A general-purpose AI assistant will draft your results announcement in seconds. The draft will read well. It will also invent a figure it cannot source, reuse a phrase you retired last year, and have no idea what your continuous disclosure obligations are. For most writing, fluent-but-unfounded is a minor irritation. For a market-sensitive announcement, it is the whole risk.
Our view: for investor communications, how an AI is governed matters more than how well it writes. Drafting quality is close to a solved problem, and it is not where the risk lives.
The question isn't "can it write"
It is "can you stand behind what it wrote". Those are different questions, and the second one is the one a board asks. A tool that produces a confident draft from nothing is not saving you work; it is moving the work to review, where someone now has to check every sentence against the record because they cannot assume any of them came from it.
So the useful test of an AI for this job is not the demo. It is what the tool can prove about its own output. Three things are worth insisting on.
Three things to insist on
Grounding you can see. The tool should draw only from sources you have approved - your disclosures, policies, prior materials - and show which source each line came from. A model that has merely seen your data is not the same as one grounded in your record, and the difference is whether you can check a claim or only trust it. If a sentence can't be traced to a source, treat it the way you would treat a sentence a junior wrote from memory.
Review that is built in, not bolted on. A person should sign off before anything reaches the market, and the tool's job is to make that review faster and better, not to skip it. Look for disclosure checking that runs before sign-off and explains itself: the flag that cites the rule or the prior statement it relates to, so the reviewer can act in seconds rather than re-read everything.
A record of who did what. Drafts, edits, approvals and sends should be logged against the person who made them. When a regulator, an auditor or a board sub-committee asks how a statement was developed, the answer should be a record, not a recollection.
Notice what these have in common: none of them is about the AI being clever. They are about the AI being answerable. That is the shift a governed tool makes - from a black box that produces text to a system that can show its work.
Why generic tools struggle here
A horizontal assistant is built to be useful for everyone, which means it is grounded in nothing in particular and accountable to no one in particular. It has no view of your disclosure record, no concept of the listing rules you operate under, and no memory of the decisions your company has already made. You can prompt around some of that, every time, from memory - which is the manual work you were trying to remove.
This is not an argument against AI in investor relations. It is an argument for the right kind. The teams getting value are not the ones with the most fluent drafts; they are the ones whose tools are purpose-built for the job, grounded in their own record, and governed so a person stays in control.
That is the standard Diolog is built to, and the security and AI governance behind it are documented in full for exactly the reader who has to sign off. If you'd like to see how it handles your own announcements, book a demo.