How to use AI for your next annual report
What to hand over, what to supervise and what's off-limits
The annual report is audited, regulated and signed off by the board. People still believe what it says — rare in a world of unverified content.
Meanwhile, the comms, reporting and governance teams behind it are under pressure to bring AI tools into their everyday workflows.
This guide shows practical examples of how to use — or avoid — AI without putting one of your last high-trust documents at risk. Through the report production process — planning, research, writing, design and proofreading — we look at example tasks and how to decide what AI can safely take on.
Can AI do that now?
Ask your AI tool of choice (or mandate) to write your entire annual report, and it'll do it in minutes. It will make assumptions, maybe fabricate some sources and settle for an answer that's the average of an average of an average. It will sound completely confident, even when it's completely wrong.
The real question is: how can we safely use the AI tools we have now?
Three patterns we're seeing in 2026
AI is already in the reporting workflow, often in ad hoc and informal ways that vary from person to person.
AI-assisted copy is getting more fluent and more readable, but less specific — it reads well but says less.
The tools are improving faster than the guardrails, and some teams don't understand the risks.
How to decide what AI can touch
Run three tests over a task to help decide where it belongs, in the context of your organisation and your report.
Reversibility
How easily can you catch and undo an error before publication?
Verifiability
Can you check the result against a source, standard or style rule?
Consequence
What happens if the error slips through to a reader?
Weigh the three answers together, and a task lands in one of three places:
Security first
Any use of AI needs to be based on your organisation's policies, tools and approach.
Unpublished report content is sensitive and should remain confidential. Use the tools your organisation has approved, and steer clear of 'shadow' tools.
Where each report task lands
These are examples from our own work on annual reports, sorted by the three tests above. Your tasks may differ, but the principles behind the tests stay the same.
Low risk, easy to check
Let AI do the repeatable, verifiable groundwork — then check it.
These are the tasks where a properly trained AI tool earns its keep. Each one is easy to verify, easy to undo and saves your time for the parts that need a human.
The goldmine of the routine and repeatable
This is likely where most teams will first see the impact of AI tools on producing annual reports. And there's very good news for those with a library of templates and examples. Your past work can serve as examples for your AI tool and save you setup time. Think contributor templates, meeting notes and action lists.
Most AI tools let you save a project or a set of standing instructions. Load yours with your brand voice, your audiences, the words you never use and the formats you write in. Examples of what you consider good are gold — they give your tool the context it needs to be useful.
Create from approved content
Content creation is usually the first on the list for comms teams.
Once your report copy is approved and locked, you have a verified source to work from. With the right prompts and guidelines, you can speed up producing a draft media release, social post or board briefing. The condition: AI adapts approved content, never a draft.
Feed it the approved text and your content brief, with specs like word count, audience and themes. Then treat what comes back as a first draft, not a done deal.
Map your source material
Board papers, program reports, interview transcripts, three years of previous reports. Get your AI to read it all and produce a working summary in minutes. The summary is your map back if you need to check a source document.
Get started with accessibility
AI can help draft your alt text, the written description of an image that a screen reader reads aloud.
Working from your approved copy, you can get help to draft:
- an Easy Read version, using short sentences and images
- a short version built for a phone screen
- a plain language summary, following the plain language standard, ISO 24495-1.
Check plain language
The key is ensuring your AI tool has the same understanding of plain language as you do. The tool can help you produce a shorter, clearer version, so your job is a 10-minute review instead of a rewrite.
AI will happily simplify a sentence it has misunderstood, so check the meaning and not just the reading level. Check your reviewed text to ensure:
- figures still match the source
- no new claims have appeared
- nothing essential has been cut
- the tone still sounds like your organisation.
Your AI tool might consider 'plain language' to just mean shorter words and a lower reading score. That's not the standard — plain language is whether your reader can find what they need, understand it and use it (ISO 24495-1). Spell that out so your tool can edit for understanding, not just a grade level.
Sweep for errors and inconsistencies
A machine reads all 120 pages at once and never gets tired on page 90. Pattern-matching is exactly what a machine is for. Targeted sweeps can catch:
- consistency — for example, one risk described three different ways in your draft. AI flags it; you make the judgement call
- tense and voice
- naming of documents, strategies and teams
- whether your report theme carries beyond the front cover, or fades by page 10
- claims that begin but never resolve.
AI sees the whole thing at once, so it catches what you can't.
If a target has slipped off the page, that's a discussion for your board and executive, not for AI to speculate on.
Draft an index
If you're still including an index (which seems to be a dying art), the right prompts and tools can significantly cut indexing time.
AI can take the tedious part — pull out the concepts, names and themes worth indexing.
You can do the thoughtful part — not just where the word appears, but its meaning and context. Treat its work as a first draft, then bring your own judgement.
Their job just got more critical. AI doesn't do nuance well, or at all. The sense check is still a human task.
Faster work, same responsibility
AI speeds the work; you still own every judgement about what's right.
Amber is where most report work lives, and where generative AI tends to earn its keep. A green task can be checked against a source or a style rule. An amber one can only be judged by someone who knows your organisation.
Find the stories
Add a year of your own media releases, newsletters and content, plus the external coverage you earned, and ask AI what the through-lines are. You'll surface the moments worth building the report around and catalogue your source material, rather than starting from a blank page.
The same works for what people told you. Feed a pile of staff survey comments or consultation responses to your AI tool. You'll see the shape of what people said in minutes. But it's a map, not a way to skip reading. It leaves out nuance and groups the loud, common answers, so read the raw responses behind any theme you publish.
Tighten the writing
AI tools can shorten your sentences and strip the jargon — and give you 10 minutes back.
But deciding what to remove is another matter. Knowing what the reader truly needs, and what can safely go, is the skill at the heart of good communication.
So if you draft with AI, make your editing more aggressive, not less. AI executes the request, but you need to supply the meaning behind it.
Draft the data visualisation
Hand AI a dense results table and ask how it might be shown. You'll likely get a bunch of chart types, groupings and ways to surface trends. It's a useful starting point when you're staring at a spreadsheet wondering how to make it make sense.
So take the options, then discuss with your designer. They should only roll their eyes a tiny bit if you explain what you're trying to do.
Prepare and process interviews
Internal interviews are how a lot of report content starts, with the executive or a program lead talking through the year. AI helps at both ends of that conversation, but not in the middle of it.
Use it to draft your questions, and then follow up with a transcript and summary. AI interviewing is gaining some traction. But using it to run an interview without your human oversight is unlikely in most organisations. It can't sit in the meeting and ask a follow-up question because of the look on someone's face — that's human.
The work that stays human
No shortcuts here — every one of these needs a human to decide and sign.
Red is where the stakes are highest. The risk isn't that AI can't do the work — it will do it, in seconds, and it will even look right. At first. The risk is that when it gets it wrong, no one notices.
Check every fact and figure
Ask AI to check your arithmetic and it'll happily oblige. The problem is it can also invent a figure that looks right, reads consistently, and is wrong — all delivered with the same easy confidence:
- A confidently wrong date
- An out-of-date job title
- A paraphrased quote
- A regulation cited by the wrong name
A wrong number in your published report can mean correcting the report, restating the figures and sometimes announcing the error to the market. A wrong name in the directors' section, or a misquoted chair, is the kind of error people notice and remember. The report is a matter of record, and it's your role to get it right.
So every number, name, date, title, quote and citation must be checked against the source, by a human.
Explain why the results happened
Why did the number move and what does it mean? They're complex questions with nuanced answers. Ask AI and you'll get a fluent, confident response in seconds — that's the danger.
Analysts and boards read these passages closely and will spot your smooth explanations that explain nothing. The connection between what happened and what it means is exactly the judgement they're paying management to make.
Avoid the temptation to draft in seconds
AI can give you a clean, competent draft of a chair's letter in seconds. Don't do it. Let it give you a suggested structure or some headings, or you'll spend twice as long editing the 'final'.
These letters are read by people who know your chair: board members, investors and staff who've been there 20 years. AI makes everyone sound the same: fluent, warm, faintly corporate, interchangeable — so write this with the CEO or chair. A reader feels the lack of human voice, even unnamed.
Say what happens next
A forward-looking statement is any claim about what will happen: your outlook, your targets, your market guidance, the climate scenarios you model.
AI can be genuinely useful in helping you describe what has already happened. But it's out of its depth when it comes to drafting what will happen next. There's a firm red line here because it carries legal weight. The Corporations Act sets the obligation, and ASIC spells out what it expects in RG 247, its guidance on the operating and financial review.
Sign it off
The decision to approve a report can't be delegated to an AI just yet. Humans remain accountable for every number, claim and promise in it.
AI can tell you the document is consistent, but not whether it's true; only the second one matters at sign-off.
The bottom line
AI shortens the distance to a first draft, but opens up other cans of worms.
AI doesn't remove the need for professional writing, editing, proofreading and design for an annual report.
Five questions for your next planning meeting
Do we agree on which of our report tasks are green, amber and red?
What are the security limits of our approved AI tools, and who can see what we paste in?
Are we tracking what we try with AI, and its impact on cost and time?
How will we verify every AI-touched fact and figure against its source, and whose job is that?
For each section, who owns the decision to publish, and does everyone know a human — not the tool — is accountable?
Sources
- International Organization for Standardization, ISO 24495-1:2023 Plain language — Part 1: Governing principles and guidelines. iso.org
- ASIC, Regulatory Guide 247: Effective disclosure in an operating and financial review; AASB S2 Climate-related disclosures. asic.gov.au
Bring AI into your reporting — safely
We help teams put the green, amber and red framework to work on real reports.
Annual Report Review
Independent, scored assessment of your report across eight dimensions — including how well it reads for both people and machines — with a prioritised improvement roadmap.
Learn moreAnnual report writing and production
Expert report content from strategy to final proof. We write, edit and project-manage annual reports for ASX-listed companies, government agencies and NFP organisations.
Learn more