AI Cannot Judge Feasibility — So It Suggests Impossible Studies
A doctoral student contacted me in tears. She’d spent four months developing a dissertation proposal with ChatGPT’s
help. The topic seemed perfect: understanding the lived experiences of adolescent sexual abuse survivors and how therapy
helped their healing process. She submitted to her IRB. Rejected immediately with a note: “Research involving minors
discussing trauma experiences poses unacceptable psychological risk. This study cannot be approved as proposed.” Four
months wasted. Not because the topic was unimportant—it addressed a genuine problem. But because AI suggested a study
that was fundamentally not feasible. No IRB would approve research asking minors to recount sexual trauma for a
student’s dissertation. Here’s what students need to understand: AI generates topics based on what sounds academically
interesting without any ability to assess whether those topics can actually be executed. It doesn’t know IRB
requirements, data accessibility constraints, timeline limitations, or measurement validity issues. It suggests studies
that are legally, ethically, or logistically impossible—and students don’t realize until months of wasted work.
Let me break down what feasibility actually requires, because students often don’t understand all the constraints they’re working within.
For a dissertation topic to be viable, it must pass all four tests: 1. IRB-Safe: Research design must be approvable by your Institutional Review Board. Studies creating excessive risk to participants, involving certain vulnerable populations, or lacking adequate protections won’t be approved—meaning you can’t conduct the research. 2. Accessible: Data or participants must be available to you within your timeline. If you can’t recruit participants, access existing datasets, or reach your research sites, the study isn’t feasible regardless of its theoretical merit. 3. Measurable: You must be able to get valid data on your constructs. If participants won’t answer honestly, behaviors can’t be observed, or phenomena can’t be measured reliably, your data will be worthless. 4. Time-Bounded: Data collection and analysis must be completable within 12-18 months typically. Studies requiring years of data accumulation or longitudinal tracking aren’t feasible for most doctoral students. If your topic fails any one of these tests, it’s not viable—even if it’s original, problem-driven, and theoretically sound. According to research from Stanford’s Graduate School of Education, feasibility issues are the leading cause of abandoned dissertation projects, with students discovering 1-2 years into their work that their topics cannot be completed as designed.
Students often focus on conceptual elements (originality, theory, problem-driven focus) while neglecting practical constraints: “This is such an important topic!” → But can you actually study it within IRB rules? “The literature gap is clear!” → But can you access data to fill that gap? “This would make a great contribution!” → But can you complete it before running out of time and funding? AI amplifies this problem by suggesting topics without any feasibility assessment at all.
Let me show you specific categories of infeasible topics AI routinely suggests.
AI frequently suggests studying populations that IRBs protect with extra scrutiny: AI suggestion: “Phenomenological study of childhood trauma survivors’ healing journeys” Feasibility problem: Trauma survivors are a vulnerable population. Asking them to recount traumatic experiences for research purposes creates psychological risk. IRBs require extensive protections: mental health screening, counseling availability, careful informed consent, often consultation with trauma specialists. Most student dissertations can’t provide adequate protections. Even with protections, many IRBs won’t approve studies asking vulnerable populations to recall trauma for student research. AI suggestion: “Experiences of undocumented immigrants accessing healthcare services” Feasibility problem: Undocumented immigrants are vulnerable (at legal risk, economically disadvantaged). Researching their experiences creates risks if documentation of participation could be discovered or subpoenaed. Most students cannot provide sufficient legal protections or anonymization to get IRB approval. AI suggestion: “Children’s perceptions of divorce and family conflict” Feasibility problem: Research with minors (under 18) requires both parental consent AND child assent. This creates recruitment challenges. Additionally, asking children about family conflict raises questions about whether they can truly voluntarily participate or may feel coerced. Studying topics that might reveal abuse creates mandatory reporting obligations that complicate research ethics.
AI suggests longitudinal or time-dependent studies without understanding timeline constraints: AI suggestion: “The long-term effects of a new teacher mentoring program on five-year retention rates” Feasibility problem: If the program started this year, you need five years of data before you can analyze retention effects. Your dissertation timeline is typically 2-3 years maximum from proposal approval to defense. You can’t wait five years for data to exist. AI suggestion: “How organizational culture changes during CEO succession” Feasibility problem: You need baseline data from before succession, then post-succession data over time to assess change. You can’t control when CEO successions occur. Waiting for appropriate succession events, then collecting multi-wave data, easily exceeds dissertation timelines. AI suggestion: “Evaluating the impact of a new state education policy on student achievement” Feasibility problem: New policies need time to be implemented, time for effects to manifest, and time for data to be collected and made available. Even if the policy was enacted recently, comprehensive impact data may not exist for 3-5 years. You can’t complete a dissertation waiting for data that doesn’t exist yet.
AI suggests topics on controversial issues without understanding that social desirability bias will make data invalid: AI suggestion: “Interviewing affluent homeowners about their attitudes toward affordable housing development in their neighborhoods” Feasibility problem: People know the socially acceptable answer is supporting affordable housing. Even if they privately oppose it, they’re unlikely to admit opposition to a researcher recording their responses. You’ll get politically correct responses that don’t reflect actual attitudes, rendering your data invalid for understanding genuine perspectives. AI suggestion: “Exploring racial attitudes and implicit biases among teachers” Feasibility problem: Teachers know discussing racial bias is professionally risky. They won’t honestly report biased attitudes in interviews, especially recorded ones. Social desirability bias makes the data worthless for understanding actual attitudes versus stated attitudes. AI suggestion: “Understanding how managers make hiring decisions when they encounter candidates from underrepresented groups” Feasibility problem: Managers won’t admit to discriminatory hiring practices, which are illegal. Self-report data on hiring decisions involving protected characteristics will be heavily filtered through social desirability, preventing you from getting valid data on actual decision-making processes.
When experienced dissertation advisors review proposed topics, they systematically assess feasibility using criteria AI cannot apply.
We assess whether data currently exists or can be collected within dissertation timelines: Existing data: If you’re using secondary data, does it exist? Do you have access? What permissions are required? How long does access take? Primary data collection: Can you realistically recruit your sample size within 3-6 months? What’s the typical response rate for your population and method? Are there seasonal factors affecting recruitment? Temporal requirements: If your research examines recent policies or events, has enough time passed for relevant data to exist? Do you need baseline data that doesn’t exist yet? AI can’t check whether data exists or is accessible—it just suggests what would be ideal to study.
We evaluate whether you can ethically and practically reach your target population: Vulnerable populations: If your population includes minors, trauma survivors, prisoners, undocumented individuals, or others protected by IRB, what special protections are needed? Can you realistically provide them? Access barriers: Do you need institutional permission (schools, hospitals, prisons)? How long do those approval processes take? What’s the likelihood of receiving permission? Recruitment feasibility: Can you reach this population? Do you have contacts or entrée? What recruitment strategies are available? What response rates can you expect? AI doesn’t know access requirements for different populations or settings.
We predict IRB concerns and assess approval likelihood: Risk level: Does your study create physical, psychological, social, legal, or economic risks for participants? Are risks minimal or more than minimal? Vulnerable populations: If studying protected populations, can you demonstrate sufficient protections and justify why the research benefits outweigh risks? Informed consent: Can participants truly consent voluntarily, or are there coercion concerns? Do you need parental consent, guardian consent, or institutional consent in addition to individual consent? Approval timeline: Will this be expedited review (2-4 weeks) or full board review (2-3 months)? Are multiple revision rounds likely? AI has no knowledge of IRB processes or requirements. It suggests topics without considering whether they’re approvable.
We assess whether you can get valid data or if social desirability will corrupt responses: Sensitive topics: Are you asking about illegal behaviors, socially undesirable attitudes, or controversial positions? Will participants answer honestly? Professional risk: Does honest answering create professional or career risks? Teachers won’t honestly report pedagogical failures. Managers won’t admit discriminatory practices. Social acceptability: Is there clearly a “right” answer participants know they should give, even if it’s not their true attitude or behavior? Alternative measurement: If self-report will be biased, are there alternative approaches (observations, archival data, indirect measures) that could yield more valid data? AI doesn’t understand measurement validity threats from social desirability.
Let me show you actual examples of topics students developed with AI help that turned out to be infeasible.
AI-suggested topic: “Phenomenological study of how incarcerated mothers maintain relationships with their children” Why it seemed good: Addresses important social issue, understudied population, clear problem focus Feasibility problems:
AI-suggested topic: “How school principals describe their decision-making about disciplining students from different racial backgrounds” Why it seemed good: Addresses equity concerns, asks about actual decisions, focuses on leaders Feasibility problems:
Don’t waste months on topics that can’t be completed. Get feasibility assessment from scholars who understand constraints you’re working within.
If your topic has feasibility concerns but might be viable with proper design: Protocol development: Designing studies that maximize approval likelihood Risk mitigation: Identifying participant protections that address IRB concerns Alternative approaches: Finding methodological modifications that reduce risk while preserving research value Documentation preparation: Helping you prepare IRB applications that anticipate and address reviewer concerns
Feasibility is one component of comprehensive planning: Get dissertation help that ensures topics are not only original and problem-driven but also completable within your program’s constraints.
AI suggests topics based on what sounds academically interesting. It has no ability to assess whether topics can actually be researched within IRB, timeline, access, and measurement constraints doctoral students face. Only experienced advisors can:
Feasibility Means More Than “Good Idea”
Let me break down what feasibility actually requires, because students often don’t understand all the constraints they’re working within.
The Four Feasibility Requirements
For a dissertation topic to be viable, it must pass all four tests: 1. IRB-Safe: Research design must be approvable by your Institutional Review Board. Studies creating excessive risk to participants, involving certain vulnerable populations, or lacking adequate protections won’t be approved—meaning you can’t conduct the research. 2. Accessible: Data or participants must be available to you within your timeline. If you can’t recruit participants, access existing datasets, or reach your research sites, the study isn’t feasible regardless of its theoretical merit. 3. Measurable: You must be able to get valid data on your constructs. If participants won’t answer honestly, behaviors can’t be observed, or phenomena can’t be measured reliably, your data will be worthless. 4. Time-Bounded: Data collection and analysis must be completable within 12-18 months typically. Studies requiring years of data accumulation or longitudinal tracking aren’t feasible for most doctoral students. If your topic fails any one of these tests, it’s not viable—even if it’s original, problem-driven, and theoretically sound. According to research from Stanford’s Graduate School of Education, feasibility issues are the leading cause of abandoned dissertation projects, with students discovering 1-2 years into their work that their topics cannot be completed as designed.
Why Students Overlook Feasibility
Students often focus on conceptual elements (originality, theory, problem-driven focus) while neglecting practical constraints: “This is such an important topic!” → But can you actually study it within IRB rules? “The literature gap is clear!” → But can you access data to fill that gap? “This would make a great contribution!” → But can you complete it before running out of time and funding? AI amplifies this problem by suggesting topics without any feasibility assessment at all.
What AI Suggests That Cannot Be Done
Let me show you specific categories of infeasible topics AI routinely suggests.
Studies Requiring Vulnerable Populations
AI frequently suggests studying populations that IRBs protect with extra scrutiny: AI suggestion: “Phenomenological study of childhood trauma survivors’ healing journeys” Feasibility problem: Trauma survivors are a vulnerable population. Asking them to recount traumatic experiences for research purposes creates psychological risk. IRBs require extensive protections: mental health screening, counseling availability, careful informed consent, often consultation with trauma specialists. Most student dissertations can’t provide adequate protections. Even with protections, many IRBs won’t approve studies asking vulnerable populations to recall trauma for student research. AI suggestion: “Experiences of undocumented immigrants accessing healthcare services” Feasibility problem: Undocumented immigrants are vulnerable (at legal risk, economically disadvantaged). Researching their experiences creates risks if documentation of participation could be discovered or subpoenaed. Most students cannot provide sufficient legal protections or anonymization to get IRB approval. AI suggestion: “Children’s perceptions of divorce and family conflict” Feasibility problem: Research with minors (under 18) requires both parental consent AND child assent. This creates recruitment challenges. Additionally, asking children about family conflict raises questions about whether they can truly voluntarily participate or may feel coerced. Studying topics that might reveal abuse creates mandatory reporting obligations that complicate research ethics.
Studies Needing Multi-Year Datasets
AI suggests longitudinal or time-dependent studies without understanding timeline constraints: AI suggestion: “The long-term effects of a new teacher mentoring program on five-year retention rates” Feasibility problem: If the program started this year, you need five years of data before you can analyze retention effects. Your dissertation timeline is typically 2-3 years maximum from proposal approval to defense. You can’t wait five years for data to exist. AI suggestion: “How organizational culture changes during CEO succession” Feasibility problem: You need baseline data from before succession, then post-succession data over time to assess change. You can’t control when CEO successions occur. Waiting for appropriate succession events, then collecting multi-wave data, easily exceeds dissertation timelines. AI suggestion: “Evaluating the impact of a new state education policy on student achievement” Feasibility problem: New policies need time to be implemented, time for effects to manifest, and time for data to be collected and made available. Even if the policy was enacted recently, comprehensive impact data may not exist for 3-5 years. You can’t complete a dissertation waiting for data that doesn’t exist yet.
Politically Sensitive Topics Producing Unusable Data
AI suggests topics on controversial issues without understanding that social desirability bias will make data invalid: AI suggestion: “Interviewing affluent homeowners about their attitudes toward affordable housing development in their neighborhoods” Feasibility problem: People know the socially acceptable answer is supporting affordable housing. Even if they privately oppose it, they’re unlikely to admit opposition to a researcher recording their responses. You’ll get politically correct responses that don’t reflect actual attitudes, rendering your data invalid for understanding genuine perspectives. AI suggestion: “Exploring racial attitudes and implicit biases among teachers” Feasibility problem: Teachers know discussing racial bias is professionally risky. They won’t honestly report biased attitudes in interviews, especially recorded ones. Social desirability bias makes the data worthless for understanding actual attitudes versus stated attitudes. AI suggestion: “Understanding how managers make hiring decisions when they encounter candidates from underrepresented groups” Feasibility problem: Managers won’t admit to discriminatory hiring practices, which are illegal. Self-report data on hiring decisions involving protected characteristics will be heavily filtered through social desirability, preventing you from getting valid data on actual decision-making processes.
The Human Feasibility Checks AI Cannot Perform
When experienced dissertation advisors review proposed topics, they systematically assess feasibility using criteria AI cannot apply.
Check 1: Is Data Available Now?
We assess whether data currently exists or can be collected within dissertation timelines: Existing data: If you’re using secondary data, does it exist? Do you have access? What permissions are required? How long does access take? Primary data collection: Can you realistically recruit your sample size within 3-6 months? What’s the typical response rate for your population and method? Are there seasonal factors affecting recruitment? Temporal requirements: If your research examines recent policies or events, has enough time passed for relevant data to exist? Do you need baseline data that doesn’t exist yet? AI can’t check whether data exists or is accessible—it just suggests what would be ideal to study.
Check 2: Can the Population Be Safely Accessed?
We evaluate whether you can ethically and practically reach your target population: Vulnerable populations: If your population includes minors, trauma survivors, prisoners, undocumented individuals, or others protected by IRB, what special protections are needed? Can you realistically provide them? Access barriers: Do you need institutional permission (schools, hospitals, prisons)? How long do those approval processes take? What’s the likelihood of receiving permission? Recruitment feasibility: Can you reach this population? Do you have contacts or entrée? What recruitment strategies are available? What response rates can you expect? AI doesn’t know access requirements for different populations or settings.
Check 3: Will the IRB Approve?
We predict IRB concerns and assess approval likelihood: Risk level: Does your study create physical, psychological, social, legal, or economic risks for participants? Are risks minimal or more than minimal? Vulnerable populations: If studying protected populations, can you demonstrate sufficient protections and justify why the research benefits outweigh risks? Informed consent: Can participants truly consent voluntarily, or are there coercion concerns? Do you need parental consent, guardian consent, or institutional consent in addition to individual consent? Approval timeline: Will this be expedited review (2-4 weeks) or full board review (2-3 months)? Are multiple revision rounds likely? AI has no knowledge of IRB processes or requirements. It suggests topics without considering whether they’re approvable.
Check 4: Will Responses Be Socially Biased?
We assess whether you can get valid data or if social desirability will corrupt responses: Sensitive topics: Are you asking about illegal behaviors, socially undesirable attitudes, or controversial positions? Will participants answer honestly? Professional risk: Does honest answering create professional or career risks? Teachers won’t honestly report pedagogical failures. Managers won’t admit discriminatory practices. Social acceptability: Is there clearly a “right” answer participants know they should give, even if it’s not their true attitude or behavior? Alternative measurement: If self-report will be biased, are there alternative approaches (observations, archival data, indirect measures) that could yield more valid data? AI doesn’t understand measurement validity threats from social desirability.
Real Examples of AI-Suggested Infeasible Topics
Let me show you actual examples of topics students developed with AI help that turned out to be infeasible.
Example 1: The Vulnerable Population Problem
AI-suggested topic: “Phenomenological study of how incarcerated mothers maintain relationships with their children” Why it seemed good: Addresses important social issue, understudied population, clear problem focus Feasibility problems:
- Prisoners are a protected class requiring special IRB protocols
- Prison access requires corrections department approval (months-long process, often denied to student researchers)
- Researching about children adds vulnerability concerns
- Timeline for securing approvals would exceed typical dissertation schedule
- Many IRBs won’t approve prison research for student dissertations due to complexity and risk
Example 2: The Timing Problem
AI-suggested topic: “Impact of a new hospital electronic health record system on medical errors one year post-implementation” Why it seemed good: Clear problem (medical errors), clear intervention (EHR system), clear timeframe (one year) Feasibility problems:- Hospital implemented the EHR three months ago; one-year post-implementation data doesn’t exist yet
- Student needs to defend within 18 months
- Can’t wait 9+ months for post-implementation period to complete, then additional months for error data to be compiled and made available
- Timeline doesn’t work
Example 3: The Social Desirability Problem
AI-suggested topic: “How school principals describe their decision-making about disciplining students from different racial backgrounds” Why it seemed good: Addresses equity concerns, asks about actual decisions, focuses on leaders Feasibility problems:
- Principals know admitting racial bias in discipline decisions is professionally and legally dangerous
- They’ll provide socially acceptable responses about treating everyone equally, regardless of actual practice
- Interview data will be filtered through extreme social desirability bias
- Research won’t capture actual decision-making patterns
Example 4: The Access Problem
AI-suggested topic: “Observational study of how emergency department teams communicate during critical trauma cases” Why it seemed good: Clear importance (patient safety), direct observation (avoiding self-report bias), specific setting Feasibility problems:- Hospitals won’t grant access to observe actual trauma cases (patient privacy, liability, HIPAA)
- IRB questions patient consent when they’re unconscious or in crisis
- Researchers cannot be present during life-and-death situations
- Sampling would be impossible (when do traumas occur? How many observations needed?)
- Staff won’t behave naturally being observed during high-stakes emergencies
Get Expert Feasibility Assessment
Don’t waste months on topics that can’t be completed. Get feasibility assessment from scholars who understand constraints you’re working within.
Our Feasibility Review Process
We systematically assess: IRB probability: Will this likely be approved? What modifications might be needed? Are there insurmountable ethical obstacles? Data accessibility: Can you access needed data within dissertation timelines? What permissions are required? How long do approvals typically take? Measurement validity: Can you get valid data, or will social desirability, access limitations, or other factors compromise data quality? Timeline realism: Can data collection and analysis be completed within 12-18 months? What could delay you? Resource requirements: What would this cost? What specialized skills or assistance would you need? Get a feasibility review before investing months in topics that won’t work.IRB Preparation Support
If your topic has feasibility concerns but might be viable with proper design: Protocol development: Designing studies that maximize approval likelihood Risk mitigation: Identifying participant protections that address IRB concerns Alternative approaches: Finding methodological modifications that reduce risk while preserving research value Documentation preparation: Helping you prepare IRB applications that anticipate and address reviewer concerns
Complete Dissertation Planning
Feasibility is one component of comprehensive planning: Get dissertation help that ensures topics are not only original and problem-driven but also completable within your program’s constraints.
The Bottom Line: Feasibility Requires Field Knowledge
AI suggests topics based on what sounds academically interesting. It has no ability to assess whether topics can actually be researched within IRB, timeline, access, and measurement constraints doctoral students face. Only experienced advisors can:
- Predict IRB concerns and approval likelihood
- Assess data accessibility within realistic timelines
- Recognize when social desirability will corrupt data validity
- Identify when populations or settings are practically inaccessible
- Suggest feasible modifications to preserve research value