Can AI grow a brand-new local travel site? The Crank experiment, five weeks in

TravelPEC is a guide to Prince Edward County, Ontario, written by someone who lives here. That’s me. I moved to Picton after years of cottage time, and I write it from there.

It covers where to stay, which village to base yourself in, wineries, restaurants, Sandbanks, events, and the practical stuff for visitors from Toronto, Ottawa and Montreal: what’s open in the off-season, and how to get around without a car. It says on the tin that it’s curated favourites, not a comprehensive directory. It isn’t a booking agent and it doesn’t sell anything.

It’s also the subject of an experiment we call Crank.

The question

Can an AI-planned strategy grow a brand-new local site to rank competitively on Google?

The rules:

• Claude plans. I approve each plan before anything runs.
• The budget is about $0.
• White-hat tactics only.
• Research uses Perplexity and SurferSEO.

I approved TravelPEC as the active site on August 27, 2026. Our other research site, aibubblequestion.com, was named the untouched control.

The choices

Prune first. We cut the content back and release it item by item as I review each piece. My belief is that pruning can help a site in search. I’m open to being told otherwise, and this experiment may be what tells me.

Write like a person who lives here. The site should read as if a human in the County wrote it, because one did. That came with an editorial rule: drop the chef and owner names visitors don’t need, and talk about price points and the kind of food or activity instead. The voice should be welcoming, not stiff, and not the same sentence shape three times in a row.

Link every business. Every business mentioned anywhere on the site links to its Google listing or its own website, in a new window. The reader is one tap from current hours and details, from the business itself.

Local rulings stay with a local. The AI can plan, but it doesn’t decide what’s open. Places that have closed come off. Places Google shows as temporarily closed stay listed with a seasonal note. The Glenora Ferry is listed as free.

Archive before rewriting. The current site gets archived before any rewrite goes live, so nothing is lost if a change goes badly.

Fix the plumbing. Back in July, every unknown address on the site returned a normal page titled “undefined — TravelPEC” instead of a “not found” error. That’s the kind of thing that quietly wastes Google’s attention. It now returns a proper 404.

What’s come in

Search Console data for the site starts on August 26. From then to September 29, 2026:

• 3,470 impressions
• 53 clicks
• 1.5% click-through rate
• average position 14.9

That’s 495 different searches and 95 different pages showing up in results, in about five weeks.

The pages that earned clicks:

• What’s open this month: 18 clicks from 225 impressions. The clear leader.
• Prince Edward County without a car: 5 clicks from 263.
• Winery tours without driving: 4 clicks from 92.
• A Thanksgiving itinerary: 4 clicks from 65.
• The Glenora Ferry guide: 3 clicks from 497 impressions.

Compare that with the dining page: 632 impressions, 2 clicks. And the Consecon village page: 173 impressions, 1 click, while the search “consecon village” showed us 85 times without a single click.

The pattern is plain enough. The pages answering a specific, time-bound question (“what’s open”, “how do I do this without a car”) get clicked. The broad overview pages get seen and skipped.

The ferry is a small story of its own. “Glenora ferry” showed us 76 times, “glenora ferry price” 26 times, and “glenora ferry dock” got 2 clicks from 5 showings. People want one fact about that ferry, and they want it fast.

What we don’t know yet

Most of it.

We don’t have a before-and-after. Search Console data starts the day before the experiment did, so there’s no clean baseline to compare against.

We don’t know what pruning did. The pages came back item by item, and we haven’t written up which were cut and which returned.

We don’t know how much is the season. Fall, apple picking and Thanksgiving pages showing up in September and October may say more about the calendar than about the strategy.

And we don’t yet have a clean comparison against the control site. Five weeks is a start, not a result.

What it means for your site

Two things are already worth taking home.

The page that won wasn’t the one about the business. It was the one about the customer’s problem this week. If you run a shop or a restaurant, “what’s on this month” or “parking and getting here” may do more for you than another page about your story.

And AI is useful for planning and drafting, but the facts that matter locally (who’s open, what’s closed, what the ferry costs) should come from someone who checks.

We’ll report again when there’s a result. Meanwhile, the build approach is on how it works, and the site sits with our other builds on the examples page.