TheAICommand is a free, independent publication. It has no named byline: the author and publisher of everything here is the Organization, TheAICommand. These standards set out how the work is done so you can judge it on its method, not on a personality. They extend the seven principles on our About page into an operational commitment.
Editorial methodology
We write in plain English, cite our sources, and lead with the Australian context. Where a primary source exists (a regulator paper, a model card, a legislative instrument, a filing), we link and quote that rather than a news write-up of it. Regulator coverage starts with APRA, ASIC, Comcare, OAIC, the Fair Work Commission, and the model WHS regulators. Overseas developments are covered only when they have a material implication for work done in Australia.
Claims are cited or cut. If a figure or quote cannot be traced to a source we can link, it does not run. We prefer to publish less and be right than to publish more and be approximately correct.
A methodology you cannot inspect is a slogan. The examples below are real ones, drawn from our own review records, showing how each rule is applied in practice. The same publish-the-rules-first discipline applies to the data our assessment tools collect: the sample thresholds and reporting rules for any future benchmark are set out in advance on the benchmark methodology page.
The same discipline runs through the workflows we teach, not just the ones we write to. TheAICommand works to the Verified Draft Method: de-identify the inputs, ground the model in your own source material, keep a person at the decision point, verify against the primary source, and log what happened.
How a claim is verified
Every load-bearing legal, statistical, or product claim is checked against the current authoritative source: the registered compilation of an Act, the published paper, the regulator's own page. Paraphrase that drifts from the source text gets pulled back to it.
- A draft on household services claims under the SRC Act described the section 29(5) exclusion as "the first 28 days after injury". The compilation text is "the 28 days beginning on the date of the injury". The article was corrected to carry the statutory wording, in the body and in the FAQ.
- A draft reported that "more than 80 per cent" of surveyed executives saw no impact from AI on employment or productivity. The paper (NBER Working Paper 34836) puts the figure at around nine in ten. Understating a finding is still misreporting it; the article was corrected to the published figure.
- A draft attributed city-level salary uplift percentages to a recruiter's salary guide that does not publish percentages. The figures were re-derived from the publisher's own city pages (for example, a national benchmark of $170,000 against $185,300 for Sydney) and presented as a worked example instead.
What counts as a source
Where a primary source exists, we cite it rather than a write-up of it. A claim that can only be traced to circulation, not to origin, is treated as unsourced.
- A four-pattern taxonomy of algorithmic collusion was attributed in a draft to an OECD report. Verification showed the framing actually came from a competition regulator's blog post. The article was corrected to carry the OECD report's own published categories, with the blog framing re-attributed to its real author.
- Consumer Data Right action dates sourced from secondary coverage were replaced with the official dates on cdr.gov.au (13 July 2026, 9 November 2026, and 10 May 2027) before publication.
- Two widely circulated statistics about performance reviews ("40 per cent of appraisals", "78 per cent of managers") could not be traced to any primary source. Both were excluded from a guide that would otherwise have leaned on them.
What gets cut
"Cited or cut" is applied literally. When verification fails, the claim does not run in weaker form; it is removed, and anything resting on it is rewritten.
- A draft cited a government guidance document called "Redesign your workflow". No such document exists. The claim was re-attributed to the agency's actual implementation guidance (October 2025), and the two specific claims that guidance does not support were cut.
- A draft claimed a workflow change had saved "thousands of hours". No source existed for the figure. It was cut at verification.
- A widely repeated quote attributed to a technology company's chief executive could not be traced to any primary record of the executive saying it. The quote was cut before the piece published.
Human verification of AI-drafted content
AI assists our research and drafting. It never gets the final word. Every article, learning module, and Brief passes through human editorial review before it is published: a person checks the load-bearing legal, statistical, and product claims against the cited primary sources, confirms the Australian framing is accurate, and signs off. Where an article carries a Last reviewed date, a human has re-checked its time-sensitive claims on that date.
How we use AI
We are an AI-education publication, so we use AI openly in our own workflow. AI tools help us scan primary sources, draft first passes, and produce brand illustrations. AI does not decide what is true, does not select what we cover on its own, and is never the final editorial voice. Every published claim is verified by a human against a source a reader can check. We do not publish AI output unreviewed, and we do not present AI-generated illustrations as photographs of real events.
Corrections policy
When we get something wrong, we correct it in public. We fix the error, note what changed, and leave the original wording visible where practical. There are no silent edits to the substance of a published piece. Credibility lives in the corrections, not the covers.
If you spot an error, tell us at hello@theaicommand.com with a link to the source you believe is correct. We aim to acknowledge substantive corrections quickly and to fix confirmed errors promptly.
Corrections in practice
What that looks like when it happens:
- A published article on tribunal review rights stated a 30 day window to apply for review. The correct statutory window is 60 days from the day the decision is received. A post-publication review caught it, and every instance across the article (the body, the step-by-step guide, and the practitioner callout) was corrected, not just the first one.
- A draft cited a superseded compilation of the Corporations Act. The citation was updated to the current registered compilation before publication, and checking compilation currency against the Federal Register of Legislation is now a standard verification step for every legislative citation.
- A source-link audit across published articles found eleven dead or wrong reference URLs, including two citations that pointed at entirely different Acts from the ones named in the text. All eleven were replaced with the correct primary documents.
Transparency notes
A dated public record of the editorial work: what was published, what was revisited, and what verification changed. Nobody audits a publication that keeps no ledger, so we publish ours. New notes are added to this section over time.
Note 1: 13 August 2026
- Pieces published: 264. Period: launch (26 April 2026) to 13 August 2026. Counts articles, news posts, glossary entries, and editions of the Brief.
- Pieces revisited after publication: 228 (an upper-bound estimate). Most of these edits came from two site-wide passes that added direct answers, FAQs, and internal links; a smaller number were factual corrections and source-link repairs.
- Claims corrected at verification: more than 150 documented (an estimated minimum). Covers review records kept between early July and mid August 2026, including pre-publication verification of scheduled articles.
- Claims or references cut as unverifiable: at least 12 documented (an estimated minimum), over the same records.
How these numbers were derived: the publish and revisit counts come from the content repository itself (publication dates in each piece's metadata, and its edit history after first publication), so they are exact for what they measure but exclude the learning modules and the interactive tools. The correction and cut counts come from written review records; verification passes before July 2026 did not keep claim-level tallies, so the true totals are higher than the documented figures shown. The revisit count includes structural improvement passes, not only corrections, which is why it is labelled an upper bound.
Reusing our charts and figures
You are welcome to reproduce our stat cards, charts, and figures, on two conditions: attribute TheAICommand with a link to the source article, and keep the content unchanged. Every stat card carries its figure, its underlying source, and that source's date on the image itself, so the attribution survives even when the card is screenshotted into a slide deck or report. Articles that carry a stat card include a ready-made embed snippet you can copy, and every article and glossary page carries a cite-this-page block with a formatted citation.
Do not alter a figure, crop away its source line, or present a card as your own work or as ours saying something it does not say. If a figure we published is later corrected, the card is regenerated and the same URL serves the corrected version, so a live embed always shows the current figure.
Cite with confidence: our URLs are permanent. Article, glossary, and image addresses on this site are stable, and if anything ever has to move, the old address will redirect to the new one rather than break. A citation made today is a citation we intend to honour for the life of the site.
Ownership and funding
TheAICommand is an independent publication based in New South Wales, Australia. It is self-funded. It carries no advertising, takes no sponsorships, and accepts no payment to cover, review, or rank any product or organisation. Nothing on the Site is a paid placement. All foundational learning, news, the Brief, and the interactive tools are free, and paid content is never required to learn. If paid content ever arrives, it will be additive rather than a paywall around what is already here.
Contact and feedback
For corrections, feedback, or editorial questions, email hello@theaicommand.com. This is the only contact channel, and it is the fastest way to reach the editorial team. Please do not send confidential, privileged, or sensitive material, and never send real case or client data.
General information and education only. Not legal, compliance, financial, or professional advice. Always confirm obligations against the primary source and current regulator guidance.