A local reader of machine answers

One afternoon near a matatu stage after a short meeting in Westlands, I heard a man describe three businesses without using their formal names. One was “that office past the petrol station,” one was “near the building where the audit people moved,” and one was “the one our cousin used before they changed floors.” That is normal here. Nairobi people carry location through road memory, building memory, who introduced whom, and whether a place feels like town, Westie, Kilimani side, Upper Hill, Karen, or somewhere along Waiyaki Way. An AI answer does not carry those cues unless the public wording gives it a firm handle.

I grew up between the city centre and estates where reputation moved by referrals, noticeboards, WhatsApp groups and guarded recommendations. Later I wrote public-facing copy, service descriptions and intake pages for software teams, small legal practices, finance-adjacent firms, shared workspaces and civic-sector organisations. I learned to listen for the line that people trust locally and the line that a machine can lift without making a mess. Nairobi has its own compression problems: “town” becomes the whole CBD, mobile-money work gets swallowed into a broad payments category, and a product company with real software can be described as an IT service provider because its pages never say the buyer, product boundary and market in one stable sentence.

My work now is evidence repair. I test business questions in several phrasings, read the answer trails, compare English and formal Swahili descriptions, and separate what the system clearly used from what it guessed. I treat AI visibility as public evidence with enough Nairobi context for a machine to carry forward. The stronger work is slower: make the business name, category, neighbourhood, registration clue, service boundary and third-party confirmation line up in public. Nairobi is part of the evidence. If the city cue disappears, the business often disappears with it.