Photograph: Levi / The Independent

Google stopped filling in your rate in September. East Africa's number now comes from the OTA.

By Onyango George (OG)

Three changes in five weeks moved the decision about what price a traveller is shown away from the hotel and into the retrieval layer. The one country where a guest can now book inside the answer is also Kenya's largest single source market.

On Tuesday 8 September, Nick Fox, Google's senior vice president for knowledge and information, said something a platform executive almost never says in public about his own product. The company had just rebuilt its European search page to satisfy the Digital Markets Act, splitting travel results into one unit stacked with booking sites and a second, narrower unit for the suppliers themselves. Out of that second unit went the date filters, the descriptive tags and, most importantly, the live prices. "These changes degrade the user experience for Europeans," Fox said.

Three weeks later, with no complaint and no blog post, Google removed something considerably more consequential for a camp in the Mara or a beach hotel in Diani. On Wednesday 30 September, under a help-centre notice of a few hundred words, the third-party rates feature for hotel ads became unavailable. Until that date a property that had never built a rate feed in its life could still appear in Google's hotel surfaces with a price attached, because Google would source one on its behalf. After it, a hotel without its own live feed, delivered through a Hotel Center account or an approved connectivity partner, stops serving. "After this date, affected campaigns will stop serving ads across all Hotel Ads inventory," the notice reads. Campaigns cannot be migrated. A property has to pass a pricing accuracy review and build them again.

East African hospitality has spent most of a decade arguing about commission. It is the right argument and it has produced real results, from the parity prohibition now sitting unused in COMESA's 2025 regulations to the eighteen thousand European hotels assembling behind a damages claim in Amsterdam. But commission is a question about what a booking costs once it exists. The three changes of the past five weeks are about something prior and, for the region, more urgent: whether the number a traveller is shown is a number the hotel supplied at all.

The clearest evidence that the answer is now usually no comes from a small, carefully bounded experiment run in June. Nicolas Sitter, a researcher whose AI visibility platform Hotelrank.ai was acquired by Lighthouse in May, froze thirty hotel price prompts, ran each eight times through ChatGPT on 25 June from a United States proxy, and logged every document the model fetched rather than every document it displayed. Across 240 captures the model searched the live web in 98.8 per cent of cases and answered from memory in none of them. It pulled 3,092 documents. Online travel agencies accounted for 1,434 of those fetches, roughly 46 per cent, and were cited back to the user 154 times, a cite rate of 11 per cent, the lowest of any real source class in the study. Booking.com was the single most-cited domain, at 64 citations. Reddit was fetched fifty-one times and cited fifty-one times.

Sitter's summary of this is the sentence East African operators should read twice. "ChatGPT treats online travel agencies as price oracles to read from," he wrote, "not as the sources it shows you." The practical reading is that an OTA listing has become the canonical record of what a property costs, invisibly, whether or not the guest ever sees the OTA's name. Sitter is careful about the limits of his own work: 240 captures, one capture day, one proxy, English only, a snapshot rather than a trend. The direction is still hard to argue with.

Then, on 27 August, the second change. Google began completing hotel bookings inside AI Mode, in the United States and in English, with payment through Google Pay and the supplier or booking platform remaining merchant of record. Ten launch partners went live: Booking.com, Expedia, Hotels.com, Priceline, Trip.com, Choice, Hilton, IHG, Marriott and Wyndham. Five retailers and five global chains. No independent property, no regional platform, no African company of any kind.

Are Morch, who writes on hotel AI adoption, put the gap in four words on 1 September: "Visibility without bookability." His longer version is sharper. "The AI surfaces you, and someone else captures the transaction." For a Kenyan or Tanzanian independent, the chain of consequence is not complicated. The lodge can be recommended inside the answer. It cannot be transacted inside the answer except through one of the ten, which in practice means through an OTA, at the OTA's gross price, with commission already inside the number the guest reads.

This is where the geography stops being abstract. Kenya's Tourism Research Institute recorded 2,394,376 international arrivals in 2024, up 14.6 per cent on the year before. The largest single source market was the United States, at 306,501 arrivals and 12.8 per cent of the total, ahead of Uganda, Tanzania and the United Kingdom. The one country in the world where a traveller can presently complete a hotel booking inside Google's AI answer is the country that sends Kenya more visitors than any other. Europe, by contrast, has just had live prices taken out of its supplier unit, which pushes rate discovery back towards the aggregator block and towards Maps, a click away. Italy and Germany together supplied another 7.5 per cent of Kenya's 2024 arrivals. Two of the region's three most valuable long-haul markets changed their price-display mechanics in the same five weeks, in opposite directions, and neither change was announced to anyone in Nairobi.

The concentration data suggests the independents were already losing before any of this. Research presented by Blake Reiter, Lighthouse's director of hospitality research, at the company's Luminate conference and published in June ran 4,545 ChatGPT prompts across nine destinations and five traveller personas. The model named hotels 49,707 times, but only 2,721 unique properties. Around 10 per cent of Tokyo's hotels and 13 per cent of Paris's were surfaced at all. In Indianapolis, independents took 6.5 per cent of mentions. Eighty-two per cent of cited sources were OTA, metasearch or editorial sites. The study is vendor-authored, by a company that sells AI visibility tools, and should be read with that in mind. Its nine destinations spanned the United States, Europe and Asia Pacific. Not one was in Africa.

Sitter's wider landscape study, published in January from 19,579 AI runs across twenty-five cities, did include two African cities, Cairo and Cape Town. The reported Cairo breakdown is the uncomfortable one: 18 per cent of hotel links went to independents and 42.4 per cent to OTAs. In Tokyo the independent share was 75 per cent, in London and Paris 69 per cent. Whatever is happening in those numbers, and city-level samples are thin, no East African city was tested at all. The region is not losing a visibility contest. It has not been entered in one.

It would be easy to read all of this as another instance of a familiar grievance, and wrong. The striking feature of the 30 September change is how cheap compliance is. A rate feed is a configuration job, not a capital project. Hotel Center is free. The connectivity layer already exists in the region: HotelOnline, founded in Nairobi, works with more than six thousand hotels and hospitality businesses across twenty-seven African countries and is an approved property management system provider in Zanzibar. The expensive fight, over commission and parity, requires a competition regulator, a claims foundation and several years. The decisive one, over whether your own number reaches the machine, requires an afternoon.

Anton Gillis, co-founder and chief executive of HAMAC and formerly chief executive of Kruger Gate Hotel, made the structural version of this point in Business Daily on 20 August, before any of the September changes. Kenyan tourism, he wrote, remains "one of the least digitally mature parts of the economy," a sector whose national statistics can report half a trillion shillings in earnings and yet "still cannot tell you with any confidence how well its hotels are pricing a Tuesday in April." His conclusion was about ambition rather than software. "A country cannot credibly target ten percent of GDP from a sector it has left off its own AI roadmap."

Two qualifications matter, and an operator who ignores them will overcorrect. The first is that the 30 September withdrawal is a paid-media rule. Google's notice concerns Hotel Ads inventory and Performance Max for Travel Goals campaigns, and does not state that free booking links are affected. Where a property's fed rate will surface organically in Europe after the DMA rebuild, Google has not said. The second is segmental. The bush product that carries East Africa's margin is still largely sold through tour operators on contracted rates nine to twelve months forward, which insulates a Serengeti camp from the shopping surface in a way it does not insulate a Mombasa or Nairobi city hotel selling independent travellers at thirty days out. The first-order exposure here is the beach and urban FIT book. The second-order exposure, which arrives later and is harder to see, runs through the operators' own visibility.

Google has published figures suggesting its earlier European changes cut free direct booking clicks for suppliers by as much as 30 per cent, and that a test in Germany, Belgium and Estonia cost hotels more than a tenth of their traffic while booking sites held flat. Those numbers come from Google, in the middle of a regulatory fight, and no one has verified them independently. They are worth noting precisely because they are self-serving: the company's own account of its own compliance is that suppliers lost traffic and intermediaries did not.

Which leaves the question a Kenyan hotelier can actually answer this week. Not what Booking.com charges, which is known, contested and slow to change. The question is narrower and entirely within reach: when a traveller in Denver asks a machine what a room at your property costs in February, where does the number come from, and did you write it.