Half your content is AI-drafted. None of your standards account for it.
Your comms team is using ChatGPT to draft press releases. Your marketing team is using Claude to write blog posts. HR is using Copilot to draft policy updates and offer letters. Legal is using AI to summarise contracts. The CEO’s executive assistant is using AI to draft the all-staff email the CEO will send out tomorrow.
This is not hypothetical. It is the state of every Australian organisation with more than two hundred staff right now. And almost none of them have written down what is and is not acceptable about it.
The result, predictably, is brand voice fragmentation. Disclosure inconsistency. Embarrassing errors that slip through because nobody assigned ownership of the “did a human actually read this?” step. Legal exposure when AI-generated content makes claims the organisation cannot back up.
AI-assisted writing standards training closes the gap between “everyone is using AI to write” and “the writing meets the standard your organisation needs.”
This post is for the comms director, P&C director, or marketing lead who has noticed the quality drift and is starting to suspect that nobody owns the problem.
Why your style guide is not enough
Most organisations have a style guide. It covers tone of voice, preferred terminology, the difference between em dashes and en dashes, when to use sentence case. It was probably last updated in 2022.
What it does not cover:
- Whether AI-drafted content needs to be flagged as AI-drafted, and to whom
- The editorial standards a human must apply to AI-drafted content before it ships
- Which categories of content cannot be AI-drafted at all (most organisations should have a list; almost none do)
- How to maintain brand voice when the first draft was written by a model that has never read your style guide
- Disclosure requirements for external audiences: clients, partners, regulators
- The legal and reputational risks of AI-generated claims, statistics, and quotes
These are the questions the style guide does not answer. The training fills the gap.
What a good AI writing standards course should cover
A custom AI writing standards course should do five things.
1. Define what content can be AI-drafted
Some content can be AI-drafted with light human editing. Internal updates. First drafts of marketing copy. Meeting summaries. FAQ entries.
Some content cannot be AI-drafted under any circumstances. Crisis communications. CEO statements about people. Anything that makes a factual claim the organisation will be held to. Anything quoting a named individual. Anything legally binding.
Most organisations have not drawn this line. The training should help them draw it and then teach staff which side of the line each piece of content sits on.
2. Teach the editorial standard for AI drafts
There is a specific editorial skill, call it “post-AI editing”, that most staff have never been taught. It is different from editing human drafts.
Human drafts tend to have errors of voice, structure, and emphasis. AI drafts tend to have errors of fact, specificity, and texture. They sound right. They use plausible-sounding statistics. They reference experts who do not exist. They smooth over the rough edges that would have flagged a problem in a human draft.
The course should teach staff what to look for: fabricated citations, over-confident hedging, the suspicious absence of specifics, the AI tells that signal a draft has not been engaged with by a human brain. The underlying skill, writing for screen, not page, is the same one we teach for any custom content. Applied to AI drafts, it is what catches the mistakes.
3. Handle disclosure
Should AI-drafted content be labelled? To whom? When?
There is no industry standard answer to this yet. Most organisations are figuring it out as they go. The training should walk through the options:
- No disclosure (high risk if the audience finds out)
- Internal-only disclosure (so colleagues know to apply extra scrutiny)
- External disclosure for certain content types (research, thought leadership, anything where authorship is part of the value)
- Full disclosure everywhere (transparent but commercially difficult)
The course should help your staff understand your organisation’s position and apply it consistently. Inconsistency is the worst outcome here.
4. Protect brand voice
Brand voice is the first thing that fragments when AI drafting becomes widespread. ChatGPT’s default voice, pleasant, hedged, slightly American, full of “it’s worth noting” and “ultimately”, starts showing up in everything.
The training should teach staff how to brief an AI tool to match your brand voice and, more importantly, how to recognise when the output has slipped into AI-default voice and needs to be pulled back. This is a practical skill that takes ten minutes to teach and never gets taught.
5. Cover the legal and factual hygiene
AI-generated content makes claims. Sometimes those claims are wrong. Sometimes they are statistically wrong in ways that are hard to spot. Sometimes they reference legislation that does not exist, court cases that did not happen, studies that were never conducted.
The course should teach staff to treat every claim, statistic, quote, and citation in AI-drafted content as unverified by default. Specific rules: every named source gets checked. Every statistic gets traced. Every quote gets verified. Every legal reference gets confirmed.
This is the section that prevents the next “we issued a press release with a hallucinated statistic” incident.
How long should the course be?
Twenty-five to thirty-five minutes. Longer than acceptable use because there are more skills to teach. Shorter than a full editorial training course because most of the audience already knows the basics of writing.
If you have specialist writers in comms, marketing, or content, they may need a longer, more advanced version. Generalist staff need a shorter one focused on the rules and the recognition skills.
What it should cost
AI writing standards training is Tier 2 work: $7,500 plus GST. The course needs scenarios: examples of AI drafts that staff edit, mark up, and assess. Without those, it stays theoretical.
If you have separate audiences (specialist writers vs general staff), the Tier 3 at $10,500 lets us build a two-version course that shares the foundational content and branches at the appropriate point.
When to commission it
After acceptable use. Likely in parallel with critical thinking training. Before your next major external communications cycle.
The longer you wait, the more bad habits set in. Staff who have been editing AI drafts for a year without standards have developed instincts that may or may not be right. The training is harder once those instincts are in place.
What to brief your developer on
When you commission AI writing standards training, bring:
- Your style guide. The actual current version.
- Five to ten real examples of AI-drafted content from your organisation. Both good and bad. Anonymised.
- The categories of content your organisation produces. Internal, external, marketing, comms, HR, legal, etc.
- Your position on disclosure. Even if it is still being worked out, bring the working version.
- The audience split. Generalists only, or generalists plus specialists.
The bigger picture
AI writing standards is the most operationally specific layer of the AI capability stack. AI acceptable use training sets the policy. AI training for managers handles the human side. AI privacy training covers regulatory risk. Critical thinking training builds the verification habit. This one applies all of it to the single most common AI use case in your organisation: writing.
If your staff are using AI to write, and they are, you need this course or something like it. The alternative is brand voice fragmentation, factual errors, and the reputational cost of looking like an organisation that does not pay attention to what goes out the door.
When you are ready, get in touch. Three weeks from kickoff to launch. Fixed price. No drama.