How the content on this site is made, which parts are machine-written, and what we do when we get something wrong.
Five streams, five different rules
Being vague about this would be the easiest thing on the site to fake, so here is the exact breakdown.
The workflow pages are never auto-generated. Agents gather candidates and draft the change; a person tests, decides, and approves every revision. Every adoption names the metric that improved, and every workflow change carries a dated changelog record with the reason. If we cannot state why a change was an improvement, it is recorded as evaluated rather than adopted.
Journal posts and the Daily AI Brief are agent-written and published without prior human review. Each day our research agents check a fixed list of sources and publish the Daily AI Brief: a round-up of the developments worth knowing, each linked to its source. When a development would change one of our own workflows, they also write a longer Journal entry on it: what it is, what it needs, which workflow it would change, and a recommended verdict. Those verdicts are the agents’ recommendations, not test results, and nothing in a Journal post changes a workflow page until a person has tested and approved it. Every post says it is machine-written and links its sources; we read them after publication and correct or remove anything wrong under the corrections policy below.
Topic pages are agent-maintained. The pages under AI Topics are the current answer on each subject we track. The research agents update them, without prior human review, when a development changes the picture, and every change is recorded in the dated changelog at the bottom of the page. Each page shows the date it was last updated.
The guide hubs — AI Tools, AI for Your Business, and Local & Private AI — are produced against a keyword plan with AI assistance, then reviewed by a person before publishing. Review means prices, model names, version numbers and any claim about what a tool can do are checked against the vendor’s own documentation on the day of publishing.
News & Updates is machine-written and published without prior human review. An automation watches a fixed list of sources, and when one of them publishes something concrete, it writes a short report on that single source and puts it straight on the site. Each of those posts is labelled as machine-written and links the source it came from. We read them after the fact, not before, and we correct or remove anything wrong under the corrections policy below.
We run the News hub this way deliberately: it is a freshness feed pointing at other people’s work, and nothing in it is a recommendation, a test result, or a claim about our own experience. Anything that reports our own measurement belongs in a reviewed stream; recommendations in the Journal are always labelled as the agents’ call, never as a tested result.
How we review tools
We recommend against things. A roundup where every tool is a winner is an advertisement with extra steps, so if the honest answer is that a category is not worth paying for yet, that is the piece we publish.
Where we have run a tool ourselves, we say so and give the specifics — what we ran it on and what we measured. Where we have not, we say that too, and the review is explicitly a synthesis of documentation and other people’s reported results rather than first-hand testing. We do not present the second kind as the first, and the News automation is instructed never to write “we tested” or “in our experience” at all.
Sources
Journal entries link their source. When the source is our own testing, it is labelled as our own benchmark and the hardware is named. When a workflow idea came from someone else’s post or video, they get the link and the credit. News posts always name and link the single source they report on.
Corrections
Mistakes get fixed in place, with a dated note saying what was wrong. We do not quietly edit a claim and leave the page looking as though it always said the right thing. If a workflow turns out to have been a bad call, it is deprecated in the changelog rather than deleted — the wrong turns are part of the record.
Because the News hub publishes without prior review, it carries the higher risk of an error, and we treat a correction there as urgent rather than routine. Found something wrong? Tell us through the contact form on Work With Us and it gets looked at.
Money
Some links are affiliate links, and display advertising may appear on the guide hubs. Neither buys a recommendation — see the affiliate disclosure for how that works and what it would take for us to drop a program.