可持续发展报告AI手册

AI playbookfor sustainabilityreporting Contents 07Toolkit 03Introduction A “starter pack” of high-impact prompts you can usetoday, plus real-world case studies from Google’s 2025reporting cycle An overview of the context and our intention forthis playbook 04Five-step framework 1 3Best practices A set of synthesized learnings to help you succeed and scale,which include emphasizing the need to keep a human in theloop, iterating, staying curious, and more A straightforward process to identify high-friction tasks,select the right tools, and move from prototypeto production 1 4Conclusion 05Opportunity landscape A map of where AI can add the most value across dataanalytics, content generation, and stakeholder interaction A closing perspective on the path forward with a callto action Introduction At Google, we believe AI has the power to accelerate progress acrossevery sector—including sustainability reporting. Currently, reporting teamsface a perfect storm of highly manual processes, unstructured data silos,and rapidly evolving standards and frameworks. AI can help you navigatethese complexities. We view sustainability as a collaborative endeavor,not a competitive one. That’s why, after a decade oftransparent disclosurein our annual environmentalreports and nearly two years of testing AI in thatprocess, we’re sharing our AI playbook. This toolkit focuses on the tangible: instead ofabstract concepts, we offer concrete examples andprompts to help your team report more efficientlyand effectively. Five-step framework How to integrate AI intoreporting processes How do you actually get started? We’ve broken down our approach into five straightforward steps. This model isdesigned to help you identify high-friction tasks and apply the right AI solutions, ensuring you can iterate quickly andscale what works. Look for your biggest time-sinks—specifically repetitive tasks, workflows involving unstructured data, orprocesses that involve information-heavy, dense documentation. Whether it’s summarizing policy updates orparsing supplier questionnaires, these highly manual processes are the prime candidates for efficiency gainswith AI. Audit manual,time-consumingworkflows Not every problem needs an AI solution; sometimes a spreadsheet formula or simple automation scriptis faster and more reliable. Save AI for complex, ambiguous tasks that rule-based logic can’t handle.Understanding this difference prevents treating AI as a universal hammer for every digital nail, ensuring youuse the right tool for the task. Decide AI,automation,or both Once you know the task’s nature, match the model type to the problem. Generative AI excels at text-heavytasks like summarizing frameworks or drafting narratives, while structured machine learning is often better forquantitative needs, like classifying spend-based emissions or gap-filling energy data. Your specific softwarechoice will likely depend on your organization’s tech stack, but in the Toolkit section below, we share thespecific AI products we use at Google. Select theappropriateAI tool Start small with a prototype rather than aiming for immediate perfection. Test the outputs against human-verified information or data, and then refine your approach based on any discrepancies. Don’t overlook theability of AI to accelerate this iteration loop. You can prompt AI to analyze its own errors and recommendspecific adjustments to improve performance. Build, test,and iteratethe solution A solution is most valuable if it’s reproducible so it can scale. Capture your successful prompts, toolconfigurations, and revised process flows in a central guide. Making your solutions easy to replicate reducesthe learning curve for colleagues and transforms a one-off team win into a scalable organizational asset. Opportunity landscapeWhere AI can add the most value in the reporting process We consulted with a range of sustainability professionals and technology experts to map the emerging landscape of AIapplications for sustainability reporting (Figure 1). The resulting non-exhaustive list highlights where AI can superchargereporting across three key areas: data analytics, content generation, and content interaction. While we haven’t built solutions for every use case, we continue to actively experiment across these opportunities.We hope this landscape sparks ideas for how you might start applying AI to your reporting workflows today. Landscape of potential AI applications across data and content. Figure 1 Content generation Content interaction Data analytics •Data management•Data review•Gap analysis•Peer benchmarking•Supplier analysis •Interactive querying•Content localization•Multimedia generation•User customization •Internal assistance•Narrative drafting•Content visualization•Content standardization•Document summarization•Accessibility enhancement •Mock scoring•Reactive communications•Inquiry response•Consistency review•Claims validation •Content visualization:Propose a