ai-disclosure-block.md

Use when the user is preparing a paper, thesis, or replication package and needs an AI-use disclosure block that conforms to a target journal's policy. Inspects git history, CLAUDE.md, and any AI usage logs to populate the disclosure with concrete uses; does not invent uses that have no evidence.

Generate the AI-use disclosure block that increasingly many journals require, populated from real evidence in the project rather than boilerplate.

This skill is set with user-invocable: true so it appears as a slash command (/ai-disclosure-block) for quick access during manuscript prep.

When to Use

Finalizing a manuscript or replication package for submission
A target journal has a specific “AI Use” section to fill in
Reviewers or co-authors ask: “what did you actually use AI for?”

If the project has no AI-assisted commits, no CLAUDE.md, no usage log, and the user can’t list any uses — there is nothing to disclose. Don’t write a block.

Non-negotiable rules

Every claim about AI use must be backed by evidence: a commit message tagged [AI], a CLAUDE.md, an ai-usage.log, or the user’s explicit statement during this session. Do not write “AI was used for X” without evidence X happened.
Disclose only what the user is comfortable disclosing. Generate a draft, then confirm each row before finalizing.
Avoid hedging defaults like “AI was used minimally” or “only for editing”. They are red flags for reviewers when the actual use was deeper. Be specific.
Match the target journal’s required format. If unknown, default to a comprehensive (JoM-style) disclosure that is the union of common requirements.

Workflow

Identify journal → Gather evidence → Categorize → Draft block → Confirm with user

Step 1 — Identify the target journal

One question: which journal (or “general/unknown”)?

Journal	Format
Journal of Marketing	Structured statement under “AI Use Disclosure”
Marketing Science	Short paragraph in acknowledgments
Quantitative Marketing and Economics	“Use of AI” section in supplementary materials
Journal of Consumer Research	Explicit list in the methods section
General / unknown	Comprehensive default (covers most requirements)

Step 2 — Gather evidence of AI use

# Tagged commits
git log --all --grep='\[AI\]' --pretty=format:'%h %s'

# Existing project memory
test -f CLAUDE.md && head -200 CLAUDE.md

# Usage log
test -f ai-usage.log && cat ai-usage.log

# AI-generated boilerplate left in outputs
grep -rn "Generated with Claude" --include="*.md" --include="*.qmd"

If nothing surfaces, ask the user directly to enumerate uses.

Step 3 — Categorize

Bucket each use:

Literature search — finding papers, building bibliographies
Code generation — writing analysis or visualization code
Code review — audit / debugging help
Data cleaning — preprocessing, transformation
Text analysis — sentiment, topic modeling, qualitative coding
Writing assistance — drafting, editing, polishing prose
Other — enumerate explicitly

Step 4 — Draft the block

Default (general) format:

## AI Use Disclosure

This research used the following AI tools:

| Tool | Version | Use | Verification |
|---|---|---|---|
| Claude Code (Sonnet 4.6) | 2025-Q4 | Wrote initial regression scripts; refactored data cleaning pipeline | All outputs cross-checked against Stata replication |
| Claude Code (Opus 4.7) | 2025-Q4 | Drafted methods section from analysis output | Manually revised; every empirical claim verified |

**What AI was NOT used for:** generating data; selecting the final sample; making theoretical claims; selecting which results to report.

**Validation procedures:**
- All AI-generated code was executed and outputs were inspected.
- All AI-generated citations were verified at Crossref.
- All AI-drafted prose was substantively revised by the authors before submission.

**Prompts and workflows:** AI prompts and the SKILL.md files invoked are archived in the replication package at `replication/ai-prompts/`.

Step 5 — Confirm with the user

Print the draft. Ask the user to confirm each row. Edit per their instructions. Save to disclosure.md (or a path they specify).

Companion skills

literature-review.md — output is a “Literature search” entry
llm-annotation.md — output is a “Text analysis” entry
revision-plan.md — disclosure should be re-run after any major R&R

Stopping criterion

If the only evidence is “I asked Claude one question once,” do not generate a disclosure block. Have the user write a single sentence by hand. The block exists to be informative, not ceremonial.

Notes for extending

Per-author disclosures. When co-authors used AI to different extents, generate a per-author table rather than a single project-wide block.
Linked verification. Auto-link to the pre-report validity check (see report → Pre-report Validity Check) and the literature-review DOI verification log (see literature-review.md → Path B checklist) in the replication package, so reviewers can audit the validation claims rather than take them on faith.