Authorship, Accuracy, & Accountability: The Ethics of AI in Clinical Psychological Reports

By Adam Lockwood, PhD, NCSP

Summary: AI-assisted report writing can offer meaningful support for clinicians, but it also raises practical ethical questions about who is responsible for the final document. In this blog, Adam Lockwood, PhD, NCSP, explores how authorship, accuracy, privacy, disclosure, and accountability remain firmly in the clinician’s hands, even when AI helps draft or organize report content.

Once a clinician begins using artificial intelligence in report writing, the ethical question changes. Before adoption, the central issue is whether a tool is appropriate for the task, the data, the workflow, and the clinician's oversight. After adoption, the question becomes more immediate: what does responsible use look like when AI-generated language becomes part of a clinical report?

That question matters because psychological reports are not ordinary documents. They communicate findings, synthesize data, support diagnoses, guide treatment, and may influence decisions made by clients, physicians, attorneys, insurers, and other stakeholders. A report may shape how a person understands themselves and how systems respond to their needs.

AI-assisted report writing cannot be treated as simple text generation; it is part of clinical practice. The clinician's ethical obligations do not lessen because a tool helped draft the language. AI can help with efficiency, organization, readability, and drafting, but the clinician remains responsible for what the report says, what it omits, how it interprets the data, and whether the final document is accurate, individualized, and clinically defensible. That balance is central to responsible integration of AI in psychological work (Farmer et al., 2025).

The clinician is still the author

One of the most important questions in AI-assisted report writing is authorship. If AI drafts a paragraph, generates recommendations, or helps organize results, who "wrote" the report?

Professionally, the answer is straightforward: the clinician did. Psychological reports have long involved tools, including templates, scoring programs, dictation systems, and clerical support. The clinician who signs the report is the professional author in the meaningful sense, and that signature communicates that the clinician reviewed the information, interpreted the findings, and stands behind the conclusions.

A clinician should not sign a report simply because the language sounds plausible or polished. The clinician should sign only what they have reviewed, understood, revised as needed, and can defend, including the data summarized, the interpretations offered, the diagnoses assigned, and the recommendations provided.

This distinction matters because AI can generate draft language that looks more finished than it is. A rough human draft often reveals its own incompleteness. AI-generated text may appear organized, confident, and professional even when it contains subtle errors or unsupported conclusions. The clinician's task is not merely to proofread. It is to evaluate.

Accuracy is more than factual correctness

Accuracy in AI-assisted reporting involves more than checking whether names, dates, and scores are correct. In psychological assessment, accuracy also includes whether interpretations follow logically from the data, whether limitations are acknowledged, whether alternative explanations were considered, and whether conclusions are stated with appropriate confidence.

AI may summarize a score pattern correctly but overstate its diagnostic meaning. It may generate recommendations that are generally reasonable but poorly matched to the client's context. It may describe a measure accurately in general terms while failing to account for validity concerns, language background, culture, medical history, effort, or referral question. The result may be text that sounds coherent while weighting one data source too heavily and another too lightly.

Psychological reports require integration across multiple data sources, and AI may accurately summarize individual pieces while failing to integrate them appropriately. The value of a report lies not in whether each sentence is correct but in whether the overall interpretation reflects the pattern of data, resolves inconsistencies, and supports clinically defensible conclusions.

Review must therefore be substantive: verify that the report accurately reflects source data, that interpretations are supported, that recommendations are individualized, and that conclusions do not imply more certainty than the evidence permits. Research comparing AI-generated and human-written psychological reports suggests AI may be useful in drafting certain sections while still requiring clinical judgment throughout (Lockwood et al., 2025). 

Watch for overgeneralization and clinical imprecision

One of the most common risks in AI-assisted clinical writing is not dramatic hallucination. It is subtle overgeneralization. AI may produce language that is broadly true but not specifically true for this client, recommend common interventions without considering feasibility, or describe patterns associated with a diagnosis without adequately explaining why those patterns apply in the present case.

This is especially important in complex or ambiguous presentations. Symptoms may reflect multiple possible explanations, and test results may be influenced by attention, motivation, anxiety, linguistic background, or other factors. In these contexts, generic language can be misleading even when it is not technically false.

Clinicians should review AI-assisted reports for specificity: Does the language connect to the referral question? Does it reflect the client's history and presentation? Does it distinguish between data, inference, and diagnosis? Are limitations clearly stated? Are recommendations realistic for this person, in this setting?

A useful report is not just well written. It is clinically precise.

Disclosure and transparency require practical judgment

Disclosure remains one of the most unsettled areas of AI-assisted practice. Many clinicians agree that clients should be informed when AI is used in clinical work, yet actual disclosure practices remain inconsistent (Lockwood et al., 2026). Clinicians may be unsure what level of AI involvement requires disclosure, as using AI for grammar editing may warrant a different approach than using it to draft client-specific report sections.

The broader principle is clear: clients should not be misled about how clinical services are provided, how their information is handled, or who is responsible for the work product. When AI is involved in documentation, report generation, or other client-specific clinical products, clinicians should consider whether that use should be described in informed consent, documentation policies, or direct communication with the client. Transparency does not require making AI the center of the conversation. It does require honesty, clarity, and respect for client autonomy. When in doubt, disclose.

Privacy and consent remain central

If an AI tool will be used with client-specific information, clinicians must ensure it meets applicable privacy, security, and compliance requirements. Depending on the setting, this typically includes HIPAA compliance, a business associate agreement, organizational approval, and clear policies regarding storage, retention, and whether user data can be used to train the model.

Informed consent also deserves attention. Clients may reasonably want to know whether AI tools are used in creating clinical documentation, especially when sensitive information is involved. Consent language should explain the purpose of the tool, what information may be entered, what safeguards are in place, and that the clinician remains responsible for the final report. The goal is not to overwhelm clients with technical details. The goal is to preserve trust.

Accountability does not transfer to the tool

When a report contains an error, the responsibility does not belong to the software. It belongs to the clinician who signed the report. AI may assist with drafting, but it cannot be accountable in the professional sense: it cannot respond to a board complaint, testify about its judgment, or revise its understanding after meeting the client.

Accountability means clinicians must have a defensible process for using AI, including appropriate tool selection, privacy safeguards, thorough review of all AI-assisted content, verification of data and interpretations, and revision of any language that is inaccurate, unsupported, or insufficiently individualized. Organizations using AI-assisted reporting tools should also establish workflow-level accountability: defining who can use the tool, what training is required, and how output is reviewed and quality monitored over time. These questions are not barriers to adoption. They are part of responsible implementation.

What ethical AI-assisted report writing looks like

In practice, ethical AI-assisted report writing is less about a single rule and more about a disciplined workflow. The clinician determines which parts of the report are appropriate for AI assistance, uses tools that meet privacy and security requirements, and reviews output carefully against source data.

AI-generated language should be treated as draft material requiring careful review, not as a finished clinical product. Clinician-in-the-loop approaches are especially important because they keep professional judgment central: AI may support the work, but the clinician must actively review, verify, and interpret the output before it informs professional decisions (Lockwood et al., 2026).

The signature still carries the weight

AI will likely become increasingly common in psychological report writing, and thoughtfully designed tools may help clinicians write more clearly and reduce administrative burden. But ethical use requires clarity about what has and has not changed.

The tools have changed. The clinician's responsibility has not.

The clinician who signs the report remains responsible for the accuracy of the data, the validity of the interpretations, the appropriateness of the recommendations, the protection of client information, and the clarity of communication. The ethical future of AI-assisted report writing cannot be one in which clinicians become passive reviewers of machine-generated conclusions. It should be one in which clinicians use well-designed tools to support better communication, more efficient documentation, and more thoughtful care.

Authorship, accuracy, and accountability remain human responsibilities. The clinician's signature still carries the weight.

About the Author

Adam Lockwood, PhD, NCSP is an Associate Professor of School Psychology at Kent State University. His research and training focus on the integration of artificial intelligence into psychological practice, including AI-assisted report writing, ethical and governance frameworks, and AI-supported research methods.

References

Farmer, R. L., Lockwood, A. B., Goforth, A. N., & Thomas, C. (2025). Artificial intelligence in practice: Opportunities, challenges, and ethical considerations. Professional Psychology: Research and Practice, 56(1), 19–32. https://doi.org/10.1037/pro0000595

Lockwood, A. B., Farmer, R. L., Shergill, G., Benson, N. F., & Gilbert, K. (2025). Human vs. machine: Comparing AI-generated and human-written psychological reports. Journal of Psychoeducational Assessment, 43(6), 559–573. https://doi.org/10.1177/07342829251346623.

Lockwood, A. B, Farmer, R. L., & McClintock, J. (2026). Helpful but not trusted: Artificial intelligence adoption and ethical concerns among health service psychologists. PsyArXiv https://osf.io/preprints/psyarxiv/gf9sq_v1