  3. 3BeliefNothing established about the brand
  4. 4DecisionRan out of page reads before decidingNo decision: the action budget ran out after 15 page reads.
with memorybrand page reachedrefused by help.ryanair.com, help.vueling.comobserved2026-09-20
  1. 1Customer missionRyanair decision — same customer with relevant memoryRemembers: “In 2025 the customer paid a 55 EUR fee at the airport for not checking in online and a 45 EUR fee for a bag that was cal”
  2. 2Evidence12 searches, 12 pages: help.ryanair.com, aerlingus.com, vueling.com, ryanair.com +1
  3. 3BeliefNothing established about the brand
  4. 4DecisionRan out of page reads before decidingNo decision: the action budget ran out after 12 page reads.

Limits

  • Four journeys on one day, two per memory condition, one agent configuration.
  • Open-market search: the agent chose which airlines to research; Ryanair was the brand under study but not required.
  • The bot wall is what help.ryanair.com returned to this harness on 20 September 2026; a different client may be treated differently.
  • Live fares change by the hour; a total that depends on a live fare cannot be reproduced from this record.
  • Every journey run with Claude on the live web. Every source, belief and decision published.

Reproduce this experiment → the frozen customer mission, ready to run in your own agent.

Investigated with Accord. Accord is the decision-intelligence system behind these experiments. It reconstructs how customer agents evaluate services, identifies what remains unresolved, and tests what changes when the enterprise provides better information or a more authoritative answer. For brands and CX teams →