05 Three teams, three shortlists
Three teams have been asked to do the same thing: produce a frequently-asked-questions resource about a new policy or service for a public-facing audience. Each will use AI for some of the work. None will use AI for all of it. The teams are illustrative composites, each a recognisable risk profile in Australian comms practice, drawn from interviews and workshop sessions with practitioners in regulators, not-for-profits, and corporate teams. As you read, watch for the team whose risk profile is closest to yours. You will use them as a starting point in the next practice.
Risk profile factors
High public visibility. Possible parliamentary scrutiny. Technical accuracy required. Sector trust expectations are high. Recent policy changes make the topic politically sensitive.
Tools chosen
- Claude Enterprise for drafting and summarising. IT has confirmed the enterprise configuration does not retain inputs for training.
- Microsoft Copilot inside the existing 365 environment, for everyday work that does not touch sensitive drafts.
- Otter.ai with consent-recording defaults, for meeting summaries during stakeholder consultation. External participants get explicit pre-meeting consent and a written notice.
Tools explicitly not chosen
- Public ChatGPT accounts, even for low-risk drafting. Drafting always happens in the enterprise tools.
- AI-generated images. Sector expectations require human-photographed or stock imagery for public communications.
What this team's experience suggests
In high-scrutiny work, the most boring decision is usually the right one. The tool whose configuration paperwork is already done beats the tool whose features are newer. Stability of process is worth more than marginal capability gains when the cost of a mistake is high.
Risk profile factors
Mission-critical accuracy in a sensitive space. Audience often researches privately and on shared devices. A crowded information environment makes credibility a constant pressure. Small team without a dedicated designer.
Tools chosen
- ChatGPT Team with training opt-out on, for drafting, summarising long evidence reviews, and translating clinical guidance into plain language. Sensitive drafts checked by a clinical adviser before publication.
- Canva Magic Studio for the social tiles and infographics the team cannot outsource. AI-generated content is disclosed.
- DeepL Write for plain-English readability editing.
- Claude with retrieval over the organisation's approved evidence base, for answering audience questions where the source can be named.
Tools explicitly not chosen
- AI-generated images of identifiable people, or any image suggesting clinical authority the AI does not actually have.
- Meeting AI for one-to-one calls with audience members.
What this team's experience suggests
They do not use AI for the writing they care about most: the personal stories that anchor their resources. They could, and the output would be technically acceptable, but the credibility of that content depends on it being unmistakably human, and the productivity gain does not outweigh the credibility cost. Being clear about this small exclusion has made their use of AI everywhere else feel less compromised.
Risk profile factors
Lower public visibility but real legal exposure on policy interpretation. Staff data may surface in drafting. The internal newsletter sometimes feeds external communications. Small and time-poor team.
Tools chosen
- Microsoft Copilot in 365, configured to the organisation's data-residency settings, for the bulk of drafting, summarising, and email, because it lives inside the existing security perimeter.
- Read.ai for meeting summaries, but only for internal team meetings, not for sensitive client or HR meetings.
- Adobe Express AI features, for visual assets in the staff newsletter and intranet.
Tools explicitly not chosen
- Personal ChatGPT or Claude accounts for any work, even draft brainstorming.
- AI-generated drafting of the formal policy itself. AI drafts the FAQ that explains the policy, never the policy text.
What this team's experience suggests
When they draft something important, they generate two versions in two different AI tools and compare them, not because they distrust either tool but because the comparison surfaces things neither draft would have caught alone. The AI tools are not replacing thinking. They are scaffolding for it.
What unifies the three teams
The three shortlists are different. That is the point. There is no universally correct AI shortlist for communicators. There is only the shortlist that fits your risk profile, your organisation's existing systems, and your work. What unifies the teams is not what they chose. It is what they did with the choice:
- Each named the tools it would use, and the tools it would not.
- Each wrote a one-line rationale per tool that connected the tool to a specific kind of work.
- Each documented an explicit exclusion: something it decided not to use, and the criteria under which that decision could be reviewed.
- Each set a cadence for revisiting its choices, and named what would make it look again earlier than planned.
What distinguished teams who use AI well from teams who use AI hopefully was not which tools they chose. The leaders in each category had wide acceptance across both groups. The difference was four elements: explicit choices, written rationales, named exclusions, and scheduled review. Teams without them made the same tool choice repeatedly without ever feeling confident about it. Teams with them could defend their practice to colleagues, to leadership, and to themselves, even when using the same tools as the first group.
These are the four elements of a shortlist you can stand behind. They are the four elements the next practices will produce.