AI that uncovers hidden gems for DMCs

For a DMC, a distinctive product has become the main defence against price comparison by booking platforms and AI agents. Finding suppliers off those platforms means searching in the local language, away from the OTAs, then building access yourself. A Claude skill connected to Linkup does that first pass in 5 to 45 minutes.

Why has product become a DMC's real value?

For a long time, a receptive agency could make a living as an intermediary. Travellers weren't very independent, booking a small posada in Valladolid from Chicago was a hassle, and the DMC fairly charged for that work.

That's over. Booking platforms, and now AI agents, let any traveller rebuild an itinerary, find the suppliers and estimate your margin line by line. If my quote lists the same hotels and the same tours as three other DMCs, there's only one way left to choose between us: price.

I'm convinced a DMC's value is moving to its product. A strong product reflects who you are as an agency, it makes your offer hard to compare and, at best, impossible to reproduce without you. At Mexikoo, that work also changed things on the ground: by building out other regions, our product diversified beyond Yucatán. A more varied product also means depending less on the same three crowded sites.

Not comparable vs not reproducible: what's the difference for an inbound tour operator?

They're two different protections, and people mix them up. I call them the two walls of your product.

The first wall is being not comparable: nobody else offers this experience, so there's no reference price. The traveller can't put your quote next to another one and do the maths.

The second wall is being not reproducible: even if they know the idea, the traveller can't arrange it alone. Access goes through a person, a negotiated time slot, a relationship. No website, no "book now" button.

Here's an honest example. In Mérida, the skill suggested taking travellers to a Leones de Yucatán baseball game at the Kukulcán ballpark. The Mexican League season opens in April, bleacher seats start at 250 pesos for a three-game series [1], and the city goes as much for the atmosphere as for the game, with cheese kibis, polcanes and hojaldras sold in the stands [2]. I'd never put it in an itinerary, and I've lived here for eleven years. First wall: solid, no traveller thinks of it when planning a trip to Mexico. Second wall: weak, tickets are sold online. Once the idea is out, anyone can copy it.

What makes it hard to copy is how you build it: who you go with, what you eat, what you come to understand about a city and its team. A good product aims for both walls. One that only has the first gets copied.

Where do you find unique experiences that aren't on the platforms?

In the local language, away from the OTAs, and with people rather than operators. In practice, four rules:

1. Search in the local language. An English query about a Mexican village returns TripAdvisor. The same query in Spanish returns the state press, local journalists' blogs and cooperative directories.
2. Exclude the platforms during discovery. Booking, Viator, GetYourGuide, TripAdvisor and Airbnb come in later, only to check whether a supplier is already everywhere.
3. Use the sources nobody uses. Wedding and film location listings for characterful places to stay, regional press, craft associations, scientific field stations.
4. Run a second wave on proper nouns. A craftsman named in an article, a hamlet mentioned once, a place name spotted at the bottom of a page. That's where the best leads hide.

An experience is almost always a person. Without a proper noun, a lead is usually thin.

What does the hidden gems skill actually do?

It's the second skill in the series I announced in my comparison of AI tools for inbound tour operators, after the roadbook quality scan. If "skill" is new to you: think of it as a recipe for AI. A set of instructions written once, which the AI follows to the letter so you get the same result whoever runs it.

It runs in seven steps:

1. Intake, in four short screens. Places to stay, experiences or both, search depth, deliverable format, language, and most importantly the profile of your usual travellers. You can paste your current supplier list so it only brings you new ones.
2. Zone recon before spending. Give it a whole country and it first proposes four micro-zones where small characterful businesses are densest.
3. Off-platform search, in the local language and in English, across six families: making and craft, art and studios, sport and local passion, music and stage, nature and science, food and land. Left alone, a search drifts towards craft and food because they're the best indexed. The skill forces the balance.
4. A second wave on proper nouns spotted at the edges of the first results.
5. Hard filters, before any score. No source or real contact, and a lead goes to an appendix. No sign of life in 18 months, and it's flagged to check. On community encounters and animals it applies strict criteria, and anything it can't judge from a desk becomes a question to ask on the phone.
6. A score out of 21: rarity, character, story, an identifiable person behind it. Plus a simple test, the dinner test: will the traveller tell this story in one sentence when they get home?
7. The call sheet: contacts, sources, capacity, where it fits in an itinerary, and three specific questions for the first call.

One detail matters: no net rates, no contract or no invoice is never held against a supplier. It's actually a rarity signal, nobody has industrialised them yet.

The last section is, to me, the most valuable: ideas for experiences to create. It crosses what the place used to live off, what's unique about it, what happens there on fixed dates, and what everyone already sells. What shows up in the first three and never in the last one is your list. It doesn't design the experience: it points to a story and a door to knock on. The experience gets built with the person who holds the know-how.

Express, Standard or Deep: how long does it take?

• Express: one town, about 5 minutes, 10 to 15 records.
• Standard: a tight region, about 15 minutes, 20 to 30 records.
• Deep: a whole region, 40 to 45 minutes. Claude really digs into the corners of the internet.

It's a hungry skill. It uses Linkup credits and a good chunk of your Claude usage, especially in Deep. Start with an Express run on a town you know well: it's the best way to judge the quality of what comes back.

What did it actually find?

Two excerpts from real deliverables: one on my home ground, one on a destination I don't know.

Valladolid, Yucatán, Standard mode: 43 candidates reviewed, 7 kept. The first finding is the most useful: the centre is saturated. Six boutique hotels on the same street came out of a single query, all with the same pitch, and the skill advises not to waste time looking there. Then it surfaces three things I wasn't selling.

• Beneath the former convent of San Bernardino de Siena, which every visitor photographs, lies the Síis Já cenote, where INAH documented 153 weapons thrown into the water in 1847 and 1848, during the Caste War [3]. The skill immediately says it's not a diving experience: the site is protected. The value is in the story, and its suggestion is ninety minutes with INAH's underwater archaeology team, to be negotiated.
• The Amazonas de Yaxunah, a softball team of Maya women who play barefoot in huipiles and beat a university team 22 to 3 in Arizona in 2023 [4]. The skill scores it very high, then sets out three precautions: they're already well known, the risk of turning them into a show is the highest in the whole document, and only a format they decide is acceptable. That's exactly what I'd expect from a good colleague.
• An economic history that contradicts the standard story: eastern Yucatán was never henequen country but maize, cattle and sugar cane, and 110 of the 117 haciendas recorded around Valladolid closed during the war [5].

Ohrid, North Macedonia, Express mode: around 35 candidates, 3 confirmed gems, 4 solid leads, 3 to dig into. A family workshop making paper by hand with a technique that reached Ohrid in the 16th century, printed on a replica Gutenberg press, with a fragility flagged: it all rests on one person. Gjomleze, a layered pastry baked under an iron lid, the city's signature dish that the local press says you can barely find in its restaurants, tasted at the end of August in a nearby village where local women compete for the best recipe. And an idea to create: a morning with the Ohrid Hydrobiological Institute, which has studied a lake millions of years old since 1935.

What it leaves out matters just as much. In Ohrid, it deliberately dropped fishing with a local fisherman, because the lake is under strict concession and endemic species are being poached. And both times it says plainly what's missing: three of six families empty in Valladolid, no local discipline found in Ohrid.

I've lived in Mexico and run a DMC here for eleven years. To me, Yucatán's economic history was henequen, the green gold of the haciendas around Mérida. This run taught me that the east of the peninsula lived off something else: maize, cattle, and sugar cane turned into panela and aguardiente. And that cane owners taking over Maya maize land was one of the direct causes of the 1847 uprising. Our travellers pass through Valladolid without anyone telling them this, and I wasn't telling them either.

I didn't know the Amazonas de Yaxunah either, an hour from Valladolid. They started in 2018 with wooden planks for bats, playing against the village boys, and in 2022 they played at Kukulcán, the Leones' ballpark, the same one the skill had suggested for Mérida. What stays with me is that they play barefoot in huipiles because a uniform cost 1,600 pesos for two sisters, not for folklore. That's the kind of detail that changes how you tell a story, and whether you decide to offer it at all.

A few weeks ago I ran it live during an AI workshop in front of around forty DMCs, each on their own destination. The reactions in the room convinced me to share it.

Why do you need Linkup, and how do you set it up?

Because Claude or ChatGPT on their own aren't good at searching the web: few queries, often in English, and results dominated by big sites. Linkup is a search engine built for AI. It searches in the local language, excludes entire domains and reads full pages. Without it, the skill stops on purpose: working from memory means risking hotels that don't exist.

1. Create a Linkup account with your professional email and copy your API key. Eligible professional accounts receive a $20 starting credit under Linkup’s terms.
2. In Claude, open Customize > Connectors > Add custom connector.
3. Name it Linkup and use https://mcp.linkup.so/mcp?apiKey=YOUR_API_KEY, replacing the last part with your personal key.
4. Enable Linkup in a new conversation and ask it to find workshops in a town you know.
5. Download the ZIP version of the skill, then open Customize > Skills > Create skill > Upload a skill. Enable it and start with Express.

Your API key is personal: never share it. Team accounts may require an administrator’s permission.

→ Get the skill and full examples

Why will tomorrow's DMC have to create its own experiences?

Travellers increasingly want places others haven't seen, and the feeling that their trip was theirs. I'm convinced the DMC that lasts is the one that creates experiences that don't exist without it. Not a catalogue of suppliers, a point of view.

That takes time, and that's exactly where I see AI as leverage. The time it gives back on research goes into what it can't do: picking up the phone, going to see, building with people. With one limit to keep in mind: every gem has a volume beyond which it changes. If two hundred clients a year visit the same embroiderer, her workshop becomes a shop. The skill estimates that threshold for every record, and it's a number worth respecting, for her and for you.

Once called and validated, these suppliers belong in your Odys product database, where your quotes go to find them.

FAQ

How do you find suppliers that aren't on Booking or Viator? Search in the local language, exclude the platforms during discovery, and dig into regional press, craft associations and wedding venue listings. Then run a new search on every proper noun you spot.

What is a non-reproducible experience for a traveller? An experience they can't arrange alone even if they know it exists, because access goes through a person or an agreement only the agency holds.

Do I need a paid plan to use this skill? Skills and connectors work best on a paid Claude plan, and searches use Linkup credits. Linkup’s $20 starting credit for eligible professional accounts lets you test in Express mode.

Does the skill replace scouting on the ground? No. It replaces the internet research that comes before the field. The best gems still come from guides, drivers and site visits.

Why does Claude need Linkup to search the web? Its built-in search is limited in the number of queries and in control. Linkup searches in the local language, excludes sites and reads full pages, which changes what you find.

A gem you find online is just a name. It becomes product the day someone picks up the phone.

Sources

1. Quadratín Yucatán, prix de la série inaugurale 2026 des Leones
2. Yucatán Today, Leones de Yucatán et la cuisine du stade
3. Tus Buenas Noticias, l'INAH documente 153 armes de la guerre des castes dans un cenote de Valladolid (avril 2026)
4. Wikipedia (es), Amazonas de Yaxunah
5. El Bable, haciendas sucrières de Valladolid et d'Espita
6. Claude Help Center, connecteurs personnalisés
7. Claude · Skills
8. Linkup · Pricing