Asked whether machines were making their hiring decisions, most employers said it was unlikely. The regulator found several of them were.

The most useful finding in the Information Commissioner's work on hiring is not about technology. It is about what employers believe they are doing.

Asked about automated decision-making in recruitment, most of the employers who spoke to the regulator considered it unlikely that they were already doing it. The regulator concluded that several of them were.

Nobody appears to have been hiding anything. These were organisations that volunteered to be looked at. That is what makes the finding worth sitting with: sincere belief, sincerely wrong, at organisations engaged enough to put their hand up.

The mechanism is ordinary and it will be familiar to anyone who has run a hiring round. A tool is bought as an assistant. It ranks, it scores, it filters. A recruiter stays in the process, so everyone involved describes it as decision-support, and they are describing the intention accurately.

Then volume arrives. Eight hundred applications for one opening, a hiring manager with a day job, and a ranked list that has already sorted them. The recruiter works down from the top. The people at the bottom are not read, because reading them is precisely the labour the tool was bought to remove. Nothing was decided to change. The tool did what it was bought to do, and somewhere in the middle of that list the process stopped being assisted and started being automatic.

Everyone in the building can still describe it as decision-support, and the description no longer matches what happens to the four hundredth application.

This is where belief turns into a legal fact for someone else. A candidate rejected inside a genuinely assisted process is owed nothing in particular. A candidate rejected by a machine acting alone is owed an explanation, a hearing, and a human review. The employer's own understanding of its process is what decides which of those two people you are, and the regulator has now established that the understanding is frequently mistaken.

The public, meanwhile, has drawn a line of its own. The Commissioner's perception research found people broadly accept that automation can reduce human bias, and comfortable enough with a machine filtering curricula vitae. They were markedly warier of the profiling end, the online assessments that score behaviour rather than experience. That is a reasonable distinction and British hiring practice does not currently oblige anyone to tell candidates which of the two they have just been through.

The behavioural point is small and it is the whole difficulty. Rights that depend on someone recognising a threshold do not fail because people are dishonest. They fail because a threshold crossed gradually, by volume, on a Tuesday afternoon, does not feel like a threshold to the person crossing it.

The candidate feels nothing at all, which is the point. They receive the same email either way.