This year is the 20th anniversary of New York Times Travel’s signature piece of journalism: the “52 Places to Go” list, which we publish each January.
To mark the anniversary, we decided to look back at how the list — and travel itself — has changed over the past 20 years.
Through old-fashioned reporting, Tariro Mzezewa, a Times Travel alumna, found that the list was born almost by chance: Toward the end of 2004, Stuart Emmrich, the Travel editor at the time, said that he and his staff members had noticed that many New Yorkers were going to Bhutan, and wondered what other destinations could become popular.
They compiled a list called “Where to Go,” with no numbers attached, and published it on Jan. 9, 2005. From that point, the list became an annual event, varying from year to year until 2014, when it finally settled in at “52 Places to Go,” one for each week of the year.
But identifying themes and inflection points in the list-making was more complicated.
Two decades of destinations add up. Through 2024, we had named 914 total places, including 145 countries, 366 cities and towns, and 41 U.S. states, as well as some major events like last year’s total solar eclipse. All told, it came to more than 300,000 words.
To process all that information, we turned to The Times’s new artificial intelligence team. They specialize in helping reporters make sense of extremely large data sets, like hundreds of hours of video recordings of private meetings that revealed how an election-related nonprofit organization created a movement to spread falsehoods, or years of transcripts from Donald J. Trump’s public appearances that show how his language has changed. The same tools can be used on recommendations for hotels and restaurants, museum openings, can’t-miss cultural events and natural phenomena.
The first hurdle was collecting all those years of recommendations into one searchable file that we could share with Zach Seward and Dylan Freedman from the A.I. team. Times publishing systems have changed over the last two decades, so the lists were in different formats. Some were published using Flash, a technology that is no longer compatible with our systems. One of our editors had to manually copy and paste text into the file. Then it all had to be cleaned so that things like the instructions for headline formats or italic style didn’t create glitches.
We used several A.I. search engines, including Gemini, a large language model that can handle files of up to 750,000 words, and Semantra, an open-source “semantic search engine” that Mr. Freedman developed. Instead of searching for specific terms — “sustainability,” say, or “climate change” — it searches for concepts or themes. “It’s a new paradigm of search, not looking at keywords but trying to capture meaning,” Mr. Freedman said.
The Times has specific policies around the use of A.I., and nothing that comes straight from an A.I. program can appear in our articles, in part because of the possibility of hallucinations — more or less the program just making things up. So after running our queries through those search engines, Mr. Seward and Mr. Freedman turned the results over to Ms. Mzezewa. From her perspective, the technology was most helpful in identifying interesting nuggets within that mountain of text, like the effect of world events on the list, seen in the 2009 inclusion of Kabul, the capital of Afghanistan (we called it a “fragile city on the way to recovery”).
She found Semantra especially helpful “because it was giving more context over time,” she said, and let her see how we’d written about topics like overtourism and the rise of social media in travel, even if we hadn’t used those exact words.
For example, we’d asked the A.I. programs to identify instances when we’d written about sustainable travel. That term didn’t really exist when the list was started, but the concept of more environmentally friendly travel did. Among the examples the search engine turned up was Star Island in the Bahamas, which first made our list as the “eco-destination of the year” in 2009.
When we humans first started looking at the years of lists, certain themes had jumped out: The impact of smartphones and social media, the growing focus on climate change and the possible negative effect of travel, including overcrowding. The A.I. programs’ analysis pretty much mirrored our own, providing a kind of high-tech backstop to our journalist’s intuition.
Picking our list each year is a team effort that requires knowledge of trends in travel, an eye for great visuals and a sense of what people are looking for now on their journeys — to name just a few of the skills brought to bear. Artificial intelligence won’t be picking our Places to Go anytime soon, but it can help us understand where we’ve been.
The post How A.I. Tools Helped the Travel Team Study Its ‘Places to Go’ Lists appeared first on New York Times.
