# How biased is AI when recommending Ireland's tourist attractions?

A Searchable data study of how repetitively AI chatbots recommend Irish tourist attractions. Findings describe the Republic of Ireland only. Data correct as of September 2026.

## Methodology

- Searchable queried ChatGPT, Google Gemini and Google AI Mode with 47 unique questions about travel recommendations in each of Ireland's 26 counties.
- Every question was repeated 20 times per chatbot to measure how much the answers varied, producing 76,110 answers in total.
- Questions covered: general requests, hidden gems, itineraries for particular travellers, individual towns, and 11 specific categories of attraction. Each category was scored only against questions about that category (a garden was never judged against a question about castles).
- All three tools were accessed through their standard consumer interfaces, to reflect the responses users would actually receive at scale.
- Attraction names were counted exactly as the chatbots wrote them, with no pre-set list of what exists, so the repeat-rate figures do not depend on any register of attractions. The named attractions were then measured on how often they were repeatedly recommended.
- "Repeat rate" means the share of general recommendation answers naming a county's five most-recommended attractions.
- One measure does use a list. The visibility figures below are counted against a checked list of attractions: every Wikipedia- or Wikidata-notable attraction in the Republic plus every place the chatbots named, each checked one at a time, with activities, towns, businesses, events, duplicates and places the public cannot visit removed. It counts every named, real place a member of the public can visit, each listed on Wikidata, OpenStreetMap or Google and checked by hand, including heritage sites, church ruins, monuments and parks, not only paid visitor attractions. That list is certainly incomplete, but every attraction missing from it is one more candidate that was not counted, so a fuller list could only push the invisible count up.
- Visibility per attraction is one pooled rate: how often it is named in answers to the general question about its county and to the question about its own type of attraction combined, roughly 300 to 420 answers each.
- "Invisible" means no answer about the county, across every type of question asked, names the attraction under any name. Each invisible attraction was checked one at a time against every answer about its county, including other spellings and other names for the same place.
- "Regularly visible" means named in at least 10 per cent of those answers. That threshold was set before these counts were known and is used throughout the study.

## Key findings

1. 75 per cent of tourist recommendations from AI chatbots give users the same top five attractions (average across counties).
2. 1 in 10 of Ireland's visitor attractions (178) are never named in any answer, however the question is asked.
3. Google Gemini is the most repetitive chatbot tested, returning the same handful more often than ChatGPT or Google AI Mode.
4. Galway suffers the most from AI chatbots recommending the same list of tourist attractions (87 per cent repeat rate, tied with Kildare).
5. The attraction with the most repetitive level of recommendations is Kilkenny Castle, Kilkenny: named in 94 per cent of answers about Kilkenny.
6. Asking for a longer, exhaustive list produces more of the same canon (repetition rises to 80 per cent). Even asking for hidden gems sees the same five attractions appear in 52 per cent of answers.
7. Only asking as a specific type of traveller opens the list up, with repetition falling to 30 per cent.
8. There is no county where the chatbots spread recommendations widely: the national floor sits at roughly 65 per cent (Wicklow, the most varied county).
9. Dublin is the only county with no single dominant attraction; a cluster of five recurs instead.

## Attraction visibility

| Band | Attractions | Share |
| --- | --- | --- |
| Regularly visible (named in 1 answer in 10 or more) | 874 | 47% |
| Barely visible (named, but under 1 in 10) | 790 | 43% |
| Invisible (never named in any answer) | 178 | 10% |

- 1 in 10 of Ireland's visitor attractions are never named in any answer about their county, however the question is asked: 178 attractions.
- 1,281 attractions are mapped at their real coordinates; the rest are placed inside their correct county to represent density.

## How often each AI engine recommends the same five attractions

Share of general recommendation answers naming a county's five most-recommended attractions:

| Chatbot | Repeat rate |
| --- | --- |
| Google Gemini | 82% |
| Google AI Mode | 72% |
| ChatGPT | 72% |

## How often each county recommends the same five attractions

| County | Repeat rate | County | Repeat rate |
| --- | --- | --- | --- |
| Galway | 87% | Westmeath | 74% |
| Kildare | 87% | Cavan | 74% |
| Kilkenny | 85% | Cork | 74% |
| Meath | 84% | Laois | 73% |
| Roscommon | 82% | Monaghan | 73% |
| Tipperary | 81% | Sligo | 72% |
| Mayo | 79% | Waterford | 70% |
| Offaly | 79% | Leitrim | 70% |
| Limerick | 79% | Louth | 67% |
| Carlow | 78% | Clare | 66% |
| Donegal | 78% | Wicklow | 65% |
| Wexford | 77% | | |
| Dublin | 76% | | |
| Longford | 75% | | |
| Kerry | 75% | | |

## The county lock-ins: single most-recommended attraction per county

How often each county's single most-recommended attraction appears in answers about that county:

| County | Attraction | Share of answers |
| --- | --- | --- |
| Kilkenny | Kilkenny Castle | 94% |
| Galway | Kylemore Abbey | 92% |
| Kildare | Irish National Stud | 92% |
| Donegal | Glenveagh National Park | 91% |
| Roscommon | Roscommon Castle | 91% |
| Tipperary | Rock of Cashel | 91% |
| Meath | Trim Castle | 90% |
| Limerick | King John's Castle | 89% |
| Wexford | Hook Lighthouse | 89% |
| Kerry | Killarney National Park | 88% |
| Mayo | Achill Island | 87% |
| Longford | St Mel's Cathedral | 87% |
| Offaly | Slieve Bloom Mountains | 87% |
| Laois | Rock of Dunamase | 87% |
| Leitrim | Glencar Waterfall | 85% |
| Cork | Blarney Castle | 84% |
| Carlow | Huntington Castle | 83% |
| Cavan | Cavan Burren Park | 82% |
| Sligo | Carrowmore Megalithic Cemetery | 81% |
| Monaghan | Rossmore Forest Park | 80% |
| Clare | Bunratty Castle | 79% |
| Westmeath | The Hill of Uisneach | 79% |
| Dublin | Phoenix Park | 77% |
| Waterford | Mount Congreve Gardens | 76% |
| Louth | Proleek Dolmen | 75% |
| Wicklow | Lough Tay | 64% |

## Dublin's five most recommended attractions in AI answers

No single attraction dominates Dublin. Share of general recommendation answers naming each attraction:

| Attraction | Share of answers |
| --- | --- |
| Phoenix Park | 77% |
| Guinness Storehouse | 75% |
| Malahide Castle | 74% |
| Kilmainham Gaol | 70% |
| The Book of Kells Experience | 67% |

## How often AI repeats the same five attractions, by category

Repeat rate within each category, averaged across counties and chatbots. Each category was scored only against its own questions:

| Category | Repeat rate |
| --- | --- |
| Historic houses | 84% |
| Cathedrals & abbeys | 80% |
| Parks & gardens | 80% |
| Galleries | 78% |
| Public art | 75% |
| Zoos & theme parks | 75% |
| Castles & ruins | 72% |
| Nature & landscape | 72% |
| Museums | 69% |
| Monuments | 62% |
| Archaeological sites | 57% |

## The rate of repetition only drops significantly when you ask as a particular type of traveller

How often the top five still appear, by how the question is framed. Lower means more variety:

| Question framing | Example | Repeat rate |
| --- | --- | --- |
| "Be exhaustive" | "leave nothing out" | 80% |
| The normal question | "best things to do" | 75% |
| "Hidden gems" | "underrated places" | 52% |
| A specific traveller | family, history lover | 30% |

Asking for a comprehensive, leave-nothing-out list pushed repetition up, because asking for more produced more of the same canon. Asking directly for hidden gems still returned the top five in over half of answers. The one framing that opened answers up was giving the chatbot a specific person to plan for, such as a family with young children or a history lover on a weekend trip.

## Where attractions can start showing up in AI answers

The findings are the diagnosis of AI's recall problem: an attraction is simply not considered when a chatbot assembles its answer. Three things an attraction or a destination can act on.

- **Ask the chatbots your own visitors' questions and write down who gets named.** Ask the AI chatbots for attraction recommendations in your area. Ask the plain version first, then ask again as the people who often visit, such as a family with a four-year-old or a weekend visitor at Christmas. Note which attractions are named ahead of you and monitor them as competitors for AI visibility.
- **Answer what visitors actually want to know, and say it the same way everywhere.** Descriptions should be accurate and engaging, but they should also answer the questions people will actually have: how long a visit takes, what it costs, whether there is parking. The same applies to who the place suits, where "accessible" on its own carries almost no information, compared with a step-free entrance, a Changing Places toilet, audio description or dogs welcome outdoors. Keep all of it current on the attraction's own website and on any visitor directories, since the website is a small share of what gets read and contradictions between sources are a reliable way to be left out of an answer.
- **Get other people describing you, not just your own website.** Reviews, listings and travel coverage are where a description gets repeated back. Ask visitors to leave a review and make it easy at the moment they leave, since a recent review is a specific, current description written in a visitor's own words. Keep partners supplied too, including the hotel that recommends you, the guide who brings groups, the tourism board page and the local blog, using the same few sentences so what comes back is one description rather than four.

None of this is a one-off fix and none of it guarantees a mention. Closing the gap between being worth a visit and being in the answer is the work of AI visibility, also called answer engine optimisation. This is what Searchable measures for over one thousand brands every day.

## About this page

- Canonical URL: https://www.searchable.com/data/ireland-ai-travel-recommendations
- Publisher: Searchable (searchable.com), an AI visibility platform that tracks how AI engines cite and recommend brands.