Managing Responsible Use of AI for Advocacy Research at G3ict

One important limitation of artificial intelligence is that it can only work with data that already exists. Yet in accessibility and digital inclusion advocacy research, some of the most important information is often not available through public datasets, government reports, or corporate disclosures. The lived experiences of persons with disabilities frequently remain undocumented, making reliable accessibility data difficult or impossible to obtain through desk research alone. In these situations, field research and direct engagement with disability advocates and persons with disabilities are the only reliable ways to understand real-world barriers, outcomes, and user experiences. For this reason, G3ict remains deeply committed to gathering primary user experience data from persons with disabilities as a core component of its research methodology. The Digital Accessibility Rights Evaluation (DARE) Indexis a clear example of this approach, combining organizational data with direct input from individuals with disabilities to generate insights that would not be available through AI systems or publicly accessible information alone.

How Can AI Best Support Advocacy Research?

Generative AI however, has opened new possibilities for research when examining published data that concerns disability rights and accessibility and technology matters.

It has become a helpful part of the research process, not as a replacement for human expertise, but as a tool that speeds up the parts of the job that used to take the most time. For researchers with long experience in digital accessibility, disability rights, and international policy, AI acts like a productivity booster. It helps organize information quickly, highlight patterns, and prepare early drafts, while human judgment remains responsible for accuracy, interpretation, and final decisions.

One of the biggest advantages of using AI in research is the ability to work faster without lowering quality. AI can scan large amounts of public information, outline documents, and map themes in minutes. This allows a small research team to produce work at a pace that normally requires a much larger staff. It also helps connect ideas across different fields, such as aging, disability, technology accountability, and government policy, making it easier to spot gaps or issues that might otherwise be missed. And because AI can generate a structured starting point, researchers can spend more time on deeper analysis and less time on basic drafting.

These benefits matter, especially when G3ict is producing global toolkits, multi-country reports, or guidance for policymakers. AI helps surface issues affecting older persons and persons with disabilities that are often overlooked in traditional policy reviews. It also frees time for leadership tasks like international advocacy, strategic planning, and framework development.

AI Risks as a Research Tool

But using AI in research also comes with real risks. AI systems sometimes produce information that sounds correct but is not. This is known as hallucination, and it can lead to inaccurate summaries of laws, policies, or country-level data. There are also privacy concerns when using consumer versions of AI tools, because information typed into them may be stored or reused by the provider. And because AI is trained mostly on publicly available data, it can reflect biases, for example, favoring Global North perspectives or outdated views of disability.

G3ict Rules for Responsible AI in Research

To protect the quality and credibility of G3ict’s work, we established and follow a clear set of rules for responsible AI use:

• The first rule is what we call the Sandwich Workflow Protocol. Every research task begins with a human setting the direction and boundaries. AI is then used for background scanning or early drafting. After that, a human reviews every detail, checking facts against official sources before anything is published. This ensures that AI never starts or finishes a research cycle on its own.
• The second rule is strict control over what information goes into AI tools. Unless we are using an enterprise-grade system with strong privacy protections, we treat AI interfaces as public spaces. That means we never upload unpublished reports, sensitive data, or internal communications.
• The third rule is transparency. Any document that uses AI support includes a short statement explaining how AI was used and what parts were completed by human researchers. This models the same transparency G3ict encourages in global AI policy.
• The fourth rule is active questioning. AI tends to give general answers unless pushed. So we use counter-prompts to ask specifically about missing perspectives, for example, how a national AI strategy addresses cognitive disabilities or whether older persons are included in a country’s digital rights framework. This helps correct for bias and strengthens rights-based analysis.
• The final rule is primary-source verification. No AI-generated claim about a country, law, or corporate policy is accepted without checking it directly against official documents such as government registries, UN sources, or corporate transparency portals.

Together, these practices ensure that AI strengthens G3ict’s research rather than weakening it. Used responsibly, AI helps us work faster, think more broadly, and produce insights that support inclusive, rights-based digital policy around the world. But it is always guided, and always verified, by human expertise.