The Hilarious Need for Adequate AI-Powered Search: A One-Query Microcosm

Image conjured by Chris Lindland

We're heading off to New Zealand and Australia in a few days, which triggered an online quest to find quirky but comfortable, cool yet peaceful, nice but affordable hotels in multiple cities. 

This sort of project was once a delight. The web used to be packed (and perhaps still is?) with countless voice-y, independent travel blogs and one-off posts about this or that city. A chorus that sounded off personally and credibly about what to do, where to stay, and what to eat in every corner of the world. Finding the voices that most spoke to me was part of the fun, and a sure route to recommendations that would truly resonate with us.

No more. 

Google itself and the SEO slums have so polluted the search ecosystem that "Best Hotels Auckland" leads to a bottomless page of listicles spewed out by click-farms feasting on affiliate fees. Virtually every page Google points to is built around links honed to maximize funnel economics — and not traveler (or travel-writer) delight. 

Some of these pages fly under the flags of big publications, faded travel brands, or even indie-like names. And no doubt a few needles in this haystack do contain authentic opinions by sophisticated travelers. But finding those needles means reading through countless gushing hotel descriptions recycling the same effusive adjectives, most of them seemingly written by the same person or algorithm. 

It's gotten really bad since I last did this. 

I didn't end up picking our hotels at random. But this project took way more time than it used to, was 98% less fun, and it's much less likely we'll land in places that truly reflect our tastes and interests.

Then, a brainstorm:

Why not turn to ChapGPT for help? Not for hotel recommendations (which would no doubt be powered by the same Googley dross). But for the voice-y independent travel blogs and posts that must still be out there somewhere.

I did that. And boom - three exhilarating links to the travel web of yore:

Megan on Location and Alex on the Map are the sorts of names that were all over Google a decade back — and Kiwi Collection just had to be a local gem carved by proud and candid locals!

To review the telepathically-perfect recommendations that all this led to, we’ll start with not just one, but every "insider tips from a local" that Megan will ever have to offer:

Not only is this page missing, but so is Megan herself, as both she and her site are OpenAI hallucinations through and through.

As for Alex, at least her 404 page has nice fonts:

Alex is also probably a real person (although constant pop-ups on her site’s actual pages asking me to chat with her pronto do make me wonder). But she has written about precisely six countries, and her closest brush with Auckland was Bangkok.

The Kiwi Collection link is another imaginary page. The site does exist — but it’s based 7,000 miles northeast of Auckland in British Columbia, and is mostly about booking hotel rooms via some unidentified white-label engine that no doubt keep the lights on by kicking back affiliate fees.

When I first saw ChatGPT’s convincing and promising links, I was SO excited to write a giddy article about a new freeway to authenticity and the end of Search as we know it. And I’m really sorry I’m unable to share something like that with you today.

But in addition to OpenAI’s unreadiness for prime time in this specific area, my experience does highlight how thoroughly broken the travel info web has become. Google has badly failed us in this area over the past decade, and the world is ripe for a solution.

And I believe that solution does actually lie in conversational interfaces. OpenAI’s language models are bewitchingly powerful, and point the way to a solution that I do think will be honed with fine-tuning and other LLM magic. An article for another day.