I watched a person try to get through a Work and Income process with no English and Google Translate open on their phone, and it was obvious that the whole apparatus had been built for somebody else. The apparatus starts early. When someone arrives in New Zealand as a refugee they are handed a stack of pamphlets: English pamphlets, about services they do not know exist, inside a system they have no way to understand, written on the assumption that somebody will be standing alongside to explain them. Usually nobody is. To a person who cannot read the pamphlet, the pamphlet is not information. It is paper.
For a long time I filed all of this as a language problem, because that is the obvious reading and because it is nearly right. The material is in English, or in officialese, the newcomer cannot read it, so you translate it and the gap closes. Language is a real barrier; the scene at Work and Income was one operating in plain view.
The translation had already happened
Then there is the scene that breaks the theory. A woman who has been in Auckland for three weeks needs to know whether the clinic down the road will see her daughter for free. The answer is not hard to find. It sits on an official page, written plainly, and the translate button on her phone renders it into her own language in about a second, accurately enough. She reads it. Then she puts the phone down, messages a man from her home town who arrived a year before her, asks him the same question, and does whatever he tells her, even though the page had already told her the truth.
If language were the barrier, this could not happen. The translation was correct, and it was not only her phone doing the work: the sector has been translating for years, and the New Zealand Red Cross publishes settlement information in Arabic, Dari, Burmese, and Farsi, the exact languages people arrive speaking. (New Zealand Red Cross 2026) The fact reached her in her own tongue, from an official source, and she went to the man anyway. The correct answer, correctly translated, lost.
So the barrier I had watched at Work and Income was real, but it was not the one doing the deciding. What was missing was not translation. It was credibility, and a credibility gap does not close by adding more correct information.
Truth and truthfulness
Onora O’Neill used her 2002 Reith Lectures to take apart the reflex I had been running on, the belief that trust is rebuilt by handing over more information. More disclosure, she argued, often lowers trust rather than raising it, because nobody can weigh the flood. What earns trust is demonstrable trustworthiness: reasons a real person can assess. (Onora O’Neill 2002) A machine translation gives the newcomer more correct information and nothing to assess.
The distinction underneath belongs to Bernard Williams. Truth is a property of the sentence. Truthfulness is a virtue of the person offering it, made of accuracy and sincerity, and we judge the teller before we accept what is told. (Bernard Williams 2002) Richard Reeves, tracing the present mess back to Williams two decades later, compresses it to a line: truth is empirical, but truthfulness is ethical. (Richard V. Reeves 2021) The page was true. The countryman was truthful. Those are different properties, and only one of them was ever going to move her.
A person beats an institution
This is not a quirk of the newly arrived. It is the direction the whole information economy has been drifting, and one number catches the scale of it. Writing in early 2026, Katie Harbath, who ran public policy at Facebook, argued that what keeps being called the decline of traditional media is really a migration of trust from institutions to individuals: on Pew Research figures she cites, about half of Americans asked to name a news influencer named a specific person, against 7 percent who named a media outlet. Her line is that this is not old media against new, it is personalities against brands. The official settlement page is a brand. The man from her home town is a personality: knowable, locatable, answerable. Between the two, the brand loses even when the brand is correct, and a newcomer, with more at stake than most of us and less ground under her, feels that pull harder, not less.
Why she is not being foolish
It would be easy to file the whole scene under bad judgement, the newcomer trusting rumour over the authorities, and that is the reading to resist hardest. During the West African Ebola outbreak, the philosopher Katherine Furman argued that trusting a neighbour over an expert can be the rational move rather than the ignorant one: when an expert’s incentives and values are opaque to you, a person from your own community is far more likely to share your situation and to give advice fitted to it. (Katherine Furman 2020) The countryman has skin in the same ground she is standing on. He has been wrong about things and corrected them out loud, in front of her, which is a kind of evidence a paragraph cannot offer. She is not misreading her situation. She is doing the sensible thing under uncertainty, which is to trust the source she can assess.
The help that exists is a person
The people who do settlement for a living worked this out long before I did, and honesty requires saying so, because my first version of this argument treated the landscape as empty. It is not. New Zealand resettles 1,500 quota refugees a year, with roughly 600 more arriving through family support, (Immigration New Zealand 2026) and the Red Cross walks beside quota arrivals for at least their first 12 months: an assigned navigator, about 130 staff and 1,000 volunteers behind them, translated materials in the languages above. (New Zealand Red Cross 2026) Look at the shape of that answer. When the sector wanted newcomers to believe what it knew, it did not print a better pamphlet. It assigned each of them a person.
But the person is rationed. The navigation is time-boxed to that first year or so, and it is built around the quota, so family-category arrivals and everyone outside the programme get thinner versions of it. The moment the navigator is not in the room, what remains is the stack: material built for the caseworker, not for the person being served, pointed at the professional who administers the system rather than the newcomer standing inside it. Nothing in the landscape is self-serve, in the newcomer’s language, and grounded in a directory of real services, the kind of thing that could say there is an English class at this address on Tuesday morning and be right.
Which is why a settlement tool that competes on accuracy is racing on a track she is not running. The countryman may be out of date, or plainly wrong, and still be the better bet, because he can be questioned, watched, and held to what he said, and a correct sentence with nobody behind it can be none of those things. Anything that wants to be believed, and not merely to be right, has to offer her what he offers: something she can interrogate, something that says the class is at this address on Tuesday morning and is right, and is right again the week after, and admits it plainly when it is not. Trust is not a property a correct page can assert. It is a record, and the countryman’s whole advantage is that she has watched him build one.