4 Illustrative Examples of AI-Assisted Work

This chapter presents illustrative examples of how generative AI can participate in research, policy, and investigative workflows. The examples are intentionally generic. Their purpose is not to prescribe workflows, endorse specific tools, or define universal best practices, but to demonstrate patterns of use that align with the conceptual and methodological commitments developed throughout this book.

Across all examples, the emphasis is not on the outputs themselves, but on how those outputs are interpreted, reviewed, revised, and incorporated into analytical work. Generative AI contributes to exploration, drafting, organization, and synthesis, while responsibility for interpretation, evidence evaluation, and final conclusions remains human.

The examples in this chapter therefore reinforce a recurring principle:

The value of AI assistance lies not in producing answers, but in supporting human sense-making, reflection, comparison, and revision.

One useful way to think about generative AI is as a form of analytical scaffolding. Scaffolding supports work temporarily while a structure is being developed, but it is not the structure itself. Similarly, AI-generated outputs can help organize ideas, surface alternatives, summarize patterns, identify assumptions, and improve readability without replacing the underlying analytical work required to validate, interpret, and justify conclusions.

This distinction matters because generative AI systems produce fluent and plausible language regardless of whether the underlying reasoning is complete, accurate, or contextually appropriate. Outputs may therefore appear more authoritative than they actually are.

Throughout the examples below, the central analytical question is therefore not:

“What did the AI generate?”

but rather:

“How should the generated material be evaluated, revised, and used responsibly?”

In the early stages of a project, research questions are often still evolving. Relevant boundaries may be unclear, important assumptions may not yet be visible, and alternative framings may not have been fully explored. In this context, generative AI systems can help broaden the space of possible interpretations or analytical directions.

For example, a researcher may ask the system to propose alternative framings, identify overlooked dimensions, suggest stakeholder perspectives, or surface competing analytical lenses. A prompt such as:

“List several alternative ways this policy issue could be framed for analysis, highlighting different underlying concerns or objectives.”

may generate framings emphasizing equity, implementation feasibility, cost containment, operational efficiency, or risk management.

The researcher may then review these framings, discard irrelevant ones, identify missing considerations, and select one or more approaches based on the project’s context and objectives.

Importantly, generated framings should not be treated as comprehensive or authoritative representations of the issue. Doing so risks premature closure and may exclude perspectives that require domain expertise, stakeholder engagement, or contextual understanding to identify. Responsible use therefore requires deliberate review, including comparison against project mandates, reassessment of missing dimensions, and explicit reflection on why certain framings were adopted or rejected.

Generative AI systems can also assist in structuring broad or ambiguous problems into smaller analytical questions. This can help researchers clarify scope, organize inquiry, and identify relationships between different aspects of a topic. The goal is not to produce a definitive research design, but to support orientation within a complex problem space.

For example, a researcher may ask:

“Break this broad research topic into distinct analytical questions that could be explored without prioritizing them.”

The resulting structure may organize the issue into areas such as implementation challenges, equity implications, financial considerations, operational impacts, or evaluation questions.

Used carefully, this type of output can function as a preliminary analytical map. However, the generated structure should not be accepted uncritically as a complete representation of the research problem. Important dimensions may still be absent, and assumptions embedded in the prompt or training data may shape the resulting categories.

Human review therefore remains necessary to reassess scope, expand missing dimensions, and ensure that the final analytical questions remain human-defined rather than AI-derived.

Generative AI can also support drafting and revising analytical writing, particularly with respect to readability, organization, clarity, structure, and tone. In this role, the system functions primarily as a linguistic and organizational aid rather than as a substantive analytical authority.

For instance, AI-assisted revision may help simplify dense passages, improve transitions between sections, reorganize arguments, or adapt writing for different audiences. A researcher might prompt the system with:

“Rewrite this paragraph to improve clarity and flow while preserving all substantive claims, limitations, and qualifications.”

The revised text may become smoother and more concise, but careful comparison against the original remains essential. Analysts must still verify that uncertainty has not been reduced, caveats have not been softened, causal claims have not been strengthened, and methodological nuance has not been lost.

Without this review, revised text may appear more definitive or more confident than the evidence actually justifies.

Generative AI systems can also help researchers work across large volumes of material such as policy documents, interview transcripts, reports, meeting notes, or literature reviews. In these contexts, AI-assisted summaries may help identify recurring themes, contrasts, or areas of disagreement across sources.

These summaries can function as navigational aids that help researchers determine where closer reading is needed, where evidence diverges, and where further verification may be required.

For example, a researcher may ask:

“Summarize recurring themes and points of disagreement across these documents without assessing which position is more valid.”

The resulting synthesis may identify recurring policy concerns, implementation tensions, conflicting assumptions, or divergent interpretations. Used appropriately, such summaries can help prioritize which documents require deeper review and where supporting or contradictory evidence should be examined more carefully.

However, generated summaries should never substitute for engagement with the original material. Relying on summaries alone risks flattening contextual nuance, obscuring methodological limitations, and overlooking differences in evidentiary quality across sources.

Responsible use therefore requires tracing claims back to primary documents, independently verifying important interpretations, and avoiding the treatment of generated summaries as evidence themselves.

Another valuable use pattern involves surfacing assumptions, uncertainties, dependencies, and information gaps embedded within draft analyses. By externalizing assumptions, researchers can better evaluate how framing choices influence conclusions and where additional evidence may be needed.

For example, a prompt such as:

“Identify assumptions, uncertainties, and information gaps implicit in this draft analysis.”

may highlight assumptions concerning data completeness, implementation feasibility, stakeholder behavior, baseline conditions, or causal interpretation.

The researcher can then assess which assumptions are defensible, which require qualification, and which indicate the need for additional investigation. At the same time, AI-generated assumption lists should not be treated as exhaustive. Important assumptions may still remain invisible without domain expertise or contextual understanding.

Across all examples presented in this chapter, a common principle emerges: interpretive control remains human.

Generative AI systems may assist with exploration, organization, drafting, comparison, synthesis, and revision, but the usefulness and legitimacy of those outputs depend entirely on how they are evaluated, contextualized, and integrated by the researcher.

In each example, the system provides material to think with rather than conclusions to accept automatically. Maintaining this distinction is essential for ensuring that AI-assisted analytical work remains methodologically rigorous, transparent, reviewable, and accountable to the standards of the research or policy environment in which it is used.

At the same time, the examples in this chapter remain intentionally simplified. Real analytical environments involve incomplete information, institutional constraints, competing stakeholder perspectives, evolving evidence, and domain-specific considerations that cannot be fully captured through generic prompts alone.

No prompt or workflow eliminates the need for evidence verification, methodological transparency, contextual understanding, or professional judgment.

The examples presented here should therefore be understood as illustrations of disciplined interaction patterns rather than universal templates.

This chapter has explored what AI-assisted analytical work can look like when generative systems are used to support inquiry rather than replace reasoning or decision-making. The next chapter shifts from examples of use to practices of oversight: how AI-assisted outputs can be reviewed, documented, governed, and evaluated in ways that preserve accountability and analytical integrity.

Where this chapter focused on patterns of interaction, the following chapter focuses on patterns of review, closing the loop between generation, interpretation, verification, and responsibility.