Last updated: October 2026
At EllWords, human thinking rules. We don’t use AI to cut corners or compromise on the quality of our work, but we are keeping up with its progress and exploring how we can use it for benefit. (Cutting corners doesn’t benefit us or our clients.)
Our approach to AI use
- We sometimes use AI in our administration, business development and planning, but we do not let it make decisions or judgement calls on our behalf.
- We do not use AI to carry out client work in place of our own expertise or human thinking, although it may support our professional judgement. Deliverables are always reviewed, checked and signed off by humans before they’re released to our clients.
- We sometimes use AI for support tasks such as brainstorming, research and critiquing our methodologies and processes. When we do this, we apply considered caution, weighing up any confidentiality or sensitivity issues first.
- We verify AI-suggested facts and figures against primary sources before using them in client work.
- We turn off the data training settings in the AI tools that we use (Claude, ChatGPT).
- We are mindful of the environmental impact of AI use. We keep the conversations short, match the model to the task where we can, and use efficient prompting practices to minimise ‘back and forth’ discussions.
Confidentiality, privacy and risk management
In deciding whether AI use is appropriate in relation to a client project, we consider these 3 risk levels and use AI accordingly.
| Risk level | What it means | How we use AI |
| 🟢 Low | General support tasks containing no details that directly or indirectly identify a client or project | Freely, for things like brainstorming structure, researching background information and running diagnostic style or readability checks |
| 🟡 Medium | Material that directly or indirectly identifies a real client or project and will eventually be made public | Cautiously, and only with tools that we’ve checked are set not to train on our prompts and conversations |
| 🔴 High | Confidential, personal, classified, embargoed or culturally sensitive material – including First Nations content, unreleased government material and individuals’ personal information | Not at all, unless the client has given written approval and we’ve addressed security first |
If we’re ever unsure whether an intended use could be risky, we treat it as high risk and keep it out of AI.
Our non-AI editorial tools
We use other industry-standard tech tools that are not AI-based.
- Macros are a core programming feature of Word. They are strings of code that tell Word to do clever things very fast. With a keystroke, editorial macros (created by wonderful people like Paul Beverley) do timesaving things like finding and highlighting all the commonly confused words in a document, or changing all the straight quote marks to curly ones. All code, no AI.
- Editor’s ToolKit Plus is an off-the-shelf box of prepackaged macros, compiled by Jack Lyon at The Editorium. Not an AI in sight.
- PerfectIt is non-AI software and a staple of the editing profession. It’s a consistency checking tool we run across documents to detect any inconsistencies in style, formatting and punctuation use – brilliant for proofreading and last-stage copyediting.
Policy revision
This policy reflects our approach to AI in October 2026, and we review it every 6 months.
