The short answer
What to know first.
Travellers increasingly arrive holding a generated Kenya-Tanzania itinerary, and the desk's welcome is warmer than you might expect: the draft states your wishes clearly, and wishes are the hard part. What it gets wrong, it gets wrong in five reliable ways. Region-month mismatch is the classic: Ndutu calving camps proposed for August, river-country crossings promised for February, the model recombining true sentences into a false calendar because the migration moves and training data does not. Invented or stale figures are second: park fees quoted from superseded tariffs or rounded from thin air, where the real numbers are the authorities' published rates our fee page maintains. Impossible transfer arithmetic is third: Serengeti to Amboseli by afternoon, three parks in a day, borders crossed like county lines, drafted by a system that has never sat on the Seronera airstrip at noon. Promised wildlife is fourth, guaranteed crossings and assured sightings in exactly the language every honest page of this journal exists to unteach. And phantom inventory is fifth: camps that closed, moved seasonally, or never existed, offered with confident charm. The fix runs in one order: dates first, then regions matched to those dates, then the scarce beds checked against live availability, then real fees from current tariffs, then transfer arithmetic done with an atlas and a watch. Keep the draft's wishes; replace its mechanics; and read our companion pages for the two-country border version. Every AI draft we receive gets this rewrite in a single pass, verified at booking.
At a glance
The planning brief.
The five reliable errors
Region-month mismatches, invented fees, impossible transfer times, promised crossings and camps that do not exist or do not exist in July.
Fix in order
Dates first, then regions to match them, then the scarce beds, then real fees, then honest transfer arithmetic.
What the draft is good for
Wishes: a generated plan states what you want beautifully and how to get it wrongly. Keep the wishes, replace the mechanics.
Sail verifies
Every AI draft against live availability, current tariffs and the ground's actual clock, in one pass.
01
Why generated itineraries fail the way they do
The failures are systematic, not random. A language model has read every migration article ever published and stood on no airstrip; it knows that crossings happen at the Mara river and that the Mara river is in the north, but not that the herds reach it in July and leave it by November, so it recombines true fragments into confident falsehoods, a calving camp in crossing season, a crossing promised at calving time.
Fees age the same way: tariffs change on the authorities' calendars, training data freezes on its own, and the draft quotes last cycle's numbers with this cycle's confidence. None of this makes the tools useless; it makes them wish-generators rather than schedulers, and the traveller who understands the division uses them well: dream with the machine, then book with a desk whose data is the ground itself.
There is also a quieter failure worth naming: the generated plan optimises for coverage, because the sources it learned from were listicles, and so it inherits their vice, five parks where two would deepen, every name included lest something be missed. The unhurried arithmetic this journal runs against human itineraries applies doubly to generated ones; the machine has read every top-ten list and none of the guest reviews that say we moved too much.
02
The five errors, worked
Region-month first, because it sinks whole trips: check every camp against our region-by-month map, and expect to move the itinerary's geography rather than its dates, since dates are usually the fixed truth. Scarce beds second: the draft's named camps were chosen for fame, which means they sold out on the horizons our booking pages describe; live availability decides what survives. Fees third: replace every figure with the current published tariff, VAT treatment included, and watch quoted totals move in both directions.
Transfers fourth, the quiet killer: real distances at real speeds with real airstrip schedules turn by afternoon into by tomorrow lunch, and the fix usually deletes a park rather than adds a day, which is the unhurried principle arriving by arithmetic. Promises fifth: strike every guarantee, then reread the plan in likelihood language and notice it is still a magnificent trip, which is the whole lesson of this journal compressed into an editing pass.
Dates, then regions
The dates are usually true and the geography usually is not: move the camps to the month, never the month to the camps.
Beds against live availability
Famous names sell out years out; the rewrite swaps fame for position on the horizons the booking pages map.
Fees from the authorities
Every figure replaced from current published tariffs: the fee page exists precisely for this pass.
Transfers with a watch
Real distances, real schedules, borders as real events: the pass that usually deletes a park and saves the trip.
03
What the draft gets right
Credit where due: the generated itinerary is often an excellent wishes document. It names the experiences that caught your imagination, balloon dawns, crater mornings, crossing vigils, a beach ending, in your own emphasis, and it sequences a narrative arc some human itineraries forget. The desk reads it the way an architect reads a client's sketch: wrong in every load-bearing particular, invaluable about what the client wants the building to feel like.
So bring the draft, unembarrassed. The rewrite conversation is faster than a blank-page one: we keep your arc, your pace and your named dreams, and swap the mechanics for ones that clear the ground truth in one pass. The gorilla version of this page proved the format; the savannah version completes it, and the two-country border case has its own page for the drafts ambitious enough to need it.
04
The rewrite, in practice
A worked example from this month's inbox: ten nights, generated, running Amboseli, Serengeti, Ngorongoro, Mara, Zanzibar, with a crossing guaranteed in the first week of March and park fees totalling a figure from two tariffs ago. The rewrite: dates held, March; geography moved south to the calving grass where March actually lives; the border crossing deleted with one country chosen per our ten-nights page; fees restated from current TANAPA and NCAA tariffs; Zanzibar kept, with the split-model arithmetic from the coast pages; and the guarantee replaced with the season's honest likelihood. The trip that emerged was shorter on paper, longer in mornings, and bookable.
That is the pattern: the rewrite almost always subtracts geography and adds time, because the generator's signature error is abundance. From the Sail operations desk, checked August 2026: AI drafts now open a meaningful share of our first conversations, the rewrite pass has become one of the desk's fastest routes to a right-sized file, and not one draft yet has survived contact with a calendar unedited.
Questions travellers ask
Useful answers, before you decide.
Should I bother making an AI itinerary before contacting an operator?
As a wishes document, genuinely useful: it states what caught your imagination in your own emphasis. Just expect the mechanics, dates-to-regions, fees, transfers, availability, to be rewritten from the ground.
What is the most common AI safari itinerary mistake?
Region-month mismatch: calving camps in August, promised crossings in February. The migration moves on rain; training data does not. Our region-by-month page is the correction.
Are the park fees in generated itineraries accurate?
Routinely stale or invented: authorities revise tariffs on their own calendars. Every figure gets replaced from current published rates, and our maintained fee page is the reference.
Can an AI itinerary be booked as written?
None we have received, yet: availability, transfer arithmetic and seasonal camp calendars intervene every time. The wishes survive; the mechanics never do, and the rewrite pass is quick.