The new first reader: what ‘good reporting’ looks like in the age of AI

When you spend so much time and resource boiling down a year’s worth of activity into one (compliant) report, you want it to deliver maximum value – to positively influence your valuation and reputation.

 

But while it has always been true that most people who open your report won’t read it cover to cover, now many won’t read it at all - they’ll ask a machine to do it for them.

 

We’ve always advocated for audience-first communications. Some might question whether that craft—from narrative to data visualisation—still matters in an AI-mediated world. The evidence suggests the opposite. People are still reading reports, and doing so with intent. But they are increasingly reading them alongside, or through, machines.

 

Reports now need to work for two audiences: the human reader, and the systems that surface, extract and summarise them. This new kind of reader wields outsize influence over the stakeholders that matter most to you, and “good” now means being legible to everyone – and every bot – that engages.

 

So, the question for reporting teams isn't just, 'who's reading this?'. Now it's 'what will the machine understand before they do?', too. In this piece, we unpack these dual demands.

 

Who is really reading your report...and how?

 

A new paper in Accounting and Business Research gives one of the first serious looks at how people behave around the new breed of CSRD‑aligned reports – and, importantly, it offers lessons for companies outside the scope of the ESRS too, given the rise of IFRS S1 and S2 globally and the growing centrality of sustainability in mainstream reporting. It tracks more than 100,000 visitors and over 400,000 page views across eight large European companies in the three months after publication.

 

A few numbers jump out: Sustainability sections account for 38% of page views but an even higher 41% of total reading time, compared with financial sections accounting for 37% of views and 35% of reading time. Users spend more time per visit on sustainability pages and return to them more persistently across the three‑month window, with climate change and own‑workforce disclosures attracting the most attention within those sections. 

 

Echoing research we did as far back as 2019 the audience for the Annual Report turns out to be much broader – and much more purposeful – than the old “this is just for investors” stereotype. The research suggests Annual and Sustainability Reports are now being used by:

 

• Investors and analysts, who still dominate traffic to financial sections and use them to understand strategy, performance, risk and forward‑looking statements. 
  
  
• Employees, who are a significant chunk of the domestic traffic and use reports to check whether external messaging matches what they experience internally. 
  
  
• Prospective hires and students, who increasingly treat the report as due diligence on leadership, culture and future direction, especially via workforce and “about us” content. 
  
  
• Customers and suppliers, who use disclosures on resilience, climate, supply chain and ethics to assess whether they want to buy from – or partner with – the company. 
  
  
• Regulators, NGOs and journalists, who mine both financial and sustainability sections for hard commitments, gaps and storylines, with particular focus on climate and workforce topics, which the study finds are among the most visited sustainability themes.

 

More intriguingly, the research uncovers how readers are navigating online reporting. For sustainability sections, almost half (42.7%) of visitors skip the main report landing pages and arrive directly on sustainability content, compared with 29.6% on financial sections and 27.7% on general pages. This suggests many stakeholders are not starting at the front at all – they are using internet searches, links, and AI tools to jump straight into specific sustainability issues, often without ever seeing your carefully crafted overview.

 

For report teams, that means the detailed sections – especially in sustainability – need to stand alone, work as entry points, and be properly tagged and structured for AI searches. These are two points worth really reflecting on: first, the annual report has become a genuinely multi‑stakeholder hub, with financials and sustainability acting as twin magnets for very different audiences. Second, more and more of those audiences are now turning up with an algorithm at their side.

 

The new first reader: not a person, but a parser 

Investors and analysts are still primary readers of reports, but what has changed is how they read. As we learnt in our recent research into what investors want from climate transition plans, many now use proprietary AI tools and search workflows to scan dozens of reports for salient datapoints and language shifts, pull CSRD/ESG disclosures into comparative dashboards, and let an internal model draft a first summary before they dive in.

 

Employees, journalists and NGOs are doing their own version of this with public AI assistants and internal search. In other words, a growing share of reading is mediated: the first thing to encounter your report is often not a human, but a crawler, a search index, a PDF parser or a large language model.

 

What this means: AI is the new control+f  

Whereas before audiences would skip to the content that matter most and consume the carefully crafted narrative your legal team signed off, now audiences read AI’s interpretation and summation of your content. And that comes with huge risk. If AI informs first layer of reading for your report, a hard question follows: what happens when LLM searches meet content that was never designed with them in mind?

 

How AI is reading your report (or not). 

This is where disciplines such as Search engine optimisation (SEO), Generative engine optimisation (GEO) and meta-tagging become more than digital hygiene. In reporting, they are what help machines find the right disclosures and messages, read them in context and connect them back to the wider story. Without them, AI might not simply miss content - it can reconstruct the report badly, turning clarity into gaps and inaccuracies.

 

When AI meets weak structure: three failure modes 

In practice, report teams see three recurring failure modes once AI and search become the first layer of reading:

1. Material disclosures become effectively invisible. 

2. Genuine progress gets drowned out by problem language. 

3. Machines assemble a parallel version of your report that diverges from what the board thinks it signed off.

 

The fictional case studies below show each of these in action.

 

Case study 1: HorizonCo and the invisible Scope 3 target 

(Failure mode 1: material disclosures machines cannot see)

 

HorizonCo publishes a carefully constructed annual report. The strategic report is coherent, the financials are clear, and the sustainability statement includes a proper transition plan and a tidy table of Scope 1, 2 and 3 targets. On paper, it is exactly what investors have been asking for.

 

Two weeks later, Investor Relations receives an email from a major shareholder:

 

“Our internal AI review shows no Scope 3 target disclosed. Can you confirm?”

 

Not a fun sentence to read on a Tuesday. After a frantic trawl through the PDF, the IR team confirms the target is very much there. So why did the investor’s proprietary model decide it was not?

 

Because HorizonCo has just hit failure mode one: the disclosure exists, but the “structure” and wording makes it invisible to machines.

• The target table is embedded as an image with weak or missing tags.
• The section heading is “Our carbon ambition” – charming for humans, meaningless for search.
• The metadata and XBRL layer never explicitly flag “Scope 3 emissions target”. 
  
  

To a human with patience, the target is findable. To an AI scanning for “Scope 3 target”, there is nothing there. GEO and tagging have turned a material disclosure into a ghost.

 

This is exactly why SEO, GEO and meta‑tagging are now reporting issues, not just digital ones: when structure fails, the machine’s answer to a simple investor question (“Do they have a Scope 3 target?”) is confidently wrong.

 

Case study 2: Northbridge and the progress that sounds like failure 

(Failure mode 2: progress drowned out by problem language) 

Northbridge Group’s situation is different. Here, the disclosure is visible – but badly framed for an AI trying to summarise it.

 

Northbridge has made real progress on its gender pay gap over three years. The annual report shows:

• A downward‑sloping chart with three years of data.
• A supporting table in the notes.
• A paragraph that acknowledges ongoing challenges.

 

Internally, everyone is relieved. Finally, a chart that bends the right way.

 

Months later, an NGO briefing appears on social media: 

 

“Northbridge’s gender pay gap has worsened over the last three years.” 

 

The comms team is baffled. Someone does the obvious 2026 thing and asks an AI assistant for a summary of Northbridge’s gender pay story based on the report. The answer: 

 

“The company reports persistent gender pay disparities and challenges, without clear evidence of improvement.” 

 

This is failure mode two: the structure and narrative make progress harder to see than the problem. 

• The “good news” is trapped inside a chart that is poorly tagged and only lightly described in the text.
• The surrounding narrative leans heavily on “challenges”, “complexity” and “structural issues”, and barely repeats “improvement” or “reduction”. 
• In the metadata, the section is parked under “People and culture”, not clearly marked as “Gender pay gap results and trends”.

 

To a model skimming for meaning, the loudest signal is “problem”; the improvement is a faint noise in the background. When asked “What’s happening to their gender pay gap?”, it answers honestly – but incompletely. 

 

Again, this is where GEO and careful meta‑tagging intersect with editorial judgement. If you want AI (and, by extension, your stakeholders) to understand that the story is “progress, with caveats”, you have to make that structure and language as obvious as you would for a time‑poor human. 

 

Case study 3: the machine‑only version of your report

(Failure mode 3: the parallel, AI‑assembled report) 

 

Taken together, HorizonCo and Northbridge point to the third failure mode: over time, weak tagging and GEO do not just cause isolated misreadings – they create a parallel version of the report that only machines see.

 

In HorizonCo’s machine‑only report, there is no Scope 3 target. 
In Northbridge’s, there is a gender pay issue with no clear evidence of improvement.

 

Now combine that with the Taylor & Francis findings: 

• Sustainability pages are a major entry point and capture a bigger share of reading time. 
• Financial pages still anchor the capital‑markets audience. 
• Little traffic flows between the two. 

 

If the machine‑only report is skewed, then: 

• Investors may come away with an overstated risk and understated progress.
• Employees may think the company’s external story does not match internal messaging.
• NGOs and journalists may pick up apparent gaps that are really artefacts of structure, not substance.
   

In other words, GEO, SEO and tagging are not bolt‑ons. They are the plumbing that determines whether the report your stakeholders read and the report their machines reconstruct stay meaningfully aligned. 

 

What does this all mean: the birth of a new kind of reporting

 

Seen in combination, the three failure modes and the web‑tracking evidence point to a clear – and fairly monumental – shift in annual reporting: away from long, investor‑only documents and towards multi‑stakeholder, AI‑ready sources of truth that have to work just as well for machines skimming in fragments as they do for humans reading the page.

 

That shift shows up in four main ways that define a broader change: 

 

1. From single audience to multi‑stakeholder hub 

Annual and sustainability reports are no longer written “for investors, with everyone else as a bonus reader.” They are now used by investors, employees, recruits, customers, suppliers, NGOs, journalists and others, each arriving with specific questions and limited time.

 

Reporting priority: purpose‑driven navigation.

Reports will increasingly be structured around the real tasks these groups are trying to complete – “understand strategy,” “check culture,” “assess climate risk,” “find commitments” – rather than around internal ownership charts or legacy layouts. Sections will need to work as standalone entry points that make sense to someone who has not read anything else.

 

2. From analyst legwork to always‑on comparability 

 

Investors in particular spend significant time comparing businesses. AI tools enable quicker and (if done well) more objective comparison by feeding reports into internal models, ratings systems, and generative AI tools. But of course, the risk of hallucinations in the data are real if these reports are not properly “discoverable” by AI.

 

Reporting priority: year-round, dual use communication 

Reports will increasingly be constructed both as narrative documents and as data products – with tagging, taxonomies and controlled language designed from the outset so that AI systems and data pipelines can ingest and reuse content cleanly. The same paragraph has to satisfy a board reader and a model that has never heard of your brand guidelines. 

 

As reports are increasingly read by AI – less fluent in translating nuances in the narrative – the need to meet human readers elsewhere and tell your story is increasing. As such businesses are increasingly looking to leverage report content across investor decks, social posts, and website content.

 

 3. From “publish and hope” to discoverability by design 

Success used to be judged on whether the report was accurate, well written and on time. Distribution meant putting it on the website and telling the market it existed. In a world where so many readers arrive via search, deep links and AI summaries, discoverability becomes part of the craft.

 

Reporting priority: authority‑based optimisation.  

GEO thinking will be baked into reporting, not as “SEO tricks” but as entity clarity, consistent terminology and answerable questions – “What is our climate plan?”, “What are our main risks?”, “What anticipated financial effects do we see by 2030?”. The aim is to help humans and machines land on the same, accurate answer quickly, wherever they start.

 

4. From disclosure volume to explainable linkages 

More pages and more metrics are no longer enough. Stakeholders – and their tools – want to see how strategy, risks, sustainability issues and financial outcomes fit together over time.

 

Reporting priority: decision‑ready explanations.  

Good reporting will focus less on listing every possible data point and more on making the causal links explicit and reusable – how issues are expected to affect the business, what management is doing about them, and how that shows up in performance, resilience and capital allocation. The clearer those links are on the page and in the underlying structure, the less room there is for AI to reconstruct the story incorrectly.

 

A different kind of “good reporting” 
The fundamentals of good communication haven't changed: know your audience. What is increasingly necessary, however, is to understand how those audiences engage with reports.  That is what changes the definition of good reporting. It is no longer just a compliant, well‑written document for investors and other stakeholders, but increasingly a discoverable, tag‑rich, AI‑ready source of truth that serves multiple stakeholders without losing its strategic spine.

 

The uncomfortable questions you need to start asking your reporting teams are: can the right people – and their machines – find what matters, understand it in context and reconstruct the story the board believes it has told? The organisations that start designing for that now, by building dual‑use design, authority‑based optimisation and explainable linkages into the heart of their reporting process, will be the ones whose narrative actually survives contact with how reports are read next.

 

What this means for your reporting – and how RY can help 
For companies, this shift means adding new muscles alongside the old ones: understanding how different stakeholders actually search and navigate; building sensible tagging taxonomies; embedding GEO/SEO into the reporting process; and routinely checking “what does AI say about us?” alongside the familiar disciplines of technical reporting, design and narrative.

 

Radley Yeldar can help you make that shift by:

• Designing reports for humans while structuring them for machines – making key sections work as standalone entry points that models can also parse and recombine. 
• Building sensible tagging, GEO/SEO and internal linking into the reporting process, rather than bolting them on at the end. 
• Treating meta‑data, summaries and authority‑building signals as part of the editorial brief, not post‑sign‑off admin. 
• Stress‑testing “what does AI say about us?” before you go live, checking whether priority topics and messages surface and stay in context when machines reconstruct the story.