I Asked AI to Plan My Day Trip. It Never Looked at a Single Destination Website.

Heidi Schlag
May 20
10 min read

What this means for your heritage organization, museum, or DMO, and why your beautiful homepage may be invisible to the tools your visitors are actually using.

I opened up Claude to draft an email, and a prompt button caught my eye: Plan a Trip.

I clicked.

As a heritage tourism consultant, I've been working to understand how AI is changing trip-planning behavior, so this felt like field research. What followed gave me a lot to think about.

Claude planned a near-perfect day trip from Frederick in about two minutes, and it never visited a destination website.

Below, I walk through the full interaction, then break down how Claude made its recommendations and what museums, byways, heritage areas, small towns, and DMOs need to change to appear in the new AI algorithms.

The Conversation with Claude About My Trip

Claude asked me three multiple choice questions -- below are my answers:

I choose my hometown of Frederick, MD, so I could better truth-test Claude's recommendations.

Next question was a list of possible destinations:

And a few more questions

Here is the itinerary it gave me (Claude output is in blue):

Harpers Ferry, WV — A Leisurely Mix

Distance: ~25 minutes from Frederick Pace: unhurried, no reservations needed

A casual day exploring where the Potomac and Shenandoah meet — a bit of history, a bit of hiking, lots of wandering, and a couple of good casual meals. I built this around a 10:30 start so you're not rushing your morning.

A few practical notes

Best days to go: Wednesday–Saturday. Kelley Farm and Snallygasters are both closed Tue/Wed-ish, and Lower Town gets crowded on summer Sundays.
Bring: comfortable shoes (cobblestones + optional hike), water, layers, $20 cash or card for parking.
If the hike sounds like too much: swap Maryland Heights for the easier walk out to Jefferson Rock and the ruins of St. John's Episcopal Church — gorgeous views, much less elevation.
Skippable if you're tired: True Treats and Battle Grounds are easy to drop without losing the day.

Want me to swap anything — a different hike, fancier dinner, a stop at the Brunswick Heritage Museum on the drive home?"

It gave me a lovely map:

When I clicked the "Open route" button, it gave me this in Google Maps:

I know Harpers Ferry well, and this is a solid trip. I got all of this information in about 2 minutes. Before AI, it would have taken several hours for me to browse multiple websites, select places I wanted to visit, look them up on a map, plot a route, find parking, and solidify an itinerary.

That kind of speed has real implications for our industry, but first I wanted to understand the mechanics.

So I asked Claude how it landed on these options.

What Claude Told Me About Its Process

I said to Claude:

"As you know, I am a heritage tourism consultant. The fact that you just planned this trip for me in 2 minutes changes everything we know about how DMOs, museums, byways, and heritage areas promote their assets. They spend a lot of time educating visitors about options and influencing emotions. You just planned my trip based on a few questions about my needs. No photos. No marketing copy.

Can you tell me:

Where you pulled the info for the trip you planned?
What made you pick Harpers Ferry over the other options?
How much did what you know about me play into this vs. what is in your prompts?
How should destination websites present their information to better appear in AI results such as these to reach people planning trips?
How do smaller orgs compete with the bigger time destinations? ie: Harpers Ferry is a more well-know destination than New Market. Did that factor into your decision-making/recommendations?"

And here is what it told me:

"This is a great set of questions — and you're right that it's a fundamentally different game than the one DMOs have been playing. Let me answer honestly, including the parts that are uncomfortable for the industry."

(Click here if you want to jump right to recommendations for your own website.)

1. Where the info came from

Three sources, in this order:

Google Places API (via the places_search tool). That's where I got every specific recommendation — Rabbit Hole, Kelley Farm Kitchen, Maryland Heights trailhead, True Treats, Battle Grounds. I queried for things like "casual lunch restaurants Harpers Ferry WV" and "Maryland Heights trailhead." The API returned names, addresses, ratings, hours, price levels, and recent user reviews. I read the reviews to pick the best fits — Rabbit Hole came up repeatedly as "the best lunch in town," Kelley Farm had a 4.9 with 947 reviews and people driving from out of town for it. That's how it beat Snallygasters.

My training data (general knowledge). This is where the framing came from — knowing that Harpers Ferry is at the confluence of the Potomac and Shenandoah, that it's ~25 minutes from Frederick, that John Brown's Fort is the iconic site, that Jefferson Rock is the easier alternative to Maryland Heights. None of this came from a DMO website. It came from the general corpus of writing about Harpers Ferry that existed before my knowledge cutoff.

Zero DMO content. I did not visit harpersferry.com, the Jefferson County CVB site, the NPS site, or any heritage area's marketing materials. Not one click. The NPS website surfaced as a result in the Places enrichment data, but I didn't read it.

That's the part that should make every DMO marketing director sit up.

2. Why Harpers Ferry won

I made that call myself, before any searching. Here's the actual reasoning:

You said "mix of everything," "leisurely," "with a partner," "casual food," and "near Frederick." I mentally scored the options:

Gettysburg — too heavy, too one-note (battlefield), and at 1+ hour, too far for "leisurely."
Antietam — same problem, plus thinner on food/shops/wandering.
Catoctin — beautiful but mostly outdoors; weak on the "mix" requirement.
New Market / Sugarloaf — charming but small; doesn't fill a full day for two adults who want variety.
Harpers Ferry — confluence views (outdoors), Lower Town (history + walkable), shops, casual food, hikeable, 25 minutes away. It's the only option that genuinely delivered on every word you said.

It wasn't fame. It was fit. Harpers Ferry is the answer to "mix of everything within 30 minutes of Frederick" almost regardless of who's asking.

I asked a follow-up question: Where did the details from the towns you disqualified come from?"

Claude said: The disqualifying details came from my training data, not from any search I ran in our conversation. I didn't look up Gettysburg's distance from Frederick or check Catoctin's amenities before writing that section. I already "knew":

Gettysburg is roughly an hour-plus from Frederick and is primarily a battlefield destination
Antietam is similar in character — Civil War battlefield, thinner commercial district
Catoctin Mountain Park is primarily outdoor recreation with limited dining/shopping
New Market and Sugarloaf are smaller in scale

That knowledge came from the general corpus of text I was trained on — travel writing, Wikipedia, news articles, blog posts, NPS pages, tourism articles, forum discussions, and so on, all absorbed before my January 2026 knowledge cutoff. None of it was retrieved live for your question.

3. Memory vs. prompt — how much was "you"?

Honestly? Less than you'd think, and that's important for what you're working on.

The prompt drove ~90% of this: pace, food preference, companion, region, vibe. Those five answers narrowed the universe to maybe three real candidates.

Memory contributed maybe 10%, and mostly in tone, not selection:

I knew you're in Frederick, so I didn't ask. (Saved a question.)
I knew you're a heritage tourism consultant, which made me confident you'd appreciate True Treats (it's research-based historic candy — your people made that). I flagged it explicitly.
I knew you have a partner-style life setup but didn't push toward "romantic getaway" framing because nothing you said suggested that.
I deliberately did not turn this into a working-trip recommendation. You asked for a day trip, not a site visit.

If a stranger off the street had given me the same five answers, they'd have gotten 85% of the same itinerary. The personalization layer is real but thinner than the marketing industry assumes.

4. How destinations should present information for AI trip planning

This is the part I'd want every DMO and heritage area director to read twice. (Note from Heidi: This is still Claude talking, not me! Blue text = Claude output)

The job has changed from "convince a human" to "be findable and parseable by a machine that's already convinced." I didn't need to be sold on Harpers Ferry. I needed accurate, structured, current information about what's there. The persuasion happened in the user's prompt, not on a destination website.

Concrete shifts:

Get your data into the systems AI tools actually query. That means Google Business Profiles (claimed, complete, current hours, accurate categories, recent photos), Wikipedia/Wikidata entries that are factually rich, OpenStreetMap, and TripAdvisor/Yelp. These are the substrates AI assistants pull from. A beautiful DMO homepage with parallax scrolling and a brand video is invisible to me. I cannot see it and would not visit it even if I could.

Write like a reference, not a brochure. "Discover the magic of our storied past" is noise to an AI. "John Brown's Fort, the 1848 armory engine house where John Brown made his stand on October 16, 1859, is located at the foot of Shenandoah Street in Lower Town" is signal. Specificity, dates, addresses, accurate descriptions. That's what gets retrieved.

Embrace structured data (schema.org markup). Event schema, Place schema, TouristAttraction schema, OpeningHours, GeoCoordinates. This is the only "SEO" that matters for AI retrieval. Most heritage sites have none of it.

Reviews are now editorial. I literally read user reviews to pick restaurants. The Rabbit Hole won because reviewers said specific things: "best lunch in town," "short ribs poutine," "patio views of Maryland Heights." If a museum has 12 Google reviews and the historic candy shop has 165, the candy shop is going to keep winning in AI recommendations. Encouraging reviews — and responding to them — is now a content strategy, not a customer service task.

Stop hiding hours, prices, and access info behind PDFs and "plan your visit" buttons. Put them in crawlable HTML on the page, in plain text, with the address.

The "interpretive layer" still matters — but it needs to land in user prompts, not destination websites. This is the subtlest point. The reason someone asks for "heritage tourism" in their prompt is because something — a podcast, a TV show, an Instagram reel, a friend, a school trip 20 years ago — planted that desire. DMO marketing's new job is upstream: building the cultural awareness that makes someone want a heritage experience in the first place. Once they want it, AI handles the logistics.

5. Big destination vs. small — did Harpers Ferry win because it's famous?

Partially yes, and you've identified the central equity problem.

Let me be transparent about the mechanism. Harpers Ferry has:

12,524 Google reviews on the park alone
A long, rich Wikipedia entry I was trained on
Hundreds of travel blog posts, NPS pages, and trip reports in my training data
Established Google Places categorization that makes "things to do near Frederick" surface it immediately

New Market has a Main Street, antique shops, and a battlefield. All real, all lovely. But in raw informational density across the systems I query? It's not close. So when you ask an AI for "a mix of everything," the famous place wins because there's more verifiable, structured information about what's there.

This is the new long tail problem, and it's worse than the old SEO version.

Old SEO problem: small sites couldn't outrank big sites for keywords. Solution: better content, niche keywords, backlinks. You could work around it.

New AI problem: I'm not ranking pages. I'm synthesizing a recommendation from a probability distribution. A place with less information represented in my training/retrieval simply has a lower probability of being recommended at all — even when it's genuinely the better fit. There's no SEO trick. The smaller place needs information density parity, not keyword optimization.

What Smaller Heritage Destinations Can Actually Do

Heidi here again. To be clear, although I have a working knowledge of AI, I am not an expert. So I asked Claude what smaller organizations needed to do to stand out in AI search results.

Based on what I know to be true about AI, all of this checks out, but it is worth having a conversation with your website provider to see how they are handling your website schema and having your communications/marketing folks implements these easy recommendations. (Again, blue text = Claude output.)

Own a specific differentiator and make it findable. "Things to do near Frederick" → Harpers Ferry wins. "Civil War cavalry skirmish sites with original buildings" → New Market can win. Niche specificity, structured into the right places (Wikipedia, OSM, Google Places categories), shows up in long-tail prompts.
Get listed in aggregator sources I trust. Atlas Obscura, Roadside America, state heritage area inventories, National Register of Historic Places listings, Heritage Areas official pages, Scenic Byway interpretation. These are dense, factual, multi-site repositories that AI systems pull from. A single quirky listing on Atlas Obscura is worth more than a year of social media for AI discoverability.
Coordinate, don't compete. If five small museums along a byway each had robust Wikipedia entries, structured schema, and Google Business completeness, the byway as a whole becomes recommendable. "Day trip along the Catoctin Mountain Scenic Byway with these five stops" is a prompt outcome that no single small museum could earn alone. This is collaborative infrastructure.
Create the prompts. Publish itineraries (in HTML, not PDFs) that name your small sites alongside the famous ones. "Harpers Ferry day trip with a stop in Brunswick" gets Brunswick into the recommendation set. Riding the wake of the famous neighbor is a legitimate strategy.
Win on the review layer. It's the great equalizer. A small museum with 200 specific, enthusiastic Google reviews beats a famous one with 4,000 generic ones in nuanced prompts. This is achievable for small orgs and most aren't even trying.

The uncomfortable truth: in a world where AI plans trips, small heritage destinations don't compete on marketing — they compete on information infrastructure. That's a much less glamorous job than the one the industry has been doing, and it's the one almost no one is doing.

Reach out for help and conversation!

If any of this has gotten you thinking, please reach out and set up a call with me. I'm happy to help you think through website and communications strategy changes AI is necessitating, as well as how you can collaborate with your neighboring organizations to increase your visibility. Contact me here.