Every other reader email I get at the moment asks some version of the same question: “I asked ChatGPT to plan my trip to [insert Australian region]. Is the itinerary any good?” Sometimes yes. Sometimes hilariously, dangerously no. So I spent the better part of two weeks running four different AI trip planners against a Great Ocean Road itinerary I know intimately, to figure out where the technology actually helps an Aussie traveller and where it falls over.

This isn’t a tech review. It’s a travel-writer’s view of what these tools do well and what they get wrong in the specific context of Australian road tripping. If you’re about to plan a trip with AI help, read this first.

The test setup

I asked four tools — ChatGPT, Claude, Gemini, and a smaller specialist travel tool — to plan a 5-day Great Ocean Road itinerary from Melbourne, mid-winter (so similar conditions to right now), for two adults who like walking and food, moderate budget, with a focus on places that are genuinely worth the stop versus just well-known.

I gave each tool the same prompt, the same constraints, and 20 minutes to refine the answer with follow-up questions. Then I checked the outputs against what I actually know about the route.

What they all got broadly right

The basic structure of the route was correct in every case. Day 1 to Lorne or Apollo Bay, day 2 to the Twelve Apostles area, day 3 through Port Campbell to the Bay of Islands, day 4 either inland to the Otways or further west to Warrnambool, day 5 return. None of the tools suggested wildly impractical drives or got the geography fundamentally wrong.

All four tools correctly noted that winter conditions reduce some of the typical activities — surf lessons, certain coastal walks in rough weather — and suggested seasonal alternatives. None of them suggested swimming at the Twelve Apostles in July, which I’d consider a baseline competence check.

Restaurant and accommodation suggestions were a mixed bag. The well-known places (Chris’s Beacon Point, the Apollo Bay Hotel, Brae if you’re spending serious money) got name-checked accurately. The smaller and newer places were sometimes accurate, sometimes hallucinated, sometimes accurate but with significant detail errors (wrong opening hours, wrong cuisine descriptions, occasionally wrong towns).

Where they fell over

The cracks showed up in the specifics. A few real examples from the tests:

One tool recommended “the historic lighthouse keepers’ restaurant” at a location where there’s no such restaurant and never has been. The lighthouse exists. The restaurant was invented.

Two tools suggested a particular walking trail in the Otways that’s been closed since the 2024 storm damage. The closure has been publicly announced and is easily checked. Neither tool flagged it.

Three tools confidently described the Great Ocean Walk as suitable for “moderately experienced day hikers” — true for some sections, dangerously misleading for others, particularly the Cape Otway to Aire River section in winter conditions.

One tool’s mid-itinerary “hidden gem” suggestion was a town that, when I checked, has a population of around 80 people, a pub that’s currently closed for renovation, and exactly zero of the restaurants the AI claimed were operating there. Pure fabrication.

This is the consistent pattern. AI trip planners are excellent at the structure and at the well-documented options. They go wrong on specifics, particularly anything that requires current information — opening hours, recent closures, weather-dependent access, condition reports.

Where they actually shine

There are a few things AI tools do better than human guidance, and it’s worth being clear about them.

They’re great at handling constraints. If you tell an AI you want walks under 5km, vegetarian-friendly restaurants only, and accommodation under $200/night, it will respect those constraints across the whole itinerary in a way that’s hard to do manually. The output may need fact-checking but the filtering work is genuine value.

They’re great at trip restructuring. “What if I had two more days?” or “What if I want to start from Geelong instead of Melbourne?” — AI tools answer those follow-ups instantly and usually well. That kind of iterative planning is tedious manually and quick with AI.

They’re useful for surfacing options you wouldn’t have searched for. Sometimes the AI suggests an attraction you’d never have heard of that turns out to be perfect. Sometimes the same suggestion turns out to be fictional. The success rate is decent for well-documented regions.

The practical workflow I’d recommend

Based on the testing, here’s what I’d actually suggest for trip planners using AI:

Use the AI for the outline. Get a 5-day skeleton, get the high-level route, get the rough budget estimate. This is where AI is fast and mostly correct.

Cross-check every named business before you commit. Google the restaurant. Check the operator’s actual website. Verify opening hours in their current Google listing, not the AI’s claim. Look for recent reviews to make sure the place is still operating.

Trust the AI less for safety-relevant detail. Walking trails, road conditions, water activities — assume the AI’s information is generic and confirm with current local sources. The Parks Victoria website is the right reference for the Otways. State tourism boards generally publish good current information. Local visitor centres are still gold.

Don’t follow the AI’s recommendations for very small or very new places without verification. Anywhere with fewer than fifty Google reviews, anywhere that opened in the last 18 months, anywhere obscure enough that you can’t easily find independent coverage — these are the places where hallucinations cluster.

A note on the tooling itself

A traveller I was chatting with recently — she runs a small Australian travel agency — was asking the Team400 team about whether they could build her something more reliable for itinerary planning specifically for the Australian market, using verified data sources rather than general training data. Her observation: the current general-purpose AI tools are good for generic trip structures but they don’t have the kind of curated, current data layer that a travel agent would need to confidently sell an itinerary. That’s a real product gap and somebody will fill it eventually.

For now, treat AI trip planners as a useful first draft, not a finished plan. They’ll save you hours of structural work and give you a decent starting point. They will also occasionally invent restaurants out of thin air. Verify, don’t trust.

Bottom line

For my Great Ocean Road test, the best AI-generated itinerary needed about 90 minutes of verification work before it was actually usable. The worst needed closer to four hours and still had problems I caught later. A from-scratch human-planned itinerary at my standards would have taken six to eight hours.

So there’s real value in the AI tools, but only if you treat them like a competent intern who doesn’t know the area and occasionally makes things up. Don’t hand over the steering wheel. Use them to draft, then verify everything that matters.

Happy travels, and as always — pack a beanie. You’re going to need it on the Otways coast in July.