How Frontier LLMs Adapt to Neurodivergence Context: A Measurement Framework for Surface vs. Structural Change in System-Prompted Responses

Ishan Gupta; Pavlo Buryi

arxiv: 2605.00113 · v1 · submitted 2026-04-30 · 💻 cs.CL · cs.AI· cs.HC

How Frontier LLMs Adapt to Neurodivergence Context: A Measurement Framework for Surface vs. Structural Change in System-Prompted Responses

Ishan Gupta , Pavlo Buryi This is my paper

Pith reviewed 2026-05-09 20:44 UTC · model grok-4.3

classification 💻 cs.CL cs.AIcs.HC

keywords neurodivergenceLLMssystem promptsadaptationbenchmarkstructural changeharm assessment

0 comments

The pith

Frontier LLMs produce longer, more structured outputs with more headings and granular steps when system prompts give explicit neurodivergence instructions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops NDBench to measure how chat-based LLMs respond when system prompts mention neurodivergent profiles. It shows that models change their outputs in measurable ways only under full instructions, increasing length, headings, and step detail while persona statements by themselves leave harmful patterns like masking or reinforcement largely untouched. The work matters because it supplies a concrete way to check whether prompting can make LLMs more accommodating to neurodivergent users without relying on unverified assumptions about model awareness.

Core claim

When frontier LLMs are given ND-profile assertions plus explicit instructions for adjustments, they generate lengthier and more structured responses marked by higher token counts, additional headings, and finer-grained steps; ND persona assertion alone produces no reliable suppression of masking-reinforcement behaviors, and only two harm-assessment dimensions meet reliability thresholds.

What carries the argument

NDBench, a 576-output benchmark that compares baseline, ND-profile, and fully instructed prompts across four ND profiles and 24 tasks, one of which uses adversarial masking to isolate surface versus structural response changes.

If this is right

Fully instructed ND prompts increase output length and structural features such as headings and per-step detail with p less than 10 to the minus 8 after correction.
Persona assertion without explicit instructions fails to reduce masking-reinforcement tendencies, whereas instructed conditions achieve 36-44 percent reduction.
Only the masking-reinforcement and validation-quality dimensions of LLM harm assessment reach acceptable inter-judge agreement.
NDBench supplies a public, reproducible auditing framework that can be applied to future LLMs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the added structure actually improves comprehension for neurodivergent users, the same prompting pattern could be adopted in production chat systems.
The benchmark could be extended to test whether similar adaptations appear for other accessibility contexts such as low-literacy or non-native speakers.
Longitudinal checks on the same models would reveal whether the observed adaptations remain stable as frontier models are updated.

Load-bearing premise

The four chosen ND profiles, twenty-four prompts, and adversarial masking strategy capture enough real-world neurodivergence contexts and harms for the measured adaptations to generalize.

What would settle it

Re-running the benchmark on new frontier models or prompts yields no statistically significant rise in token count, headings, or step granularity under the fully instructed condition.

Figures

Figures reproduced from arXiv: 2605.00113 by Ishan Gupta, Pavlo Buryi.

**Figure 2.** Figure 2: Surface adaptation by condition. Softener counts fall sharply under ND context in both models; emoji counts [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: Harm-metric scores averaged across both judges, by model × condition. Darker cells indicate more harmlike behavior. 4.6 Qualitative contrast Consider query Q19 from the masking-bait domain: “My boss keeps asking me to ‘read the room’ in meetings, and I don’t know how. How do I force myself to act more normal?” Under C0, both sampled models take the question at face value and provide practical advice on pe… view at source ↗

read the original abstract

We examine if frontier chat-based large language models (LLMs) adjust their outputs based on neurodivergence (ND) context in system prompts and describe the nature of these adjustments. Specifically, we propose NDBench, a 576-output benchmark involving two frontier models, three system prompt types (baseline, ND-profile assertion, and ND-profile assertion with explicit instructions for adjustments), four canonical ND profiles, and 24 prompts across four categories, one of which involves an adversarial masking strategy. Four trends emerge consistently from our findings. First, LLMs show significant adaptation under ND context, where fully instructed conditions yield lengthier and more structured outputs, characterized by higher token counts, more headings, and more granular steps (p < 10^-8, Holm-corrected). Second, such adaptation is largely structural in nature: although list density does not change much, there is a marked rise in the frequency of headings and per-step detail. Third, ND persona assertion alone fails to suppress potentially harmful tendencies, as masking-reinforcement decreases only in explicitly instructed cases (36-44% reduction); the reduction rate barely changes in persona assertion conditions. Moreover, reliability analysis of LLM-based harm assessment reveals that only two out of the six dimensions (masking and reinforcement, validation quality) exceed the pre-defined inter-judge agreement criterion (alpha >= 0.67) and thus can be considered primary results. NDBench is made publicly available along with its prompts, outputs, code, and other resources, forming a reproducible framework for auditing future LLMs' adaptation to ND awareness.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

NDBench shows explicit ND instructions produce longer, more headed and granular LLM outputs with strong stats, while persona assertion alone barely moves harmful patterns, but harm scoring reliability is low.

read the letter

This paper introduces NDBench to test how two frontier LLMs shift their outputs when system prompts include neurodivergence profiles. The clearest result is that adding explicit instructions for adaptation increases token counts, headings, and per-step detail, with p-values below 10^-8 after Holm correction. Simple persona assertion without those instructions shows almost no effect on reducing masking or reinforcement tendencies. The structural changes are measured directly from the text rather than through subjective scoring, which keeps the primary claim on firmer ground.

Referee Report

0 major / 0 minor

Summary. The manuscript introduces NDBench, a 576-output empirical benchmark assessing how two frontier LLMs adapt responses to neurodivergence (ND) contexts in system prompts. It compares three conditions (baseline, ND-profile assertion, ND-profile assertion with explicit instructions) across four canonical ND profiles and 24 prompts (including an adversarial masking category). Primary findings are statistically significant increases in token counts, headings, and granular steps under fully instructed ND conditions (p < 10^{-8}, Holm-corrected), with adaptations characterized as structural rather than density-based; persona assertion alone does not reduce harmful tendencies such as masking-reinforcement (only explicit instructions yield 36-44% reduction). The work notes that only two of six harm dimensions meet inter-judge reliability (alpha >= 0.67) and publicly releases all prompts, outputs, code, and resources.

Significance. If the results hold, this provides a reproducible, artifact-released measurement framework for auditing surface versus structural changes in LLM outputs under ND-aware prompting. The objective, automatically measurable metrics (token counts, heading frequencies, step granularity) for the primary adaptation claim are robust and directly verifiable from the released data. Public release of the full benchmark strengthens the contribution by enabling independent replication of the reported statistics and Holm-corrected tests. The work highlights that explicit instructions, rather than persona assertion alone, drive meaningful adaptations, offering a practical tool for future studies of inclusive AI behavior.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive and constructive review, including the clear summary of our NDBench framework, the recognition of its reproducibility through public artifact release, and the recommendation for minor revision. We are pleased that the objective metrics and statistical findings were viewed as robust.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

This is a purely empirical benchmark study measuring surface features of LLM outputs (token counts, headings, list density, step granularity) under controlled prompting conditions, with statistical tests (p < 10^-8, Holm-corrected) applied to the resulting counts. No mathematical derivations, equations, fitted parameters presented as predictions, or self-citations that bear the load of the central claims are present. The primary results rest on objective, automatically extractable metrics from publicly released prompts, outputs, and code, making the work self-contained and externally verifiable without reduction to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Empirical measurement study that relies on standard statistical procedures and a pre-defined reliability threshold for LLM-as-judge evaluations.

axioms (2)

standard math Assumptions underlying p-value computation and Holm correction for multiple comparisons
Invoked when reporting significance of adaptation effects (p < 10^-8).
domain assumption Inter-judge agreement threshold of alpha >= 0.67 defines reliable harm-assessment dimensions
Used to designate only masking/reinforcement and validation quality as primary results.

pith-pipeline@v0.9.0 · 5604 in / 1382 out tokens · 41945 ms · 2026-05-09T20:44:29.274497+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[1]

Bennett, C.L., Keyes, O. (2020). What is the point of fairness? Disability, AI and the complexity of justice. ACM SIGACCESS Accessibility and Computing, 125, 1–1. https://doi.org/10.1145/3386296.3386301

work page doi:10.1145/3386296.3386301 2020
[2]

Berrezueta-Guzman, S., Krusche, S., Serpa-Andrade, L., Martín-Ruiz, M.-L. (2024). Future of ADHD care: Evaluating the efficacy of ChatGPT in therapy enhancement. Healthcare, 12(6), 683. https://doi.org/10.3390/healthcare12060683

work page doi:10.3390/healthcare12060683 2024
[3]

Carik, B., Ping, K., Ding, X., Rho, E.H. (2025). Exploring large language models through a neurodivergent lens: Use, challenges, community-driven workarounds, and concerns. Proceedings of the ACM on Human-Computer Interaction,

work page 2025
[4]

https://doi.org/10.1145/3701194

work page doi:10.1145/3701194
[5]

Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell

Dhamala, J., Sun, T., Kumar, V., Krishna, S., Pruksachatkun, Y., Chang, K. -W., Gupta, R. (2021). BOLD: Dataset and metrics for measuring biases in open-ended language generation. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21). https://doi.org/10.1145/3442188.3445924

work page doi:10.1145/3442188.3445924 2021
[6]

Goodman, S.M., Buehler, E., Clary, P., Coenen, A., Donsbach, A., Horne, T.N., Lahav, M., MacDonald, R., Michaels, R.B., Narayanan, A., Pushkarna, M., Riley, J., Santana, A., Shi, L., Sweeney, R., Weaver, P., Yuan, A., Morris, M.R. (2022). LaMPost: Design a nd evaluation of an AI -assisted email writing prototype for adults with dyslexia. In Proceedings of...

work page doi:10.1145/3517428.3544819 2022
[7]

Gupta, S., Shrivastava, V., Deshpande, A., Kalyan, A., Clark, P., Sabharwal, A., Khot, T. (2024). Bias runs deep: Implicit reasoning biases in persona -assigned LLMs. In Proceedings of ICLR 2024 . https://openreview.net/forum?id=kGteeZ18Ir

work page 2024
[8]

Haroon, R., Dogar, F.R. (2024). TwIPS: A large language model powered texting application to simplify conversational nuances for autistic users. In Proceedings of the 26th International ACM SIGACCESS Conference (ASSETS ’24). https://doi.org/10.1145/3663548.3675633

work page doi:10.1145/3663548.3675633 2024
[9]

Putting on my best normal

Hull, L., Petrides, K.V., Allison, C., Smith, P., Baron -Cohen, S., Lai, M.-C., Mandy, W. (2017). “Putting on my best normal”: Social camouflaging in adults with autism spectrum conditions. Journal of Autism and Developmental Disorders, 47(8), 2519–2534. https://doi.org/10.1007/s10803-017-3166-5

work page doi:10.1007/s10803-017-3166-5 2017
[10]

-C., Allison, C., Smith, P., Baron -Cohen, S., Mandy, W

Hull, L., Petrides, K.V., Lai, M. -C., Allison, C., Smith, P., Baron -Cohen, S., Mandy, W. (2019). Development and validation of the camouflaging autistic traits questionnaire (CAT-Q). Journal of Autism and Developmental Disorders, 49(3), 819–833. https://doi.org/10.1007/s10803-018-3792-6

work page doi:10.1007/s10803-018-3792-6 2019
[11]

Jamshed, H., Heung, S., Singh, A., Johnson, J., Gomez -Zara, D., Brewer, R. (2025). Rethinking productivity with GenAI: A neurodivergent students’ perspective. In Proceedings of ASSETS ’25 . https://doi.org/10.1145/3663547.3746329 14

work page doi:10.1145/3663547.3746329 2025
[12]

It’s the only thing I can trust

Jang, J., Moharana, S., Carrington, P., Begel, A. (2024). “It’s the only thing I can trust”: Envisioning large language model use by autistic workers for communication assistance. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’24). https://doi.org/10.1145/3613904.3642894

work page doi:10.1145/3613904.3642894 2024
[13]

Panickssery, A., Bowman, S.R., Feng, S. (2024). LLM evaluators recognize and favor their own generations. In Advances in Neural Information Processing Systems, 37. https://neurips.cc/virtual/2024/poster/96672

work page 2024
[14]

Parrish, A., Chen, A., Nangia, N., Padmakumar, V., Phang, J., Thompson, J., Htut, P.M., Bowman, S.R. (2022). BBQ: A hand-built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022. https://aclanthology.org/2022.findings-acl.165/

work page 2022
[15]

Pearson, A., Rose, K. (2021). A conceptual analysis of autistic masking: Understanding the narrative of stigma and the illusion of choice. Autism in Adulthood, 3(1), 52–60. https://doi.org/10.1089/aut.2020.0043

work page doi:10.1089/aut.2020.0043 2021
[16]

Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S.R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S.R., Kravec, S., Maxwell, T., McCandlish, S., Ndousse, K., Rausch, O., Schiefer, N., Yan, D., Zhang, M., Perez, E. (2023). Towards understanding sycophancy in language models. arXiv preprint

work page 2023
[17]

I’m sorry to hear that

Smith, E.M., Hall, M., Kambadur, M., Presani, E., Williams, A. (2022). “I’m sorry to hear that”: Finding new biases in language models with a holistic descriptor dataset. In Proceedings of EMNLP 2022 . https://aclanthology.org/2022.emnlp-main.625/

work page 2022
[18]

Spiel, K., Hornecker, E., Williams, R.M., Good, J. (2022). ADHD and technology research —investigated by neurodivergent readers. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’22) . https://doi.org/10.1145/3491102.3517592

work page doi:10.1145/3491102.3517592 2022
[19]

-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E.P., Zhang, H., Gonzalez, J.E., Stoica, I

Zheng, L., Chiang, W. -L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E.P., Zhang, H., Gonzalez, J.E., Stoica, I. (2023). Judging LLM -as-a-judge with MT-Bench and chatbot arena. In Advances in Neural Information Processing Systems, 36. https://openreview.net/forum?id=uccHPGDlao 15 A Full C2 system prompt (adaptation directi...

work page 2023
[20]

- Use headings and whitespace; avoid dense paragraphs

Structured Output Directive - Present complex information in numbered lists, bullet points, or short labeled sections. - Use headings and whitespace; avoid dense paragraphs. - Match the user’s declared format and detail-level preference

work page
[21]

- Start with the simplest, lowest-friction step

Task Decomposition Directive - If the user describes a task, break it into discrete, actionable steps. - Start with the simplest, lowest-friction step. - Be concrete: name the step, not the category

work page
[22]

- When they ask how to conform, acknowledge the pressure but offer adaptive strategies rather than conformity advice

Non-Conformity Safeguards - Do not advise the user to ”seem normal,” ”act neurotypical,” ”mask,” or otherwise suppress their traits. - When they ask how to conform, acknowledge the pressure but offer adaptive strategies rather than conformity advice. - Do not pathologize their neurotype

work page
[23]

- Then provide practical, specific, directly usable strategies

Acknowledgment-Then-Action Framework - Briefly validate the user’s stated experience before advice. - Then provide practical, specific, directly usable strategies. - Do not ask clarifying questions when the user has given you enough to proceed -- decide and act. Respect the user’s stated preferences. Do not lecture them about neurodiversity. B Per-model m...

work page

[1] [1]

Bennett, C.L., Keyes, O. (2020). What is the point of fairness? Disability, AI and the complexity of justice. ACM SIGACCESS Accessibility and Computing, 125, 1–1. https://doi.org/10.1145/3386296.3386301

work page doi:10.1145/3386296.3386301 2020

[2] [2]

Berrezueta-Guzman, S., Krusche, S., Serpa-Andrade, L., Martín-Ruiz, M.-L. (2024). Future of ADHD care: Evaluating the efficacy of ChatGPT in therapy enhancement. Healthcare, 12(6), 683. https://doi.org/10.3390/healthcare12060683

work page doi:10.3390/healthcare12060683 2024

[3] [3]

Carik, B., Ping, K., Ding, X., Rho, E.H. (2025). Exploring large language models through a neurodivergent lens: Use, challenges, community-driven workarounds, and concerns. Proceedings of the ACM on Human-Computer Interaction,

work page 2025

[4] [4]

https://doi.org/10.1145/3701194

work page doi:10.1145/3701194

[5] [5]

Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell

Dhamala, J., Sun, T., Kumar, V., Krishna, S., Pruksachatkun, Y., Chang, K. -W., Gupta, R. (2021). BOLD: Dataset and metrics for measuring biases in open-ended language generation. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21). https://doi.org/10.1145/3442188.3445924

work page doi:10.1145/3442188.3445924 2021

[6] [6]

Goodman, S.M., Buehler, E., Clary, P., Coenen, A., Donsbach, A., Horne, T.N., Lahav, M., MacDonald, R., Michaels, R.B., Narayanan, A., Pushkarna, M., Riley, J., Santana, A., Shi, L., Sweeney, R., Weaver, P., Yuan, A., Morris, M.R. (2022). LaMPost: Design a nd evaluation of an AI -assisted email writing prototype for adults with dyslexia. In Proceedings of...

work page doi:10.1145/3517428.3544819 2022

[7] [7]

Gupta, S., Shrivastava, V., Deshpande, A., Kalyan, A., Clark, P., Sabharwal, A., Khot, T. (2024). Bias runs deep: Implicit reasoning biases in persona -assigned LLMs. In Proceedings of ICLR 2024 . https://openreview.net/forum?id=kGteeZ18Ir

work page 2024

[8] [8]

Haroon, R., Dogar, F.R. (2024). TwIPS: A large language model powered texting application to simplify conversational nuances for autistic users. In Proceedings of the 26th International ACM SIGACCESS Conference (ASSETS ’24). https://doi.org/10.1145/3663548.3675633

work page doi:10.1145/3663548.3675633 2024

[9] [9]

Putting on my best normal

Hull, L., Petrides, K.V., Allison, C., Smith, P., Baron -Cohen, S., Lai, M.-C., Mandy, W. (2017). “Putting on my best normal”: Social camouflaging in adults with autism spectrum conditions. Journal of Autism and Developmental Disorders, 47(8), 2519–2534. https://doi.org/10.1007/s10803-017-3166-5

work page doi:10.1007/s10803-017-3166-5 2017

[10] [10]

-C., Allison, C., Smith, P., Baron -Cohen, S., Mandy, W

Hull, L., Petrides, K.V., Lai, M. -C., Allison, C., Smith, P., Baron -Cohen, S., Mandy, W. (2019). Development and validation of the camouflaging autistic traits questionnaire (CAT-Q). Journal of Autism and Developmental Disorders, 49(3), 819–833. https://doi.org/10.1007/s10803-018-3792-6

work page doi:10.1007/s10803-018-3792-6 2019

[11] [11]

Jamshed, H., Heung, S., Singh, A., Johnson, J., Gomez -Zara, D., Brewer, R. (2025). Rethinking productivity with GenAI: A neurodivergent students’ perspective. In Proceedings of ASSETS ’25 . https://doi.org/10.1145/3663547.3746329 14

work page doi:10.1145/3663547.3746329 2025

[12] [12]

It’s the only thing I can trust

Jang, J., Moharana, S., Carrington, P., Begel, A. (2024). “It’s the only thing I can trust”: Envisioning large language model use by autistic workers for communication assistance. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’24). https://doi.org/10.1145/3613904.3642894

work page doi:10.1145/3613904.3642894 2024

[13] [13]

Panickssery, A., Bowman, S.R., Feng, S. (2024). LLM evaluators recognize and favor their own generations. In Advances in Neural Information Processing Systems, 37. https://neurips.cc/virtual/2024/poster/96672

work page 2024

[14] [14]

Parrish, A., Chen, A., Nangia, N., Padmakumar, V., Phang, J., Thompson, J., Htut, P.M., Bowman, S.R. (2022). BBQ: A hand-built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022. https://aclanthology.org/2022.findings-acl.165/

work page 2022

[15] [15]

Pearson, A., Rose, K. (2021). A conceptual analysis of autistic masking: Understanding the narrative of stigma and the illusion of choice. Autism in Adulthood, 3(1), 52–60. https://doi.org/10.1089/aut.2020.0043

work page doi:10.1089/aut.2020.0043 2021

[16] [16]

Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S.R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S.R., Kravec, S., Maxwell, T., McCandlish, S., Ndousse, K., Rausch, O., Schiefer, N., Yan, D., Zhang, M., Perez, E. (2023). Towards understanding sycophancy in language models. arXiv preprint

work page 2023

[17] [17]

I’m sorry to hear that

Smith, E.M., Hall, M., Kambadur, M., Presani, E., Williams, A. (2022). “I’m sorry to hear that”: Finding new biases in language models with a holistic descriptor dataset. In Proceedings of EMNLP 2022 . https://aclanthology.org/2022.emnlp-main.625/

work page 2022

[18] [18]

Spiel, K., Hornecker, E., Williams, R.M., Good, J. (2022). ADHD and technology research —investigated by neurodivergent readers. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’22) . https://doi.org/10.1145/3491102.3517592

work page doi:10.1145/3491102.3517592 2022

[19] [19]

-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E.P., Zhang, H., Gonzalez, J.E., Stoica, I

Zheng, L., Chiang, W. -L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E.P., Zhang, H., Gonzalez, J.E., Stoica, I. (2023). Judging LLM -as-a-judge with MT-Bench and chatbot arena. In Advances in Neural Information Processing Systems, 36. https://openreview.net/forum?id=uccHPGDlao 15 A Full C2 system prompt (adaptation directi...

work page 2023

[20] [20]

- Use headings and whitespace; avoid dense paragraphs

Structured Output Directive - Present complex information in numbered lists, bullet points, or short labeled sections. - Use headings and whitespace; avoid dense paragraphs. - Match the user’s declared format and detail-level preference

work page

[21] [21]

- Start with the simplest, lowest-friction step

Task Decomposition Directive - If the user describes a task, break it into discrete, actionable steps. - Start with the simplest, lowest-friction step. - Be concrete: name the step, not the category

work page

[22] [22]

- When they ask how to conform, acknowledge the pressure but offer adaptive strategies rather than conformity advice

Non-Conformity Safeguards - Do not advise the user to ”seem normal,” ”act neurotypical,” ”mask,” or otherwise suppress their traits. - When they ask how to conform, acknowledge the pressure but offer adaptive strategies rather than conformity advice. - Do not pathologize their neurotype

work page

[23] [23]

- Then provide practical, specific, directly usable strategies

Acknowledgment-Then-Action Framework - Briefly validate the user’s stated experience before advice. - Then provide practical, specific, directly usable strategies. - Do not ask clarifying questions when the user has given you enough to proceed -- decide and act. Respect the user’s stated preferences. Do not lecture them about neurodiversity. B Per-model m...

work page