How Do You Define a Document Corpus for Research Analysis?

How Do You Decide Which Documents Belong in Your Research Dataset?

You are planning to analyze university policies on generative AI. You begin searching institutional websites and quickly discover a mixture of formal policies, academic integrity guidelines, teaching resources, frequently asked questions, administrative memoranda, and public announcements.

Some universities publish one comprehensive document. Others distribute their requirements across several webpages. A few have multiple versions, while some relevant materials are difficult to locate.

Should every document mentioning generative AI be included? Should you restrict the study to official policies? How do you prevent the corpus from becoming a collection of whatever happened to appear in your searches?

The challenge is to establish a documentary dataset whose boundaries are conceptually justified, whose collection procedures are transparent, and whose contents can support the intended analysis.

Define the Corpus Through the Research Question and Explicit Eligibility Criteria

Building a Coherent and Defensible Document Corpus

What Is a Document Corpus?

A document corpus is a defined collection of documentary materials assembled for systematic analysis.

The materials may include policies, reports, correspondence, institutional records, webpages, archival documents, or other recorded artifacts.

For example, a corpus for a study of institutional AI governance might contain official university policies and related guidance documents issued during a specified period.

The defining feature is not the number of documents but the existence of a coherent relationship between the collection and the research question.

A folder containing hundreds of downloaded PDFs is not necessarily a defensible research corpus. Its contents must be selected according to a transparent methodological rationale.

Why Corpus Definition Matters

The boundaries of a corpus influence which patterns become visible and which remain outside the analysis.

For example, a study examining only formal AI policies may identify institutional obligations but overlook the more practical guidance communicated through teaching resources.

Conversely, including every institutional webpage mentioning AI may introduce promotional content unrelated to governance.

Neither selection is automatically wrong.

The appropriate boundary depends on whether the research concerns formal policy, public communication, institutional discourse, or another phenomenon.

Corpus construction is therefore an analytical decision rather than merely an administrative step.

Start by Defining the Documentary Population

The documentary population is the broader set of materials that would be relevant to the research question under the study's conceptual and temporal boundaries.

For example, the population might consist of all official generative AI policies issued by accredited universities within a particular jurisdiction during a defined academic year.

That differs from a population consisting of all public institutional communications about AI.

The first emphasizes formal authorization. The second includes a wider range of representations.

Researchers should specify the population before deciding which accessible documents to collect.

Otherwise, the available materials may inadvertently determine the research question rather than the reverse.

Distinguish the Target Corpus from the Accessible Corpus

The target corpus contains the documents that meet the conceptual eligibility criteria.

The accessible corpus contains the eligible materials that the researcher can actually retrieve.

These may differ because of restricted access, incomplete archives, removed webpages, or unavailable institutional records.

For example, a university may maintain internal AI guidance that is not publicly accessible.

Failure to retrieve that guidance does not establish that it does not exist.

Researchers should distinguish the absence of a document from the absence of access to it.

This distinction becomes particularly important when comparing institutions or making claims about policy coverage.

What Should Inclusion Criteria Specify?

Inclusion criteria establish which documents qualify for analysis.

Depending on the research question, they may address document type, issuing authority, geographical scope, publication period, institutional status, language, subject matter, and intended audience.

For example, a study of formal university AI assessment requirements might include documents issued or formally endorsed by institutional authorities that contain provisions governing AI use in assessment.

The criteria should be sufficiently precise that another researcher could understand why a particular document was included.

However, precision should not be confused with unnecessary restriction. Criteria should follow the research question rather than arbitrary preferences for particular formats.

How Should Exclusion Criteria Be Developed?

Exclusion criteria identify materials that fall outside the intended documentary population.

For example, a study of authorized institutional AI policies might exclude personal blog posts, commercial advertisements, informal commentary, and documents produced by external organizations without institutional endorsement.

Superseded policy versions might also be excluded if the study concerns current requirements.

However, those versions would become relevant if the research investigated policy change over time.

Exclusion criteria should therefore be justified conceptually rather than selected merely to reduce workload.

Researchers should document significant exclusions and avoid silently removing materials because their content complicates an emerging interpretation.

How Do You Decide Which Document Types Qualify?

The term document can encompass written, digital, visual, and other recorded materials.

However, a particular study may require only some of these types.

For example, a study of official policy provisions might include signed policies, institutional regulations, and formally approved guidelines.

A study of public institutional communication might also include webpages, infographics, announcements, and social media statements.

The broader question of what counts as a research document should be resolved in relation to the study's intended evidence.

Documentary eligibility should not depend solely on whether a file is a PDF, Word document, or webpage.

Should the Corpus Include Only Official Documents?

Not necessarily.

Official status matters when the research concerns formal institutional requirements or authorized positions.

However, unofficial materials may be essential when investigating informal communication, professional practices, or competing interpretations.

For example, teacher-generated guidance documents may reveal how institutional AI policies are interpreted locally.

These materials should not automatically be treated as official policy.

Researchers can include both categories when their distinct functions are analytically relevant.

The corpus should preserve information about authorship, authorization, and intended audience.

How Do You Establish Temporal Boundaries?

Documentary research often requires a clearly defined time period.

For example, a study may examine policies effective during one academic year or compare documents issued before and after a major institutional change.

Researchers should distinguish publication date, effective date, revision date, and retrieval date.

These dates do not necessarily refer to the same event.

A policy published several years earlier may remain in force during the study period.

Conversely, a recently uploaded document may reproduce an older policy.

Temporal eligibility should therefore follow the relevant status of the document rather than assuming that the date displayed on a webpage establishes its substantive period of application.

How Do You Handle Multiple Versions of the Same Document?

Documents may be revised, replaced, corrected, or republished.

Researchers should determine whether the study concerns current content, historical development, or both.

For a cross-sectional analysis of current policies, the relevant version may be the one effective at a defined reference date.

For a longitudinal analysis, successive versions may constitute distinct analytical observations.

Researchers should record version identifiers, dates, retrieval sources, and relationships between documents.

Failing to distinguish versions may produce duplicate counts or misleading conclusions about policy change.

How Should Duplicate Documents Be Managed?

The same document may appear in multiple locations, such as an institutional policy repository, faculty webpage, and downloadable archive.

Researchers should distinguish duplicate copies from substantively different versions.

For example, identical copies of one policy should not ordinarily be counted as separate policies merely because they were downloaded from three webpages.

However, repeated publication across institutional channels may be relevant when the research concerns document circulation or public communication.

The treatment of duplicates should follow the unit of analysis.

Researchers should preserve enough provenance information to explain how duplicates were identified and handled.

What Is the Difference Between a Document and an Analytical Case?

A document is a collected material, while an analytical case is the entity about which the study intends to make conclusions.

These may be identical, but they need not be.

For example, a study may collect 120 documents from 30 universities while treating each university as one institutional case.

Another study may treat each policy provision as the unit of analysis.

Researchers should distinguish the unit of collection, coding unit, and unit of analysis.

Otherwise, institutions with more published documents may appear to have more policies or greater commitment simply because they produce more textual material.

How Should Documents from Different Institutions Be Compared?

Institutions may organize information differently.

One university may issue a comprehensive AI policy, while another distributes similar requirements across academic integrity regulations, assessment guidelines, and teaching resources.

If the study compares institutional governance, examining one document per institution may create artificial differences.

Researchers should determine whether multiple documents need to be combined to represent an institutional case.

However, combining documents should not erase differences in authority, audience, or purpose.

The analysis should retain relevant information about each document's function.

How Do You Develop a Document Retrieval Strategy?

A retrieval strategy specifies where and how eligible documents will be located.

Sources may include institutional repositories, official websites, archives, administrative offices, databases, and authorized organizational records.

Researchers should document search terms, source locations, retrieval dates, and relevant access procedures.

For example, a university policy study might search official institutional policy repositories and academic integrity webpages using consistent terminology.

However, institutional websites differ in structure and search functionality.

Researchers may need supplementary navigation or direct institutional inquiries, with procedures documented transparently.

What If Search Results Depend on the Search Engine?

General search engines may return different results depending on indexing, ranking, search terms, and changes to webpages.

Consequently, search results should not automatically be treated as a complete documentary population.

Researchers should prioritize appropriate authoritative repositories and establish repeatable retrieval procedures where feasible.

For example, a policy might be available in an institutional repository but absent from a general web search.

Search engines can support discovery, but their results should be interpreted as a retrieval mechanism rather than evidence that every eligible document has been identified.

How Do You Document the Corpus?

A corpus register can record the identity, provenance, eligibility, and retrieval status of each document.

Useful fields may include document identifier, title, issuing organization, document type, publication date, effective date, version, source location, retrieval date, inclusion decision, and reason for exclusion.

Researchers may also record language, access restrictions, document relationships, and relevant contextual notes.

The register should distinguish metadata from analytical findings.

For example, recording that a policy was issued by a university is metadata, while interpreting its allocation of responsibility is part of analysis.

A clear register supports transparency and reduces confusion during corpus development.

What Does a Corpus Register Look Like?

These entries are illustrative. The actual register should reflect the study's documentary population and analytical requirements.

Should the Corpus Be Fixed Before Analysis Begins?

Some studies benefit from establishing a fixed corpus before substantive coding.

This is particularly useful when the researcher intends to compare predefined categories or estimate the prevalence of documentary features.

Other studies use iterative qualitative procedures in which analysis reveals additional relevant document types or sources.

Both arrangements can be defensible when consistent with the research design.

However, changes to corpus boundaries should be documented.

Researchers should avoid selectively adding or removing documents merely because they support or contradict an emerging interpretation.

How Do You Handle Documents in Different Languages?

Multilingual corpora require decisions about eligibility, translation, coding, and comparability.

For example, institutional policies may be available in English and local languages.

Researchers should determine whether both versions are authoritative, whether one is a translation, and whether substantive differences exist.

Translation may alter terminology or obscure distinctions important to the analysis.

Researchers should document language decisions and avoid treating translations as identical without appropriate evaluation.

Excluding documents because of language limitations may also restrict the documentary population.

How Do You Address Missing or Inaccessible Documents?

Missing materials can affect corpus completeness and the interpretation of institutional differences.

For example, a university may have internal AI guidelines that are unavailable to external researchers.

The absence of an accessible document should not automatically be coded as the absence of institutional guidance.

Researchers should record unsuccessful retrieval attempts, access restrictions, and uncertainty about whether documents exist.

The treatment of missing or unavailable documents should be considered when evaluating corpus adequacy.

When Should Documents Be Sampled Rather Than Collected Exhaustively?

A complete census of eligible documents may be feasible when the documentary population is small and clearly bounded.

For large or open-ended populations, sampling may be more practical or methodologically appropriate.

For example, a study of institutional policy documents may feasibly include all eligible policies from a defined set of universities, while a study of public discussion posts may require sampling.

The decision about sampling documents versus analyzing all available documents should reflect the research question, corpus size, analytical depth, and intended inference.

A large corpus is not automatically more representative, and a small purposive corpus is not automatically inadequate.

How Do You Assess Corpus Adequacy?

Corpus adequacy concerns whether the assembled materials provide sufficient coverage and relevance for the intended analysis.

For quantitative document analysis, adequacy may depend on the sampling frame, selection probabilities, and the number of independent analytical units.

For qualitative documentary analysis, it may concern the relevance, diversity, contextual richness, and analytical contribution of the materials.

Researchers should not treat the number of documents alone as evidence of adequacy.

Ten highly relevant institutional policies may be more informative for one question than hundreds of loosely related webpages.

The evaluation should follow the intended claims.

How Do You Avoid Corpus Selection Bias?

Selection bias can arise when documents are included or excluded in ways related to the phenomenon being investigated.

For example, collecting only institutions with easily accessible AI policies may overrepresent universities with more developed public communication systems.

Similarly, selecting documents because they contain explicit references to ethical responsibility may exclude policies that address responsibility using different terminology.

Researchers should establish clear retrieval and eligibility procedures and examine how these procedures might systematically exclude relevant materials.

Where limitations remain, conclusions should be appropriately bounded.

How Should Corpus Construction Be Reported?

The methodology should identify the target documentary population, eligibility criteria, retrieval sources, search procedures, time boundaries, and final corpus composition.

Researchers should report how duplicates, versions, inaccessible materials, and changes to criteria were handled.

Where useful, a document selection flow diagram or corpus inventory may support transparency.

However, researchers should not assume that every documentary study must follow a systematic-review reporting framework.

The reporting procedure should fit the study's design and make its evidence-selection decisions understandable and auditable.

Defining a Corpus of University Generative AI Policies

The resulting corpus has explicit conceptual, institutional, and temporal boundaries. Its defensibility comes from transparent selection and alignment with the research question rather than the sheer number of collected documents.

Common Mistakes When Defining a Document Corpus

How to Establish Defensible Corpus Boundaries

Define the documentary population before collecting materials, then establish retrieval and eligibility procedures that reflect the intended analysis.

A strong corpus definition makes it possible to explain why each document belongs and why other materials do not.

Check Your Document Corpus Before Analysis

Frequently Asked Questions About Defining a Document Corpus

A Defensible Corpus Has Clear Boundaries and a Clear Analytical Purpose

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.