Overview

Lesson 5.4: Building Your Personal AI Research Protocol

This lesson teaches researchers how to synthesize learning from the entire Level 2 certification into a personal AI research protocol, a customized standard operating procedure that documents your approach to AI in every phase of research. A personal protocol transforms ad hoc AI use decisions into principled, consistent practice. It specifies which tools you use for which tasks, what verification and documentation standards apply to each task category, how you disclose AI use, and how you update the protocol as experience and external standards evolve. The lesson walks through the eight-step protocol development process, provides a complete worked example of a graduate researcher's protocol, and shows how team protocols enable consistent AI use across a research group.

Title

Lesson 5.4: Building Your Personal AI Research Protocol

Purpose

This lesson teaches researchers how to synthesize learning from the entire Level 2 certification into a personal AI research protocol, a customized standard operating procedure documenting your approach to AI in research. You will learn to document your personal standards, create decision frameworks, establish verification procedures, and develop a living document you will refine as experience grows.

By the end of this lesson you will be able to: (1) articulate the research values that should guide your AI use decisions; (2) categorize your research tasks by AI suitability using the RISK matrix; (3) specify verification and documentation standards for each task category; (4) draft a complete personal AI research protocol covering all phases of your research workflow; and (5) plan a protocol review cycle that keeps your protocol current as tools and standards evolve.

Core Concepts

A personal AI research protocol is not a policy imposed by an institution. It is a statement of how you choose to work. That distinction matters because it means the protocol reflects your values and your specific research context, not a generic set of rules. It is also a commitment: researchers who articulate their principles explicitly are more likely to act consistently with them under pressure than those who rely on in-the-moment judgment.

Why Personal Protocols Outperform Ad Hoc Decisions

Ad hoc AI use decisions have three failure modes. First, inconsistency: the same task gets different AI treatment in different projects, making it impossible to reproduce your methods or explain them to reviewers. Second, scope creep: without explicit boundaries, AI use tends to expand into tasks where it adds risk without proportionate value. Third, verification gaps: without a documented verification standard, high-stakes AI outputs may receive inadequate checking simply because no explicit protocol requires it.

A written protocol prevents all three failure modes. It also communicates your standards to collaborators, research assistants, and students who work within your research group.

The Five Components of a Personal Protocol

Component 1 - Research Values Statement: A 200-500 word articulation of what you care most about in how research is conducted. For example: 'I prioritize rigor over speed. I will use AI to accelerate routine tasks but will not sacrifice verification thoroughness to gain time. I prioritize transparency and will always disclose AI use proactively.' Or: 'I prioritize innovation and am willing to use AI in exploratory ways, provided I apply verification proportional to the stakes involved.' The values statement is not public-facing. It is the foundation for all other protocol decisions.

Component 2 - Task Classification: A categorized list of your most common research tasks, each assessed using the RISK matrix and classified as AI-appropriate, hybrid, or human-primary. This classification drives the decision tree for each task category.

Component 3 - Verification Standards: Specific verification requirements for each task category. Not a single standard for all tasks, a differentiated standard calibrated to risk. Example: 'All citations in main arguments: full verification against primary source. Background citations: spot-check 25% random sample. Code for primary outcome analyses: test on sample with known results + second-person review. Exploratory code: test on sample with known results.'

Component 4 - Documentation Procedures: How AI use is logged, what format the log uses, where it is stored, and how it is retained. Specifically: the AI use log template, the tool version tracking approach, and the procedure for drafting disclosure statements from the log at submission time.

Component 5 - Disclosure Standards: Pre-drafted template disclosure statements for the four main submission contexts (journal manuscript, grant application, dissertation/thesis, conference abstract), ready to be customized for each submission.

Protocols as Living Documents

A protocol that is written once and never updated is a liability rather than an asset. It creates the appearance of systematic practice without the substance. The protocol should be reviewed and updated at least quarterly (or after each major project completion) as you accumulate experience about which AI applications produce reliable results in your workflow and which produce errors that require significant correction. External reviews, when your institution, target journals, or major funders update their AI policies, should also trigger a protocol update.

Team Protocols

Research groups benefit from a shared team protocol that establishes consistent standards for all members. A team protocol addresses: which tools team members may use, what the documentation standard is, how verification responsibilities are distributed (e.g., research assistant generates AI output, PI reviews), what the standard disclosure language is for team publications, and what the process is for proposing protocol updates. Creating a team protocol collaboratively, with input from all members, increases buy-in and often surfaces important considerations that the PI alone would miss.

Practical Applications

The following worked examples illustrate what complete protocols look like at different career stages and research contexts.

Example 1: Graduate Researcher Protocol (First Year)

Values: 'I am still developing my research expertise and need to be careful not to outsource learning to AI. I will use AI to accelerate routine tasks but will always read primary literature myself, conduct my own analyses, and write my own arguments.'

Task Classification:
- Literature screening: AI-appropriate [RISK score 5], Use Semantic Scholar AI features for initial screening; verify against database search
- Individual paper summarization: AI-appropriate [RISK score 4], Use for papers outside my specialty; always read papers in my specialty myself
- Survey item drafting: AI-appropriate [RISK score 4], Generate candidates; select and refine myself
- Writing literature review: Hybrid [RISK score 8], AI drafts paragraph summaries; I synthesize themes
- Statistical analysis: Human-primary [RISK score 10], Conduct analysis myself; AI can explain methods I am learning
- Novel theoretical contribution: Human-primary [RISK score 12], No AI

Verification Standards:
- All citations: Full verification before including in any draft
- AI-drafted text: Read for accuracy and verify against sources; rewrite any passage I cannot independently defend
- Code: Test on sample data with manually calculated expected results

Disclosure: Log all AI use with tool, task, and date. Draft disclosure statement before each submission.

Review cycle: Monthly review with advisor.

Example 2: Established Researcher Protocol (Based on Experience)

Values: 'I prioritize rigor in data analysis and statistical interpretation. These are my areas of expertise and I do not want AI to displace my judgment. I am comfortable using AI extensively for literature tasks and writing support because I have sufficient expertise to verify these efficiently.'

Task Classification (refined through two years of experience):
- Literature tasks: AI-appropriate for all sub-tasks, Developed efficient spot-checking procedures; 5-10% error rate historically, manageable
- Writing assistance: AI-appropriate for clarity edits, abstract polish, response-to-reviewer letters, Human-primary for all novel arguments
- Data analysis: Human-primary for all primary outcome analyses, AI-appropriate for exploratory visualization and code generation, with verification
- Grant writing: Hybrid, AI assists with specific aims narrative clarity; human-primary for scientific content

Verification Standards (refined):
- Citations: Full for main arguments; 20% spot-check for background (based on historical 4% error rate)
- Analysis code: Test on sample + second-person review for all primary outcome code

Documentation: AI use log maintained in Notion; transfer to research folder at project close.

Review cycle: Quarterly, with team; full revision annually or when major policy changes occur.

The Eight-Step Protocol Development Process

Step 1: Write a 300-500 word research values statement. What do you care most about in how research is conducted? Where do you prioritize rigor over speed, or innovation over caution?

Step 2: List your 15-20 most common research tasks.

Step 3: Score each task on the RISK matrix and classify as AI-appropriate, hybrid, or human-primary.

Step 4: For each task category, specify: which AI tools, what the verification standard is, and what the documentation requirement is.

Step 5: Identify the top 3-5 time-consuming AI-appropriate tasks. These are your highest-return AI use cases and the priority for implementation.

Step 6: Draft template disclosure statements for each submission context you use (journal, grant, dissertation).

Step 7: Compile components 1-5 into a single document. Format it as a quick reference: values first, then decision tables for each task category, then verification checklists, then disclosure templates.

Step 8: Schedule the first protocol review. If this is your first protocol, schedule a review after your first major project using it (typically 3-6 months). Add a recurring reminder to your calendar.

Sharing the Protocol with Collaborators

When beginning a collaboration, share your protocol proactively. This prevents misunderstandings about AI use (collaborators may have very different practices), enables the team to discuss and align on standards early rather than after a conflict arises, and demonstrates methodological transparency. You can share your full protocol or a simplified one-page summary covering the most important decision rules and verification standards.

Key Takeaways

A personal AI research protocol transforms ad hoc AI use decisions into principled, consistent practice. Researchers who articulate their principles explicitly are more likely to act consistently with them than those relying on in-the-moment judgment, particularly under deadline pressure when shortcuts are tempting.

The five core protocol components are: a research values statement, a task classification using the RISK matrix, differentiated verification standards calibrated to risk, documentation procedures, and pre-drafted disclosure templates. Each component serves a specific function; omitting any one creates gaps in practice.

Protocols are living documents, not set-and-forget artifacts. Regular review, at least quarterly, incorporating actual experience from completed projects is what transforms a good first draft into a reliable operational guide.

Team protocols enable consistent AI use across research groups. Establishing shared standards reduces inconsistency in how team members handle AI use, creates a shared accountability structure, and surfaces important considerations that individual protocols may miss.

The research values statement is the foundation of the entire protocol. Decisions about task classification, verification intensity, and disclosure practice all follow from values. A researcher who values rigor above speed makes different protocol decisions than one who values innovation above caution, both can be valid; both should be explicit.

A completed personal protocol is the synthesis product of this certification. It represents not just what you have learned but a commitment to how you will work. The goal is not perfect compliance. It is thoughtful practice that evolves as you learn more about what AI does well in your specific research context and what it does not.

Personal AI Research Protocol Template

Use this template as a starting framework. Customize every section to reflect your specific research context, field norms, and values.

---
[Your Name] - Personal AI Research Protocol
Version: 1.0
Date created: _______________
Last reviewed: _______________
Next scheduled review: _______________

Section 1: Research Values
[Write 300-500 words articulating what you prioritize in research practice and how those values should shape your AI use decisions.]

Section 2: Task Classification
| Task | RISK Score | Classification | AI Tools Permitted | Notes |
|---|---|---|---|---|
| Literature search | | | | |
| Paper summarization | | | | |
| Survey/instrument drafting | | | | |
| Writing assistance | | | | |
| Data cleaning code | | | | |
| Statistical analysis | | | | |
| Grant writing | | | | |
| [Add your specific tasks] | | | | |

Section 3: Verification Standards
High-stakes outputs (primary analyses, main argument citations): [Specify verification requirement]
Moderate-stakes outputs (background literature, methods text): [Specify verification requirement]
Low-stakes outputs (structural outlines, first drafts for revision): [Specify verification requirement]

Section 4: Documentation Procedure
Log format: [Describe format or link to template]
Log storage location: _______________
Retention period: _______________

Section 5: Disclosure Templates
Journal manuscript: 'AI tools were used in the preparation of this manuscript as follows: [tool] was used for [task]. All AI-assisted content was reviewed by [authors/PI] for accuracy and verified against primary sources.'

Grant application: 'AI writing assistance was used to improve the clarity of narrative sections. All scientific content, budget rationale, and compliance statements were independently verified by the investigators.'

Dissertation/thesis: 'AI tools were used for the following tasks: [list]. Use was disclosed to and approved by the dissertation committee. All research methods, analysis, interpretation, and conclusions are the author's own.'

Section 6: Review and Update Plan
Scheduled reviews: [List dates or frequency]
Triggers for unscheduled review: institutional policy update, major journal guideline change, significant funder guidance revision, completion of major project
---