The Three Letters That Changed How an MP's Team Used AI
- Elaine Schillinger
- Jun 16
- 3 min read
Most professionals know AI can hallucinate. Far fewer know how to catch it before it leaves their desk.
In February this year I ran a pilot program with an MP and his team - smart and busy people working under public scrutiny every day. I gave them a lot of useful things that session, but the thing that landed hardest was surprisingly simple.
Three letters: A, B and C.
By the end of the session, those three letters had changed the way the whole room was looking at AI output. And now that the Federal Court of Australia has published its first formal guidance on AI use in legal proceedings, I keep coming back to that moment.
Most professionals already know that AI can hallucinate. It can make things up and present false information with the same confidence it presents accurate information.
The problem is that knowing this in theory and actually catching it before something goes out the door are two very different skills.
AI output looks authoritative. It is well-structured, fluent and confident. A human draft looks like a draft, with rough edges and obvious gaps. AI output looks like a finished product, even when the foundations are shaky. That is where people get into trouble, and it is rarely because they are careless. The tool simply does not make its own gaps visible.
AI output looks like a finished product, even when the foundations are shaky.
So I built a verification step.
I built a verification step into the RAPID prompt framework tailored specifically for the team. RAPID is a structured approach to prompting AI tools responsibly, designed for people working in high-stakes communication environments. After every AI output, before anyone does anything with it, they run through three questions:
What claims require fact checking?
Is anything speculative or assumed?
Does this align with previous public statements?
And then they verify AI output by labelling every claim:
A | DIRECTLY SUPPORTED Present in the source material you provided. Verified. Can stay. |
B | IMPLIED BUT NOT STATED The AI inferred this. Reasonable, but needs a human judgement call. |
C | NOT SUPPORTED AI generated this independently. Verify it, or cut it. |
The room got quiet the first time they did it. Not because the AI had done something dramatic. Because they could suddenly see the difference between what they had given the tool and what the tool had added on its own. For people whose words are scrutinised publicly, that visibility matters a lot.
Then the regulations caught up.
Two months later, the Federal Court of Australia published the GPN-AI Practice Note, signed by Chief Justice Mortimer. It is the Court's first comprehensive statement on AI use in proceedings. It says plainly that presenting false or inaccurate information to the Court is unacceptable. It requires disclosure when AI has been used to summarise or analyse material relied upon by a witness. It draws a clear line between public AI tools and enterprise systems when confidential information is involved.
The Victorian Law Reform Commission tabled its own report in February, the first inquiry by an Australian law reform body into AI use in courts and tribunals. Over a third of Victorian lawyers are already using AI. Fewer than a third have read the existing guidelines.
That gap, widespread use alongside low awareness of what responsible practice actually requires, is exactly the gap I have been working in.
The regulations are catching up. They describe the problem well. What they do not do is give practitioners a working tool for solving it in the moment, under pressure, before something goes out the door with their name on it.
That is what the A/B/C labelling does.
Label what the AI knew because you told it. Label what the AI inferred. Label what the AI invented.
Then decide what stays, what needs verifying, and what gets cut.
Three minutes. Every time. Before anything goes anywhere.
That is what responsible AI use in high-stakes communication actually looks like in practice. A habit, not a policy.
If you work in legal, government, policy, or corporate communications and want to understand how the RAPID framework applies to your context, get in touch.
Comments