The short answer
What to know first.
A generated Uganda itinerary usually fails on one structural fact: Bwindi Impenetrable Forest sells gorilla permits by sector, four of them, each with its own gate, and the lodge must serve the gate the permit belongs to. AI drafts name a famous lodge and a trek day as separate line items, invent or misremember prices, promise sightings no honest operator promises, and compress road hours that mountain murram does not compress. The rewrite therefore runs in a fixed order: live permit availability first, which fixes the sector; then a lodge shortlist drawn only from that gate; then the days rebuilt around real road and flight times; and finally the deletions, every invented tariff and guaranteed sighting struck out and replaced with what can be stated, which is availability, terms confirmed at booking, and the family's actual wishes carried through. The draft's enthusiasm survives; its operations are replaced.
At a glance
The planning brief.
The test it fails
Generated itineraries name lodges and days; Bwindi sells permits by sector, and the lodge must serve the permit's gate.
The rewrite order
Live permit availability first, sector fixed, then lodges that serve that gate, then the days around them.
What gets deleted
Invented prices, guaranteed sightings and impossible road timings. What survives is the family's actual wishes.
Sail verifies
Every line against live inventory and real road hours, then returns a version that can actually be operated.
01
Why the drafts fail the same way
Generated itineraries are assembled from the public internet, where lodge marketing and permit operations live on different pages. The model learns that people trek gorillas and stay at well-reviewed lodges; it does not learn that the permit is sold for a sector and the lodge must serve that sector's gate, because almost nobody writes that sentence plainly.
So the classic draft pairs a Buhoma address with an unspecified permit, quotes a tariff remembered from an old blog, and promises the family will 'encounter mountain gorillas' as if the forest took reservations. Each error is invisible to a reader and immediate to an operator.
02
The rewrite, in order
First, live availability: which sectors hold permits for the party's size on the possible dates. That answer, not preference, fixes the sector, because a preference for a sold-out gate is a plan for disappointment. Second, the lodge shortlist is drawn from the fixed gate, and only then does taste choose among real options.
Third, the days are rebuilt on real timings: mountain road hours as driven, flight schedules as flown, and the buffer rule that keeps borders and long transfers away from the briefing morning. The draft's shape often survives this surprisingly well; its arithmetic almost never does.
03
The deletions
Prices are struck first: permit tariffs belong to the authorities and journey prices to a live quote, so a proposal carries 'confirmed at booking' where the draft carried a number. Sighting promises go next, replaced by the honest sentence that trackers find the family on the day and the hour is regulated.
Then the quiet impossibilities: the same-day border-and-briefing, the lodge that serves a different gate, the activity below its minimum age. What remains after the deletions is smaller and true, which is the only kind of itinerary worth signing.
04
What to keep from the machine
The drafts are genuinely useful as appetite: they show what the client read, what excited them, and how they imagine the pace. A family whose AI sketch lingered on forest walks and community visits is telling their advisor something no form would have captured.
Sail's working method is to honour that signal: keep the wishes, keep the tone, and return a version where every line can be operated. The client rarely mourns the deleted promises once the real journey is in front of them; what they wanted was never the arithmetic.
Questions travellers ask
Useful answers, before you decide.
Can Sail work directly from a client's ChatGPT itinerary?
Yes, and it is increasingly common. Send the draft as the brief: it carries the client's wishes clearly, and Sail returns a version tested against live permits, sectors, lodges and real timings.
What is the single most common AI itinerary error for Uganda?
A lodge and a trek that do not share a sector. The permit fixes the gate, and a beautiful lodge at the wrong gate means a dawn of mountain roads. It is the first thing Sail checks.
Are AI price estimates ever close enough to use?
No. Permit tariffs are the authorities' to publish and journey prices move with season, occupancy and design. Anything numerical in a generated draft is replaced with confirmed-at-booking figures.