In corporate communications, the first genuine benefit of an AI assistant is not a finished announcement. It is a starting point. A model can convert a lengthy technical brief into an organised set of claims, highlight stakeholders who have not been consulted, or rebuild a scattered email thread into a timeline that colleagues can actually verify.
That speed is valuable, yet a tidy summary can still fail in the most consequential way: it can present an exploratory conversation as an approved contract, an outdated quotation as a current position, or a tentative roadmap as a firm commitment. The prose may read fluently while the underlying status is untrue.
I use generative models as drafting and triage aids, never as the source of record. Before any sentence leaves my desk, I want clear answers to three questions: What exactly are we asserting? Which source supports it? What is its legal and commercial standing?
Begin with structure, not a paragraph
A communications brief is rarely a clean set of approved facts. It often blends verified detail, market commentary, unresolved decisions and aspirational language. Ask a model to write an announcement straight away and it will tend to flatten those categories into one persuasive paragraph; the uncertainty vanishes from the text even though it remains present in the evidence.
I get a stronger first pass by asking the model to act as a parser, extracting every assertion into a table that records the claim, the supporting document, that document’s author or owner, and whether written clearance exists. This reframes the task from “make it sound exciting” to “show me what we can prove.”
The model can surface missing sources, inconsistent dates and repetitive phrasing. A person still has to judge whether a claim is accurate, approved and fit for public release. A polished draft creates pressure to publish; a visible list of open checks makes it easier to stop.
Audit the verbs of commitment
Verbs such as “launched,” “partnered,” “selected,” “acquired” and “secured” carry real weight in corporate communications, and each describes a different state of affairs. A model that treats them as stylistic synonyms can recast an exploratory discussion as the announcement of a completed relationship.
An AI-assisted draft can draw on genuine historical correspondence and present it as a current partnership. The source material may exist; the implication may not. The remedy is not a better prompt. It is to return to the primary correspondence, verify the relevant agreement and approvals, and delete the unsupported assertion.
That discipline applies to our own communications as well. The HKMA’s published list names Votee AI among the technology partners selected for the first GenA.I. Sandbox++ cohort, and the SFC’s published annex records the same cohort. Likewise, our CTO appears as a contributor to OWASP’s Agentic Skills Top 10 (AST10), a framework that documents security risks in AI agent skills. “Selected” and “contributed” are distinct claims, and each must match the public record exactly.
The governance question for agents
The industry is shifting from scripted chatbots toward systems that can plan and invoke tools across multiple steps. That can make software far more capable, but it also makes permissions, oversight and accountability harder to reason about.
An agent with access to company data should never be treated as safe merely because its output sounds plausible. Systems need constraints on what an agent may access and do, clear records of the actions it takes, and a means for people to review or halt consequential steps. A practical maker-checker pattern helps here: one component gathers information or prepares a draft, while a separate control process tests the action against defined rules and escalates higher-risk decisions to a human.
Review should scale with risk. An internal recap email does not demand the same scrutiny as a public statement, a regulatory filing or an earnings disclosure. Yet every process should state plainly which facts are verified, who approved them, and what remains uncertain.
Spend the saved time on the record
The useful promise of AI in communications is not that it replaces human judgement. It is that it reduces the manual effort of sorting, collating and summarising material drawn from many documents.
The important question is what we do with the time that frees up. I spend it tracing claims back to their sources, asking subject-matter colleagues what a document actually authorises, and confirming that a draft reflects the decision that was made rather than the decision someone hopes will be made.
A strong draft is not the one that sounds most certain. It is the one whose confidence matches the evidence behind it. In communications, the source check is not a final proofreading step; it is part of the work.
Also Read
- U.S. and Russia Set to Resist Binding Rules for Autonomous Weapons at U.N. Meeting
- Washington and Moscow Expected to Block Bid for Binding Rules on Autonomous Weapons
- More than 158,000 Tech Executives Sign Open Letter Demanding Nvidia’s Board Restore Jensen Huang
- Coles Investigates Cybersecurity Incident and Apologises to Customers

