8 New Deep Research Proposals

Published: 15 February 2026 | Category: research

8 New Deep Research Proposals

Category: research
Date: 15 February 2026, 17:36 UTC
Original File: PROPOSED_DEEP_RESEARCH_PROMPTS.md

Based on Claude reports 1 & 2 + research portfolio

Proposed Deep Research Prompts

Based on Findings from Reports 1 & 2 + Your Research Portfolio

PROMPT C: The Detection Tool Equity Crisis

Research Question: How do AI detection tools perform across different English language proficiency levels, and what are the implications for international students in UK higher education?

Background: Liang et al. (2023) found systematic bias against non-native English writers. Your portfolio includes detection tool evaluations. This extends that work with focus on equity.

Specific Investigation:

  1. Replicate Liang et al. methodology with current detection tools (Turnitin, Originality.ai, GPTZero, Copyleaks)
  2. Test with student writing samples from:
    • Native English speakers (UK, US, Australia)
    • Non-native speakers (Chinese, Arabic, Spanish, Hindi first-language)
    • Different proficiency levels (IELTS 6.0, 7.0, 8.0)
  3. Measure false positive rates by language background
  4. Analyze linguistic features triggering misclassification
  5. Document consequences for students (anxiety, appeals, academic consequences)

Data Sources:

Output: Policy brief on equitable assessment + academic paper

PROMPT D: The Essay Mill Legislation Enforcement Gap

Research Question: Why have UK, Australian, and Irish essay mill laws produced virtually no prosecutions, and what enforcement mechanisms would be effective?

Background: Report 1 noted legislation enacted but almost no prosecutions. Your 2018-2020 work mapped the contract cheating industry.

Specific Investigation:

  1. Document all legislation:
    • UK Skills and Post-16 Education Act 2022
    • Australian TEQSA legislation
    • Irish provisions
    • Any others enacted 2022-2026
  2. Interview/key informant:
    • TEQSA (Australia) enforcement data
    • UK Department for Education
    • University legal/compliance officers
    • Why no prosecutions?
  3. Analyze successful prosecutions in other jurisdictions (if any)
  4. Model effective enforcement:
    • Criminal vs civil penalties
    • Extraterritorial jurisdiction
    • Payment processor blocking
    • ISP blocking (as TEQSA does)
  5. Compare to successful enforcement in similar areas (essay mills vs piracy, etc.)

Output: Policy recommendations + comparative law analysis

PROMPT E: The Performance-Learning Paradox Replication Study

Research Question: Is the “performance-learning paradox” (higher homework scores, lower exam scores with AI use) replicated across UK universities, and what assessment designs prevent it?

Background: Report 2 found this pattern at Stanford and CMU. Your teaching module puts you in position to investigate this directly.

Specific Investigation:

  1. Obtain anonymized grade data from:
    • Imperial College modules (your own teaching data)
    • Partner UK universities (Russell Group)
    • Pre-2023 vs 2023-2026 comparison
  2. Correlate with:
    • Assessment type (exam vs coursework)
    • AI policy permissiveness
    • Student self-reported AI use
  3. Identify assessment designs that:
    • Resist AI assistance
    • Maintain validity with AI available
    • Protect learning outcomes
  4. Cost-benefit analysis of redesign

Output: Best practice guide for assessment design + journal article

PROMPT F: Humanizer Tools Efficacy Arms Race

Research Question: How effective are “humanizer” tools (Undetectable AI, StealthWriter, HIX Bypass) at evading current detection systems, and what countermeasures exist?

Background: Report 1 mentioned these tools exist but limited systematic testing. Your technical background enables rigorous evaluation.

Specific Investigation:

  1. Systematic testing protocol:
    • Generate text with GPT-4, Claude, Gemini
    • Process through 5+ humanizer tools
    • Test against 5+ detection tools
    • Measure detection rates before/after humanizing
  2. Document humanizer techniques:
    • Synonym replacement
    • Sentence restructuring
    • “Perplexity/burstiness” adjustment
    • Adversarial patterns
  3. Test detection countermeasures:
    • Can detectors be trained to catch humanized text?
    • What linguistic features persist?
  4. Track tool evolution over 6 months

Ethical Note: Document for defensive purposes only (improving detection, policy)

Output: Technical report + detection tool improvement recommendations

PROMPT G: International Student Conference Meta-Analysis

Research Question: What do 3 years of student-led academic integrity research reveal about student perspectives on cheating, AI, and integrity that faculty research misses?

Background: Your student conferences (2023-2025) contain underutilized data. Student voice often more credible to peers.

Specific Investigation:

  1. Transcribe and code all 3 conference videos
  2. Thematic analysis:
    • What topics do students choose?
    • How do they frame integrity? (positive/negative)
    • What solutions do they propose?
    • How has this changed 2023→2025?
  3. Compare to faculty research:
    • Different questions asked?
    • Different methodologies?
    • Different conclusions?
  4. Identify student research suitable for:
    • Journal publication (with student co-authors)
    • Policy recommendations
    • Further development
  5. Create student research showcase

Output: Meta-analysis paper + student co-authored publications

PROMPT H: CS Curriculum AI Integration Comparison (Student Outcomes)

Research Question: Which of the 26 university approaches to GenAI in CS education produces the best student learning outcomes, and what can others learn from them?

Background: Report 2 documented 26 different approaches. Now test which work.

Specific Investigation:

  1. Partner with 5-10 universities from the 26
  2. Collect outcome data:
    • Exam performance (AI-resistant assessment)
    • Project quality
    • Graduate employment
    • Student satisfaction
    • Academic integrity violations
  3. Correlate with approach:
    • Prohibition vs integration
    • Infrastructure investment
    • Assessment redesign
    • Training provided
  4. Cost-effectiveness analysis
  5. Identify transferable best practices

Output: Evidence-based recommendations for CS educators

PROMPT I: Multilingual Academic Integrity Discourse

Research Question: How do discussions of academic integrity, contract cheating, and AI differ across languages and cultures on social media?

Background: Your portfolio includes Twitter analysis. Extend to multilingual + your multilingual framework.

Specific Investigation:

  1. Collect social media data:
    • English (UK, US, Australia)
    • Spanish (Spain, Latin America)
    • Chinese (Weibo, Zhihu if accessible)
    • Arabic (Twitter MENA)
    • Other languages as possible
  2. Code themes:
    • Attitudes toward cheating
    • Views on AI use
    • Trust in institutions
    • Suggested solutions
  3. Compare:
    • Cultural differences in integrity concepts
    • Different AI adoption patterns
    • Varying policy preferences
  4. Translate key insights

Output: Cross-cultural analysis + policy implications for international education

PROMPT J: Assessment Redesign Efficacy Database

Research Question: Which specific assessment redesign strategies successfully maintain integrity and learning outcomes in the GenAI era?

Background: Report 2 noted shift to assessment redesign but limited evidence of what works.

Specific Investigation:

  1. Systematic review:
    • Identify all papers on assessment redesign 2023-2026
    • Extract redesign strategies
    • Categorize by discipline, level, assessment type
  2. Efficacy coding:
    • Which maintain integrity?
    • Which protect learning?
    • Which are feasible at scale?
    • Cost to implement?
  3. Create database:
    • Assessment type
    • Redesign strategy
    • Evidence of efficacy
    • Implementation guidance
  4. Gap analysis:
    • What hasn’t been tried?
    • What needs evaluation?

Output: Searchable database + best practice guide

Immediate (Start This Week):

  1. C - Detection Equity (builds on Liang et al., high policy relevance)
  2. G - Student Conference Analysis (data already exists, unique contribution)

Short-term (Next 2-4 Weeks):

  1. E - Performance-Learning Paradox (uses your teaching data)
  2. F - Humanizer Tools (technical, timely)
  3. D - Legislation Enforcement Gap (policy impact)

Medium-term (1-2 Months):

  1. H - CS Curriculum Outcomes (requires partnerships)
  2. I - Multilingual Discourse (scale up gradually)
  3. J - Assessment Redesign Database (systematic review)

SELECTION CRITERIA

Each prompt selected based on:

Which 2-3 should we start immediately?

Original file: PROPOSED_DEEP_RESEARCH_PROMPTS.md