Travel
Last reviewed
Sources
53 citations
Review status
Source-backed
Revision
v2 ยท 5,492 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
53 citations
Review status
Source-backed
Revision
v2 ยท 5,492 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Travel ChatGPT Plugins
Artificial intelligence in travel covers the use of machine learning, natural language processing, computer vision, and generative AI across the travel and hospitality industries. The technology touches almost every stage of a trip: research and inspiration, search and booking, pricing, customer service, airport identity checks, in-stay services, and even the vehicles that carry passengers from A to B.
Travel was an early adopter of computational decision support. Airlines began using yield management systems in the 1960s, and large hotel groups followed in the late 1980s. The current wave of public-facing AI products dates to the launch of OpenAI's ChatGPT in late 2022. Within a few months, Expedia, Kayak, Booking.com, Skyscanner, and Tripadvisor had all announced generative AI features, and a wave of trip planning startups (Layla, Mindtrip, GuideGeek, Wonderplan, OutOfOffice and others) had launched.
Not every rollout has gone smoothly. In February 2024 the British Columbia Civil Resolution Tribunal ordered Air Canada to compensate a passenger after its support chatbot had invented a refund policy that the airline did not honour. Researchers have documented tourists being sent to landmarks that do not exist, and consumer surveys still find that fewer than half of travellers fully trust AI recommendations.
The industry typically groups AI use into a few broad buckets:
| Area | Typical use of AI |
|---|---|
| Inspiration and planning | Chatbots, recommender systems, image and video tagging |
| Search and booking | Conversational search, ranking, fare and rate prediction |
| Pricing and revenue management | Demand forecasting, dynamic pricing, market segmentation |
| Operations | Disruption recovery, crew scheduling, demand-driven staffing |
| Customer service | Self-service chatbots, intent detection, voice IVR |
| Identity and security | Facial recognition, document checks, biometric boarding |
| In-trip experience | Translation apps, voice assistants, hospitality robotics |
| Mobility | Ride-hail dispatch, route optimisation, autonomous vehicles |
McKinsey's 2024 review found that 35 percent of the largest public travel companies referenced AI in their annual reports, up from only 4 percent in 2022. In a survey of 86 mostly US travel executives, 59 percent credited AI with increased employee productivity, and a majority reported more than 6 percent annual revenue growth and cost savings attributable to AI over the previous three years.
Consumer adoption has moved even faster. Phocuswright's 2024 research showed that the share of US travellers using generative AI for any topic jumped from 22 percent in 2023 to 39 percent in 2024, with travel being the single most common use case among generative AI users.
Long before chatbots, the travel industry was already using algorithms to set prices. American Airlines began researching reservation inventory in the early 1960s, but the watershed came after US airline deregulation in 1978. Faced with low-cost competitors such as People Express, American Airlines launched its Ultimate Super Saver fares on January 17, 1985, supported by a system called DINAMO (Dynamic Inventory Allocation and Maintenance Optimizer). Robert Crandall, then CEO, called yield management "the single most important technical development in transportation management since we entered deregulation." American Airlines Decision Technologies later won the 1991 Franz Edelman Award for the work, which generated an estimated $1.4 billion in incremental revenue over three years.
Hotels were watching. The Cornell Hospitality Quarterly published the first article applying airline yield management to lodging in 1988, and Marriott built a revenue management team to forecast demand across its 100,000 plus rooms. The term softened from "yield management" to "revenue management" as the practice spread to car rental, cruise, and resort businesses. Vendors such as IDeaS (founded 1989), PROS Holdings, and Sabre's hospitality unit grew up around these markets.
The 2000s and 2010s brought a second wave: collaborative filtering, gradient-boosted trees, and deep learning. Hopper, founded by ex-Expedia executives Frederic Lalonde and Joost Ouwerkerk in Montreal in April 2007, spent close to a decade in stealth building a price database before launching a mobile app in 2015 that predicted whether to buy or wait. The company claims roughly 95 percent prediction accuracy up to a year out, based on trillions of historical fare records.
During the same period, Airbnb rolled out Smart Pricing (using gradient-boosting machines to predict per-listing demand), search ranking models, and image classification for property photos. Online travel agencies refined recommendation systems for hotels and flights, and revenue management vendors began branding their products as "AI" rather than "operations research."
The public launch of ChatGPT in November 2022 reset the conversation. Within four months, Kayak and Expedia announced ChatGPT plugins (March 2023). Booking.com followed with its own AI Trip Planner on 28 June 2023. Google introduced travel extensions for Bard at Google I/O 2023, eventually wiring it into Google Flights, Google Hotels, and Google Maps. By late 2023 standalone trip planning bots, white-label assistants for tourism boards, and an avalanche of AI features inside existing apps were live or in beta.
Generative AI made the "natural language trip planner" a standard product category. Most of these tools take a destination, dates, traveller count, budget, and interests, then produce a multi-day itinerary, sometimes with hotels and flights linked through partner APIs.
| Product | Launched | Origin | Notable features |
|---|---|---|---|
| Roam Around | 2022 | Independent, later acquired by Layla | Early viral itinerary bot built on the OpenAI API |
| GuideGeek | April 2023 | Matador Network | Messaging-first via WhatsApp, Instagram, Facebook Messenger; white-label for tourism boards |
| Vacay (Vacay Chatbot) | 2023 | Independent | GPT based travel advisor focused on destination ideas |
| Wonderplan | 2023 | Independent | Free web planner with PDF export and offline access |
| Tripnotes | 2023 | Independent | Note-taking style planner with auto-mapping and budget tracking |
| Mindtrip | May 2024 (public) | Silicon Valley, founded 2023 | Inspiration from screenshots, TikTok, YouTube; investors include Amex Ventures, Capital One Ventures, United Airlines Ventures |
| Layla | November 2023 | Berlin, Jeremy Jauncey of Beautiful Destinations and Saad Saeed | Trained on the Beautiful Destinations creator library; acquired Roam Around |
| OutOfOffice (OOO) | August 2021, AI trip generator added later | Chicago | Personalised quiz, packing lists, recommendations across 3,500 cities |
| Tripadvisor AI Trip Builder | July 2023 (rebuilt August 2024) | Tripadvisor | Custom recommender model trained on Tripadvisor reviews; latency cut from about 40 seconds to 6.5 |
Tripadvisor reported in late 2023 that users of its AI itinerary tool generated roughly three times the revenue of an average user. The 2024 rebuild replaced a generic large language model layer with an in-house recommender that searches Tripadvisor's review corpus directly, which doubled the rate at which travellers saved recommendations.
Funding around these products has been substantial for an early category. Mindtrip raised about $7 million in seed funding led by Costanoa Ventures in 2023 before its 2024 public launch. Layla closed a 3 million euro seed round in late 2023 with backers including Paris Hilton and a group of travel industry investors.
The large OTAs took two distinct paths after ChatGPT launched. Some integrated with OpenAI's plugin system, exposing inventory to a third-party assistant. Others built conversational search inside their own apps using the OpenAI API or in-house models.
Expedia Group was on both sides. In late March 2023 it launched an Expedia plugin for ChatGPT, giving Plus subscribers access to live flight, hotel, car, and activity availability inside a ChatGPT conversation. A few days later it added a ChatGPT-powered trip planning experience to its iOS app, automatically saving hotels discussed in the conversation to a "trip" object inside the app.
Kayak shipped a ChatGPT plugin around the same time and later opened a sandbox called KAYAK.ai. By 2025, Kayak had also released its own consumer-facing AI travel planner. The company emphasised that its ChatGPT integration let users ask things like "Where can I fly to from New York for under $500 in April?" and get personalised answers grounded in Kayak's historical search data.
Booking.com shipped its AI Trip Planner on 28 June 2023, beta-launching to a slice of US Genius loyalty members. The tool combined Booking's existing machine learning stack with the OpenAI API, surfacing destinations and properties with live pricing inside a chat interface. Booking rolled out a substantial update in October 2024 that added support for more granular conversations, including ranking specific hotels against each other. Booking later announced an expanded partnership with OpenAI on personalisation.
Skyscanner launched its Savvy Search tool, which sits on top of an estimated 18 million unique flight routes and 80 billion prices searched daily. Users can type prompts like "foodie city breaks" or "short flights next weekend" and receive up to three curated destination ideas that handoff to a regular flight search.
Tripadvisor launched its AI Trip Builder in July 2023 and shipped a major rebuild in August 2024. The 2024 version uses a recommender model trained directly on Tripadvisor's review corpus rather than passing prompts through a general purpose LLM, which cut latency from around 40 seconds to about 6.5 seconds. Tripadvisor also rolled out a dedicated Tripadvisor app inside the OpenAI GPT Store and announced a partnership with Perplexity AI.
Google's travel push at Google I/O 2023 wired Bard (later renamed Gemini) into Google Flights, Google Hotels, Google Maps, Gmail, and YouTube. A user can ask Gemini to find dates that work for everyone in a Gmail thread, look up live flight and hotel prices, get walking directions to the airport, and watch destination videos in the same conversation. Google has since expanded this with an "AI Overviews" treatment for travel queries in regular search.
New entrants such as Perplexity AI and OpenAI's Operator agent have also pushed into travel. Skyscanner launched a dedicated Skyscanner app inside ChatGPT in 2026, joining a wave of conversational booking integrations that move the booking session out of OTA websites and into general purpose chat assistants.
Dynamic pricing is the area where AI has the longest track record in travel. Modern revenue management systems use a stack of forecasting models, optimisation engines, and competitive intelligence feeds to set or recommend rates by day, market segment, channel, and length of stay.
Major revenue management vendors include:
| Vendor | Founded | Notes |
|---|---|---|
| IDeaS (SAS Institute) | 1989 | Industry incumbent in large-property hotel revenue management; uses neural forecasting and operations research methods |
| Duetto | 2012 | Pioneered "Open Pricing" with continuous rate movement instead of restricted rate plans; acquired by Accel-KKR in 2019 |
| PROS Holdings | 1985 | Airline pricing and shopping optimisation; pricing for cargo, B2B, and travel |
| Revenue Analytics | 2005 | Hotels, cruise, casinos, and broader hospitality |
| Atomize, Pace, RoomPriceGenie | 2010s | Smaller property and short-term rental segments |
| Sabre | 1960 (as joint IBM/AA project SABRE) | Distribution and revenue tools used by airlines and hotels; SynXis hotel platform |
On the airline side, the airline-specific stack still descends from American Airlines' DINAMO and the broader research program at American Airlines Decision Technologies. Modern systems blend rule-based fare classes with deep learning demand forecasts and reinforcement-learning style willingness-to-pay models. PROS, Sabre, Amadeus, and Lufthansa Systems are the largest off-the-shelf vendors. Larger carriers usually pair these with in-house data science teams. Several airlines, including Lufthansa, have announced moves toward dynamic offer creation under the IATA New Distribution Capability standard, where AI generates a tailored bundle (fare, ancillaries, loyalty earnings) for each shopping request rather than picking from a fixed catalogue.
For hotels, the dominant trend has been the shift from once-a-day rate decisions to continuous repricing across hundreds of rate plans, with AI handling segmentation and competitor monitoring. Vendors increasingly position their products as advisers to revenue managers rather than full autonomous pricers; in practice, only a minority of hotels accept system recommendations without review.
Airbnb is the canonical machine learning case study in short-term rentals. Smart Pricing, introduced in 2015, uses a gradient boosting model to estimate a separate demand curve for each listing using property characteristics, neighbourhood features, seasonality, local events, and competitor prices. The model maps the predicted curve onto a recommended price for each future date, subject to host-set minimum and maximum bounds. Airbnb's published case study reports that listings using Smart Pricing get 5.7 percent lower average prices for guests but more bookings overall.
Beyond pricing, Airbnb uses machine learning for search ranking, fraud detection, photo categorisation (recognising kitchens, pools, bedrooms), automatic translation of listings and reviews, customer service routing, and host matching. The company prototyped a more proactive AI assistant called Aiden around 2017, but never released it as a flagship product. In 2024 and 2025, CEO Brian Chesky talked publicly about turning the Airbnb app into a more agent-like "personal concierge," and in 2026 Airbnb shipped an AI-powered customer service assistant for guests and hosts.
Competitors have followed. Vrbo, Booking.com's home rental segment, and dedicated short-term rental tools such as Beyond, PriceLabs, and Wheelhouse all offer dynamic pricing for individual hosts.
Facial recognition has spread quickly through airports since the late 2010s. In the United States the rollout has come in two channels: government programs run by the Transportation Security Administration and Customs and Border Protection, and private programs run by carriers and partners such as CLEAR.
The TSA began using facial recognition at security checkpoints in 2019 through its Credential Authentication Technology version 2 (CAT-2) kiosks. By the end of 2024 the agency had deployed CAT-2 kiosks at roughly 80 US airports and signalled plans to expand to more than 400 federalised airports. Travellers can opt out and request a manual ID check. Department of Homeland Security testing in 2023 and 2024 found the system worked more than 99 percent of the time across a 1,600 person test population, although the same testing flagged that self-identified Black volunteers had the lowest match success rate of any demographic group. In November 2023 a bipartisan group of US senators led by John Kennedy and Jeff Merkley introduced a bill to ban TSA's use of facial recognition.
Customs and Border Protection runs a separate program called Simplified Arrival, which uses facial comparison to verify travellers against existing passport and visa photos at the inspection booth. An initial pilot ran at Washington Dulles in 2015. By the early 2020s CBP had completed rollout at every US international airport. The agency reports that more than 171 million travellers have been processed through facial comparison and that the system has identified more than 1,450 imposters using legitimate travel documents that belonged to other people. Photos of US citizens are deleted within 12 hours; photos of most non-citizens are retained in a DHS system. Simplified Arrival has expanded to land and sea ports of entry and to facilities in cooperating countries.
CLEAR, founded as Verified Identity Pass in 2003 and relaunched after bankruptcy in 2010 by Caryn Seidman-Becker and Ken Cornick, operates dedicated biometric lanes at security checkpoints. As of 2024 the company reported around 38 million enrolled members and lanes at more than 50 US airports, plus stadiums and arenas. CLEAR began rolling out biometric eGates in partnership with the TSA in 2025.
Airlines run their own biometric programs. Delta Air Lines launched Digital ID at Atlanta and Detroit in 2021, in partnership with the TSA, allowing eligible TSA PreCheck members to clear bag drop and security by face match alone. Delta has since extended it to Salt Lake City, Los Angeles, JFK, LaGuardia, and Reagan National. The airline reports that biometric bag drop typically takes about 30 seconds instead of two minutes.
In Europe, Lufthansa Group and Swiss became the first carriers to use Star Alliance Biometrics in November 2020, starting at Frankfurt and Munich and later extending to Hamburg and Vienna. The system is built on NEC Corporation's I:Delight platform; passengers enrol in the Star Alliance app and then pass security, lounge, and boarding touchpoints with a face match. Other large airports in the Middle East and Asia (Singapore Changi, Dubai, Doha) have rolled out similar end-to-end biometric corridors, sometimes in cooperation with home carriers.
These systems remain controversial. Civil liberties groups including the ACLU and Electronic Frontier Foundation have repeatedly pushed back on TSA's use of facial recognition, citing accuracy disparities across demographic groups and the risk of normalising face matching as a baseline identity check. Many programs are formally voluntary, but advocates argue that signage and time pressure can make opting out effectively impractical.
Long before generative AI, travel was one of the busiest categories for rule-based chatbots and voice IVR. Several systems are well documented in industry case studies:
| Product | Launched | Operator | Channel | Notes |
|---|---|---|---|---|
| Amtrak Julie | 2001 | Amtrak, built by Next IT | Voice IVR and later web chat | Reported to generate roughly 30 percent more revenue per booking than human agents and save Amtrak around $1 million per year in customer service costs |
| Alaska Airlines Jenn | 2008 | Alaska Airlines, built by Next IT | Web chat | Reported as the first US airline chatbot |
| KLM BlueBot (BB) | 2017 | KLM | Facebook Messenger | Helps customers book tickets, supported by 250 human agents who take over when the bot escalates |
| Lufthansa Mildred | 2016 | Lufthansa | Facebook Messenger | Built by a single developer; helps customers search fares and find general flight information |
| Marriott chatbots | 2017 | Marriott | Facebook Messenger, Slack, later WeChat and Google Assistant | Loyalty-focused bot tied to Marriott Rewards |
| Aloft ChatBotlr | 2017 | Aloft Hotels (Marriott brand) | SMS | In-room service requests; Marriott reported that two thirds of Aloft guests interacted with it within early pilots, with a five second average response time |
| Hilton Connie | 2016 | Hilton and IBM | In-lobby robot | First Watson-enabled hotel concierge, replaced over time at the Hilton McLean pilot location |
Since 2023 most of these have been rebuilt on top of large language models or replaced with new generative AI experiences. The category is increasingly opaque from outside: hotel groups such as Marriott, Hilton, IHG, and Accor have all announced internal copilots for staff, and most large airlines and OTAs have replaced rules-based chatbots with LLM-driven systems that hand off to humans when uncertain.
Operators have learned the limits the hard way. Voice IVR systems still misroute callers, and generative chatbots can confidently invent policies. Adoption is uneven; even when bots are present, a large share of high-value calls still ends up with a human agent.
Lobby and room service robots have been an area of fascination and disappointment. The most famous example is the Henn-na Hotel chain in Japan, which opened its first property at Huis Ten Bosch in Nagasaki in July 2015 and was certified by Guinness World Records in 2016 as the world's first robot-staffed hotel. The original lineup included animatronic dinosaur receptionists, luggage robots, and an in-room doll named Churi.
In 2019 the hotel reduced its 243 robot workforce by more than half. The doll was decommissioned after guests complained it could not answer basic questions. The dinosaur receptionists were retired because they could not photocopy passports, which is a legal requirement at Japanese check-in. Aging robots also became expensive to repair. The chain has continued to operate, but with fewer robots and more humans.
Hilton's 2016 partnership with IBM produced Connie, named after Conrad Hilton, the first Watson-enabled hotel concierge. The 23 inch SoftBank NAO robot used IBM Watson APIs and WayBlazer's travel knowledge graph to answer guest questions about nearby attractions, restaurants, and hotel services, learning from each interaction. Connie was piloted at the Hilton McLean in Virginia. The robot was eventually replaced by other devices, and Hilton later focused its AI investment on its Connie loyalty data platform and digital key features instead of physical robots.
Other experiments include Savioke's Relay robot, which delivered items to rooms at Aloft and Crowne Plaza properties starting in 2014, and SoftBank's Pepper, which was deployed in some hotel lobbies in Asia. Most of these projects have ended or shrunk. The current generation of hospitality robotics is more mundane: floor cleaners, automated kitchen equipment, and warehouse-style robots in back of house, often built by Bear Robotics, Pudu Robotics, or Mira Robotics.
Machine translation has been one of the most useful AI tools for individual travellers. Google Translate launched on 28 April 2006 and reached 250 supported languages and one billion monthly users by its 20th anniversary in 2026. Its mobile camera mode, which overlays translations onto signs and menus in real time, became iconic in the late 2010s. The app supports offline downloads, handwriting input, conversation mode, and a phrasebook.
Skype Translator launched in beta on 15 December 2014, initially for English and Spanish, after Satya Nadella demonstrated it at the Code Conference earlier that year. The system used deep neural networks for speech recognition combined with statistical machine translation. Skype Translator was eventually folded into Microsoft Translator, whose APIs now power most Microsoft real-time translation features.
DeepL, Apple Translate, and on-device LLMs (Apple Intelligence translation, Samsung Live Translate) have since pushed the quality of on-the-go translation higher. Earbuds with on-board translation (Google Pixel Buds, Timekettle WT2, Pocketalk) have become a small but growing category.
On-trip AI assistants also extend into navigation. Google Maps uses machine learning for ETA prediction, lane-level guidance, and immersive view (a generated 3D fly-through of a route or landmark). Uber and Lyft use deep learning for demand prediction, dispatch, and surge pricing. Airbnb, Tripadvisor, and Yelp use AI for personalised recommendations.
Autonomous and semi-autonomous vehicles affect travel through three channels: airport transfers and city rides at the destination, long-haul freight (which indirectly affects supply chains and air cargo), and consumer experience features such as adaptive cruise and automated parking in rental cars.
Waymo, Alphabet's robotaxi unit, is the most active in the consumer travel context. As of mid-2026 Waymo offered paid rides in roughly ten US cities including the San Francisco Bay Area, Los Angeles, Phoenix, Austin, Atlanta, Miami, Houston, Dallas, San Antonio, Orlando, and Nashville. The company reports around 500,000 paid rides per week and was valued at about $126 billion in a 2025 funding round. Waymo has expanded into airport pickup zones at Phoenix Sky Harbor and Austin-Bergstrom, and has announced plans for Philadelphia, JFK, and Newark with mixed reactions from city officials. The company issued a voluntary recall of 3,791 vehicles in May 2026 after one of its robotaxis drove into a flooded road in San Antonio.
Other robotaxi efforts are at different stages. Cruise, GM's robotaxi unit, was shut down in late 2024 after a 2023 incident in San Francisco. Zoox, owned by Amazon, has been running limited service in Las Vegas. Apollo Go (Baidu) and Pony.ai operate paid robotaxis in several Chinese cities. Tesla has talked about a Cybercab service but as of mid-2026 had not launched a free of supervisor commercial robotaxi network. Several airports, including Las Vegas Harry Reid International, have run autonomous shuttle pilots inside terminal areas.
In February 2024 the British Columbia Civil Resolution Tribunal issued a closely watched decision in Moffatt v. Air Canada. After the death of his grandmother in late 2022, Jake Moffatt asked the AI chatbot on Air Canada's website about bereavement fares. The chatbot told him he could buy a regular ticket and apply for a bereavement refund within 90 days of travel. The airline's actual policy on a separate linked page said bereavement claims could not be made retroactively. Air Canada refused the refund.
Moffatt sued in the small claims tribunal. Air Canada argued that the chatbot was "a separate legal entity" responsible for its own actions, and pointed at the linked policy page as the authoritative source. The tribunal rejected the defence, ruling that Air Canada was responsible for the information on its own website regardless of whether it came from a static page or a chatbot, and that the airline owed Moffatt a duty of care that it had breached. Moffatt was awarded approximately CA$650 in damages plus interest and tribunal filing fees.
The case has been cited extensively in legal commentary as the first published consumer-rights decision finding a company directly liable for misinformation generated by its AI chatbot. It is taught in business law classes and referenced by enterprise legal teams when reviewing internal AI policies.
Reporting in 2024 and 2025 documented several incidents in which generative AI sent travellers to places that do not exist or gave dangerously wrong logistics. A pair of tourists hiking in Peru were stopped by a local guide on their way to a non-existent "Sacred Canyon of Humantay" that ChatGPT had described in detail. A Japanese couple followed ChatGPT's advice to start a Mount Misen hike at 3 pm to catch the sunset, only to discover that the ropeway closed earlier than the chatbot claimed. A Malaysian couple drove 400 kilometres to a fictitious "Kuak Skyride" generated in an AI travel video. A 2026 survey by booking platform Klook of about 11,000 global travellers found that around 91 percent had used AI for trip planning but only 35 percent fully trusted its outputs.
DHS's own testing has flagged demographic disparities in face matching accuracy at TSA checkpoints, with self-identified Black travellers experiencing lower match rates than other groups even when overall accuracy is high. Civil rights groups and a bipartisan group of US senators have called for the program to be paused, and several airports have published opt-out signage in response to advocacy pressure.
A 2024 study by Originality.AI estimated that the share of likely AI-generated Tripadvisor reviews grew roughly 137 percent between 2019 and 2024. Tripadvisor, Booking.com, Yelp, and Google have all invested in machine-learning content moderation to filter out generated reviews, but the cat-and-mouse dynamic continues.
A few themes recur in industry and academic discussions:
The industry consensus heading into the second half of the decade is that AI is moving from a copilot role to an agent role: from suggesting itineraries to actually executing bookings, negotiating cancellations, and rebooking disrupted travel without human prompting. Whether consumers, regulators, and operators are ready for that shift is still being worked out.