Short answer: In August we ran nine searches across five African cities and found that in every single hotel or safari search, the number-one organic result belonged to an aggregator — never to the business itself. We re-ran the same nine searches seven weeks later. This time one did not: a locally-owned operator in Arusha took position one, displacing the directory that held it in August. Across hospitality, businesses’ own sites rose from 23% to 28% of page one. Professional services slipped from 68% to 62%, so the gap between the two sectors narrowed from 45 points to 34. Both datasets are published below.
Seven weeks ago we wrote about a lodge owner outside Arusha typing their own trade into a search box at 11pm and finding everybody except themselves. Review sites, booking platforms, a “top 20 operators” listicle written on another continent.
We said we would re-run the searches rather than let the finding calcify into a talking point. So we did, on 17 October, with the same nine queries in the same five cities.
Most of it came back the same, which is its own kind of answer. One thing did not.
Illustrative composite — not a client, not a testimonial
The crack: one operator took position one
In August, the number-one organic result for best safari operator in Arusha was yourafricansafari.com — a directory. In October, it was safarisolestours.com: Safari Soles Tours, a locally-owned operator in Arusha itself.
That is one search out of five. It is not a revolution and we are not going to dress it up as one. But it is the first time in this study that a local African business has held the top organic spot in a hospitality search, and it is worth more than the five percentage points of aggregate movement that came with it, because it demonstrates the thing the August data could only imply: the wall is not structural. Somebody climbed it.
What that operator actually did
We went and read their site, because “a local business won” is only interesting if you can say how. Here is what is on it, all of it visible to anyone:
- A title tag that says who, what and where — Safari Soles Tours | Tanzania Safaris & Tours — and a meta description of a sensible length underneath it.
- An FAQ block on the homepage answering seven real questions: what is included, how booking works, whether you need insurance, how safe it is.
- A section headed “Your Tanzanian safari in numbers” — summit success rate, over 100 sample itineraries, guaranteed departures, review ratios. Figures, stated plainly, on the page.
Organization,PersonandArticlestructured data in the page source.- Third-party corroboration they point at themselves, including a Trustpilot profile, and a flat statement of identity: “100% locally owned, middle-men free… a trusted licensed company.”
- A real
llms.txtat the root — served astext/plain, with sectioned links to their tours, Kilimanjaro pages and travel information. - A
robots.txtthat namesGPTBot,ChatGPT-User,OAI-SearchBot,PerplexityBot,ClaudeBotandClaude-SearchBotexplicitly.
Now the caveat, and it matters more than the list. None of this proves causation. Google has never said it reads llms.txt, and we have no evidence that file moved a Google ranking. We cannot tell you Safari Soles took position one because of any single item above, and anyone who tells you otherwise is selling something.
What we can tell you is narrower and, we think, more useful: the one business in our sample that beat the directories is also the one that treated machine readers as an audience. Same posture, not a proven mechanism. A correlation with a sample size of one.
It is also the mirror image of what we found the same week in Cameroon. In a separate study of 30 of that country’s largest organisations — published here in French — not one of the 26 with a working website publishes an llms.txt, and not one names a single AI crawler. Those are the two things the Arusha operator did. Across an entire national economy, nobody had done either.
What did not budge
TripAdvisor still owns the hotel searches. It was the number-one organic result in all four city-hotel queries — Nairobi, Lagos, Accra and Johannesburg. The Arusha crack was in safari, not in hotels, and that distinction is probably the whole story: a safari is a bespoke trip people research, a hotel room is inventory people compare.
The sector still predicts more than the effort. Accounting firms’ own sites were 62% of page one; hotels and safari operators managed 28%. A professional services firm remains roughly twice as likely to own its own result as a hospitality business, in the same country, on the same day.
Lagos and Accra hotels are still the worst in the sample. One own-site result in ten, each. In Accra, positions 1, 2, 7, 9 and 10 were all aggregators, and position 8 was an Instagram profile. If you run a hotel in either city, your website is not competing for page one right now; it is competing for position 8 against somebody else’s social account.
All the numbers, both waves
| Metric | August 2026 | October 2026 | Change |
|---|---|---|---|
| Hospitality — own-site share of page one | 23% (11 of 48) | 28% (14 of 50) | +5 pp |
| Professional services — own-site share | 68% (25 of 37) | 62% (25 of 40) | −6 pp |
| All nine searches — own-site share | 42% (36 of 85) | 43% (39 of 90) | +1 pp |
| Hospitality searches where an aggregator held position one | 100% (5 of 5) | 80% (4 of 5) | −20 pp |
| Gap between the two sectors | 45 pp | 34 pp | narrowed 11 pp |
| Searches returning a Google AI element | 100% (9 of 9) | not measurable | see below |
Two waves is two data points. The convergence between the sectors is the interesting line in that table and also the one we are least willing to call a trend — ask us again in wave 3.
What we could not measure this time
The instrument changed between waves, and that costs us a row.
Wave 1 ran on the DataForSEO SERP API, which reports whether Google served an AI Overview. That account has no balance, our sanctioned replacement had insufficient credit, and so wave 2 ran on Firecrawl — which returns organic results reliably but does not return SERP features.
So: the share-of-page-one figures above are comparable across both waves, because they are counted the same way from the same kind of result list. AI Overview presence is not measurable in October. In August it was 9 out of 9. We are not going to carry that number forward as though we re-measured it, and we are not going to delete the row and hope nobody notices. It is dark, and restoring an instrument that captures it is the first job before wave 3.
One more honest wrinkle: wave 1 returned 9 to 10 organic results per query and wave 2 returned 10 uniformly. Percentages compare cleanly. Raw counts do not.
What to take from this
If you run a hospitality business, the useful news is that the top spot in a safari search was winnable by a locally-owned operator with a well-built site, in the same month that four of five such searches still went to a platform. That is not a promise. It is an existence proof, which is more than we had in August.
The things the winner had in place are the same unglamorous list we keep publishing: state your facts on your own site, answer real questions in plain language, put numbers on the page, add structured data, keep one consistent identity, and let the crawlers read you. That is answer engine optimization and GEO work, and none of it requires a platform’s permission.
If you are in professional services, you are still starting from double the odds — and still drifted backwards six points this wave while nobody was watching. Page one is not a possession. For the specifics by sector, our guides on getting a hotel cited by ChatGPT, reducing OTA commission and small business SEO go further than this study does.
Method and limits
We collected wave 2 on 17 October 2026: nine searches of the form best {service} in {city}, {country} — boutique hotels and safari operators, plus accounting firms — in Nairobi, Lagos, Accra, Johannesburg and Arusha, retrieved through the Firecrawl search API with country-level location targeting. Every page-one organic result was hand-classified as the business’s own website, an aggregator, media, or user-generated content, on the same rules as wave 1.
The limits, stated plainly. It is nine searches, not nine hundred. Desktop, English, country-level geolocation. The instrument changed between waves, which is disclosed in the dataset itself and costs us the AI Overview row. Search results move, which is the entire reason this page carries a date and a wave number.
Both datasets are published for anyone to check or reproduce: wave 2 (October 2026) and wave 1 (August 2026). The original write-up is here. Wave 3 runs in roughly seven weeks, same nine queries.
If you quote these figures, please cite them as: AfriShield AI, “Who does AI recommend when someone searches for an African business?”, wave 2, October 2026, with a link to this page.
To see where one business sits against these numbers, run the free AI-visibility check — the same method, applied to a single name.