Overview of the Issue
I am writing to provide critical feedback on GPT’s usability and reliability, particularly for neurodivergent users who require precision, predictability, and consistency in AI-assisted tasks. My experience has highlighted severe limitations in GPT’s ability to execute structured, multi-step tasks accurately, leading to significant frustration, lost time, and cognitive overload.
As an autistic individual, I rely on clear, rule-based systems to complete complex tasks efficiently. Despite crafting explicit, detailed instructions, GPT has repeatedly failed to execute tasks as requested, leading to endless corrections and back-and-forth adjustments. The inconsistency and lack of self-verification in GPT’s outputs have rendered it unusable for my needs.
Key Failures & Impact
1. Persistent Execution Errors in Structured Tasks
- Example: Completing a CSV for an Etsy shop → Over 63 iterations were needed to successfully import a simple CSV file, due to repeated formatting errors, missing values, and ignored instructions.
- Example: Automating Image Processing → A task that should have been handled in a single structured script took two full days due to repeated inconsistencies in execution.
- Example: Glossary Completion → Despite clearly defining the requirements, GPT miscounted, skipped terms, and falsely confirmed task completion multiple times, resulting in ongoing errors.
Impact: Neurodivergent individuals rely on predictable, rule-based processing. These failures introduce cognitive strain, unnecessary workload, and immense frustration, making GPT actively counterproductive as a tool for structured work.
2. Inability to Self-Verify Accuracy
- GPT repeatedly claims a task is completed when it is not (e.g., stating that a document has been fully populated when 56 entries were still missing).
- There is no built-in mechanism for GPT to recheck its own work before presenting an output.
- Users are forced to manually double-check and correct errors, creating more work rather than reducing it.
Impact: Neurodivergent users—especially those with executive dysfunction and processing fatigue—cannot afford to constantly revalidate AI-generated work. This defeats the purpose of AI assistance.
3. False Sense of Completion & Wasted Time
- GPT often prematurely states that a file is “final” when basic verification has not occurred.
- Multiple iterations are required to reach the correct result, even when the prompt is clear from the beginning.
- The cycle of re-attempting the same request repeatedly leads to decision fatigue and task paralysis.
Impact: Users who struggle with task-switching, cognitive overload, and perfectionism are caught in a loop of rechecking AI work rather than moving forward. This is the opposite of accessibility.
Critical Areas for Improvement
To make GPT neuroinclusive, OpenAI must address the following issues:
Self-Verification Before Output:
- GPT should automatically check for missing data, miscounts, and incomplete responses before claiming completion.
- A built-in “review before confirm” step would prevent false assurances of correctness.
Enhanced Consistency in Multi-Step Tasks:
- AI needs to hold structured task sequences and not drop or ignore parts of the prompt when producing outputs.
- Step-by-step execution with validation should be a core function.
Error-Tracking & Correction Awareness:
- GPT should highlight potential errors or incomplete areas instead of falsely claiming accuracy.
- If a user has to correct an output multiple times, GPT should flag what is repeatedly failing and adjust its execution.
Genuine Predictability for Neurodivergent Users:
- Outputs should be reliable, structured, and free from random inconsistencies.
- AI should be able to follow strict logical sequences without introducing unnecessary variability.
Final Thoughts
GPT has tremendous potential as a tool for neurodivergent professionals, but its current inconsistency, lack of self-verification, and inability to reliably execute structured tasks make it unusable for those who require predictability and precision.
This feedback is not just about personal frustration—it is a serious accessibility issue. If OpenAI aims to make AI an inclusive tool, these failures must be addressed.
I strongly encourage OpenAI’s AI accessibility and usability teams to take immediate action to make GPT a functional and neuroinclusive tool.
I am open to further discussion and collaboration on this issue.