How The Bearing Is Made

Most of the words on this site are drafted by AI. We think you deserve to know that, and to know exactly how that happens.

The Bearing exists because good policy analysis takes hours of reading government documents, cross-checking claims, and thinking honestly about trade-offs — and almost nobody has the time to do that on every story that matters. We built a process that does the hard reading properly, and we're publishing it here in full, because a black box asking for your trust isn't worth trusting.

Where a story starts

Every article begins with real source material. We pull from government releases (Treasury, the RBA, the ABS, the Productivity Commission, the PBO - as well as ministerial media statements), established research institutions, and mainstream reporting. We rank government-data and peer-reviewed sources above press releases and advocacy content. An editor can pitch it directly and the system researches it from scratch before writing a word.

The analytical lens

Every candidate story is read the same way: what does this policy actually do, not just what does it claim to do. Who gains, who pays, and is that the group it was sold as helping. What happened the last time something like this was tried. The goal is not to arrive with a conclusion already written — a piece that starts with the answer is not analysis, it is a reflex wearing analysis as a costume. Where a policy is well-designed, we say so. Where the evidence is genuinely mixed, we say that too, instead of resolving it into something tidier than the facts allow.

Writing and verification

A draft is written against that research, grounded in the source material and our own research library, never asserting a claim the evidence doesn't support. Every draft is then read a second time by a separate, more careful editorial pass — one that checks the argument holds together, that nothing significant has been left underweighted, that headings and structure serve the reader, and that a related second story hasn't been buried inside the first. That second pass also writes the FAQ block you'll sometimes see at the bottom of a piece, and generates the header image using a consistent rough-oil-paint-impasto style.

Checking the facts

After a draft is reviewed, yet another AI pass checks its specific claims — dates, figures, named events — against real evidence. If a claim doesn’t match what the writer was given, the system searches for it live. A claim that turns out to be true gets a citation at the bottom of the article. A claim that’s wrong gets corrected — just the fact, not the sentence around it — and logged in an audit trail. The rare claim nobody can confirm or deny gets held for a person to check before it publishes.

This doesn’t make The Bearing infallible. It’s still AI, and AI can still be wrong. But it’s worth saying why this holds up anyway: it looks for evidence about one claim and reports what it finds, whichever way that goes. There’s a lot of political writing that does the opposite — it starts with a conclusion and looks for evidence to fit it. That difference is why we think this bar is as higher than lazy journalism, not despite being AI-checked, but because of it.

What a human actually does

Nothing publishes itself. A human decides which stories get commissioned in the first place, can redirect the angle or headline before writing starts, can rewrite or regenerate the image at any point, and makes the final call to publish. The system does the reading and the first draft. The editorial judgment about what The Bearing says is still made by a person.

What this doesn't mean

It doesn't mean the process is finished improving, or that a piece is beyond question. If you think we've gotten something wrong — a fact, a framing, an omission — we want to hear about it. The whole point of building this in the open is that it should be able to withstand scrutiny, not avoid it.

You can read more about why this exists on the About page.