The Invisible Industry

Every day I speak to someone in transport about AI. Every day I hear the brilliant, the genius and the absolutely absurd.

ChatGPT had 1.8 billion UK visits in the first eight months of 2025, up from 368 million in the same eight months of 2024 (Ofcom). The stories I hear from the industry don’t add up, so I went snooping.

500 companies, five questions

I scored 500 transport companies, one to five, on five questions: can a crawler get in, is there an AI briefing file, a stated crawler policy, structured data a machine can parse, and real text rather than a JavaScript shell, in that order.

The average is 3.58. It flatters. Doors are open because the software opens them. But come and peek behind the curtain…

• 462 of 500 sites have no llms.txt, the plain text briefing file for AI models. Whether the models actually read those files is still unproven and I wouldn’t build a strategy on one. But nobody writes one by accident, so it acts as a clean measure to see who is thinking about AI at all. Only 38 of the 500 are.
• 232 of 500 carry no Schema.org structured data at all and 81 have no structured data of any kind.
• 27 companies block AI crawlers by name. 17 of them are selling AI to the rest of us.
• Operators are last on four of five. Their passengers already ask an AI which bus to catch.

Ten bus, ten rail, ten authorities

So I went deeper. Not scraping, real browsing, on ten major bus operators, ten major rail operators and, with franchising underway, ten major authorities’ websites.

To a machine, bus and rail are on two different levels. Every rail site gives a crawler a map, four bus sites give it nothing. Eight rail sites tell a machine what they are, not one bus site does, in its page source.

One major bus company gives an AI no map, no structured data and no description of itself. Another major bus company’s structured data describes its app, not the company.

Sadly, the authorities are closer to bus than rail. Three have structured data, but it’s inherited from a WordPress plugin. Across the authorities that are currently running franchising, deliberate decisions about how a machine reads them… zero!

Franchising makes the authority the answer

Franchising moves the passenger relationship to the authority and its network. When someone asks an AI about their bus, the authority is the answer (or it isn’t).

One authority, which takes its buses into public control from 2027, has no robots file, no sitemap, no structured data and no description of itself.

Crawlable, mapped, structured and briefed

If I were writing a bus franchising spec this autumn, I would add a line to ensure the network’s public information is machine readable, always. Crawlable, mapped, structured and briefed. I would make it a KPI. It really isn’t hard work.

Most of transport is legible to AI by accident. Nobody has been asked to decide how a machine reads their organisation. Somebody should ask.

The way people search is changing.

Transport needs to keep up.

Method: 500 companies scored on five factors by automated fetch of the homepage, robots.txt, llms.txt and sitemap, September 2026, own rubric, non-random sample. The 30 bus, rail and authority sites were checked by hand in a real browser with JavaScript running, homepage only. Individual company results are not published.

For the bid-team version of this, three levers and what I’d check today, see can AI increase your chance of winning a bid.

Benjamin Ross. Practical application of AI in public transport. Founder, economAIcs. Benjamin Ross is the founder of economAIcs, the company behind YourTransportBrain.

Related: AI in UK public transport: what’s actually real in 2026 and How authorities and operators can use AI in bus franchising: six moves.