Should you change your annual report for AI?
Most of the advice you'll get is wrong. Here's why.
You may have already been told your annual report needs to change for AI.
- Add question-style section headings
- Use shorter words
- Restructure around what someone might ask ChatGPT
- Optimise for 'generative engines'.
Some of it is well-intentioned. Most of it doesn't survive contact with how large language models actually work.
We're in the awkward early years of an industry-wide guess. Annual reporting is barely on the AI advisory radar, and the advice that does exist is often borrowed wholesale from a different discipline.
The recycled-SEO problem
A lot of current AI advice for corporate communications is search engine optimisation with the labels changed:
It's largely the same 2015 playbook in a fresh wrapper. The underlying assumption is that LLMs are search engines with a chatbot bolted on. Strictly speaking, they're not — but there are two things going on, and it's worth pulling them apart.
Tools like ChatGPT, Perplexity and Google's AI Overviews have a search step and a model step.
The two steps behind an AI answer
The search step fetches pages, and it still rewards many of the SEO techniques we've used for over a decade. That's the part GEO is really talking about.
The model step is the new bit. A large language model is trained on enormous bodies of text and learns patterns from how that text is written. You can't game it with keywords. Here, clarity is key and genre matters more than it ever has.
Why genre matters more than format tricks
When you hand an LLM an annual report, the model recognises it as an annual report. It draws on the patterns it learned from thousands of others: formal register, hedged forward-looking statements, audited financial language, third-person past-tense reporting, conventional section ordering.
That recognition is doing a lot of work. It tells the model to treat your numbers carefully, to preserve qualifiers, to summarise responsibly rather than glibly.
A report headed with 'How much money did we make?' and stripped of professional vocabulary reads as something between an explainer, a FAQ and a marketing brochure. The model is less certain of what it's looking at, and its summary becomes less reliable as a result.
Many consultants will tell you that your next annual report should use question-style headings throughout and, for good measure, swap longer words for shorter ones to help AI tools.
Neither change would help an AI tool. Both would change what the document is. The audit committee report stops reading like an audit committee report. The directors' review reads like a Q&A blog post. The investors, regulators and board members who actually have to read the document carefully are worse off, and so is the AI tool the change was supposed to help.
The transparency flip
Here's the part that gets missed: when an LLM can't follow your report, it doesn't admit defeat. It fills in the gaps. A confused model produces a confident-sounding summary that may or may not bear any relationship to what you actually said.
It's a governance issue, not a comms one
For ASX-listed companies and government entities, that isn't a communications inconvenience. It's a governance issue.
Transparency obligations and AI legibility point in the same direction.
A simple test
Defining acronyms on first use, descriptive section titles, leading with the point, explicit causation where legally safe. It was good advice before AI existed.
Question-style headings, dumbed-down vocabulary, keyword stuffing, schema gymnastics. You're probably looking at recycled SEO methodology in a new wrapper.
Why we don't sell 'AI report optimisation'
WorkWords reviews annual reports across eight categories, including AI Readability (whether the file itself parses cleanly when uploaded) and AI Platform Accuracy (how AI tools represent the organisation publicly when asked about it).
We deliberately don't have a category for AI narrative comprehension, or whether an AI tool understands what your report is trying to say.
If we built one, its criteria would be indistinguishable from the criteria for good annual reporting:
- Clear narrative
- Defined terms
- Logical structure
- Honest self-assessment
- Plain language
The work already exists. It just isn't labelled 'for AI'.
The legitimate alternative
If you genuinely want to give AI tools the best chance of representing your organisation accurately, the answer isn't restructuring the report.
It gives the tools a clean, machine-readable version to work from, without touching a word of the report your human readers rely on.
The bottom line
Write a good annual report.
It's the only AI strategy that survives the next model release.
Make your report work for people and machines
The same work that helps a human reader is what helps an AI tool represent you accurately.
Annual Report Review
Independent, scored assessment of your report across eight dimensions — including AI Readability and AI Platform Accuracy — 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