What APA’s June 2026 AI Guidance Changes for Personality Reports

What was actually updated in June 2026?

APA published a practical overview for responsible AI use in assessment in June 2026. The document, titled “Responsible Use of AI in Assessment: A Practical Overview,” identifies areas to consider when artificial intelligence is used in psychological or educational assessment. It does not introduce a new personality scale, provide new norms or cutoffs, or report validation results for a named personality instrument. The overview is attributed to APA’s Committee on Psychological Tests and Assessment and is dated June 2026. Its eight topics are transparency and accountability; bias and fairness; privacy and confidentiality; informed consent; competence and training; human oversight; impact on practice; and continuous improvement. Taken together, these topics point to questions about how AI may be involved, what information it uses, how its output is checked, and how problems may be noticed or addressed. They concern responsible practice around assessment. They do not tell a reader how a specific personality questionnaire converts answers into a score, which comparison group supports an interpretation, or whether that interpretation fits a proposed use. The overview itself describes its role as informational rather than as a source of binding professional standards, so its considerations should not be presented as a newly imposed universal rule. A separate date on APA’s “Testing and Assessment” index can easily blur the distinction. The overview’s practical scope is relevant even when a report does not disclose every technical detail: it gives readers a reason to ask how assessment work was handled, without implying that AI necessarily changed the reported result. The index lists the AI overview under reports and says “Last updated: June 2026”; it also directs readers elsewhere for news and revision updates about the Standards for Educational and Psychological Testing (2014). The index does not say that a personality instrument’s norms or interpretation rules were revised. Its date alone cannot establish a change in scoring practice, the evidence behind a report, or the status of any individual result. That distinction is useful when a report presents a confident conclusion. A reader can ask whether AI contributed to assessment or reporting and what safeguards surrounded that contribution. But a favorable answer about process cannot, on its own, show that the report’s score means what its prose says. Nor does the June date establish that an existing score has become wrong. It marks a practical resource relevant to AI in assessment, not a finding about every personality report. Read the update as a prompt to inspect how a report was produced, not as evidence that its scoring rules changed.

Sources: Responsible Use of AI in Assessment: A Practical Overview; Testing and Assessment

How is this resource different from the testing standards?

The June overview is an APA practical resource about responsible AI use in assessment. The Standards for Educational and Psychological Testing are a separate 2014 publication jointly developed by the American Educational Research Association (AERA), the American Psychological Association (APA), and the National Council on Measurement in Education (NCME). AERA describes the Standards as addressing professional and technical issues of test development and use in education, psychology, and employment. The new overview does not announce a new edition or replace the Standards. A reference to “new APA testing guidance” could mean the AI resource, the joint publication, or evidence for a particular instrument; those are different things to check.

Here, “standards” refers to that named joint publication, rather than a loose synonym for advice. The AERA publisher record identifies its 2014 edition and its three sponsoring organizations; it recommends that edition as the authoritative source of testing standards. The record establishes the publication’s identity and broad scope, but it is not the full text and cannot show that any particular personality report meets a specific standard. Nor does the date on the APA testing index establish that the joint Standards were revised: a webpage’s maintenance date is metadata about that page. The June 2026 item and the 2014 book therefore should not be collapsed into one update. The available APA summary describes its overview as informational and says it does not establish binding professional standards. It is not a new edition of the joint work. The distinction is not a ranking of which source is more useful. A practical overview can make emerging process questions easier to notice, while a standards publication offers a broader reference for professional and technical practice. Neither description answers whether an individual report has evidence for a claim about a reader. That question still depends on the instrument, the score interpretation, and the use being proposed.

Their practical questions differ. The June overview names considerations for responsible practice when AI is involved in assessment. It can help a reader ask where a system entered the process, what role it played, and what oversight or safeguards were considered. The Standards address broader professional and technical questions about developing and using tests. Instrument-specific documentation and validation evidence then address whether a defined interpretation of a defined score is supported for a particular population and purpose. The layers inform one another but cannot substitute for one another: process explanation does not show what a score means, and general standards do not validate each instrument. This distinction does not assign either document a particular legal force for every provider.

Sources: Standards for Educational and Psychological Testing (2014 Edition); Testing and Assessment; Responsible Use of AI in Assessment: A Practical Overview

Where can AI enter a personality report's production?

AI can enter a personality report at several points, and the point matters more than the broad label “AI-assisted.” The American Psychological Association’s Responsible Use of AI in Assessment: A Practical Overview names tool selection, administration or delivery, scoring, interpretation, and reporting as distinct decision junctures. That description is about assessment generally; it does not establish that a particular personality-report provider uses AI, or that every report passes through all these stages. A reader can trace the report as a chain. First, a tool might be chosen to support an assessment, or software might deliver the questions. These uses can affect which instrument is used and how a person encounters it, but they do not necessarily mean the software calculated the score or wrote the conclusion. Ask whether its role was limited to logistics, and whether it changed the assessment or the way a person encountered it. The overview points to selection and delivery as stages for consideration; these are reader questions drawn from that structure, not a quoted APA disclosure form. The next link may be scoring. A system could calculate a score from answers, check or classify response patterns, or summarize them. Those operations differ: a calculation applies scoring rules, pattern recognition flags response features, and a summary describes answers. Ask what input it received, whether it changed the score or merely supported review, and whether a person checked the result against the scoring method. The overview does not say that using AI automatically makes a score wrong. It identifies accuracy, bias, validity, reliability, fairness, and transparency as concerns to address when AI is part of assessment. A further step is interpretation: turning a score or response pattern into a statement about a tendency. A system might suggest language or choose among descriptions, shaping the report even if the underlying number remains unchanged. “The score was calculated automatically” and “the narrative was generated from a score” describe different processes; neither phrase alone tells a reader whether the interpretation is supported. Ask whether it proposed wording or determined it, what information shaped the language, and who owned the final interpretation. Finally, the report may combine those outputs into advice or recommendations. The key distinction is whether a person reviewed the material before it reached the reader and could correct, reject, or override it. In its human-oversight discussion, Responsible Use of AI in Assessment: A Practical Overview describes qualified people reviewing AI outputs, interpreting results, and making final decisions; it also notes that review can help identify errors and account for context. That is the overview’s account of responsible practice, not proof that a particular report received meaningful review. A human name or signature, by itself, does not show what was checked. So when a provider says “AI-assisted,” ask for a plain account of the function: where in the chain AI was used, what information it received, whether its output could change the score or wording, and who could review or correct an error. The answer may reveal a narrow administrative role or a contribution to scoring and interpretation. Either way, it clarifies how the report was made. It does not, by itself, establish what the score means or whether the conclusion fits a proposed use.

Sources: Responsible Use of AI in Assessment: A Practical Overview

What should a reader ask about transparency, consent, privacy, and review?

Ask the provider to describe the system’s role in ordinary language: what task it performed, why it was included, what information it received, and whether its output could change a score or report wording. Then ask who checked that contribution and what happens if someone finds an error. These questions follow the topics in APA’s “Responsible Use of AI in Assessment: A Practical Overview”; they help clarify a particular process, but are not a universal consumer checklist APA says every provider must use.

The questions separate concerns that can otherwise blur together. Transparency is about making the system’s role understandable. “AI-assisted” leaves open whether software scheduled an assessment, helped score responses, suggested an interpretation, or drafted narrative text. Those are different points in the path from answers to conclusions. A useful explanation identifies the point of use and the contribution. Without it, the reader cannot tell what part of the report to scrutinize or what to ask next.

Consent concerns what the person was told before their information was used and whether alternatives were explained. Ask what notice was provided, what participation entails, and whether there was another way to complete the assessment. The APA overview includes informed consent among its considerations, but that does not establish a particular legal entitlement or a universal opt-out route. The practical aim is to understand the choice offered and whether it fits the reason for taking the assessment.

Privacy is a separate question: what information entered the system, how it was handled, and who could access it. Ask whether answers, identifying details, or report text were sent to an AI tool, and what the provider can explain about their handling. APA’s overview names privacy and confidentiality as concerns; it does not establish any named provider’s practices. Do not infer safe handling merely from a polished report or a human signature.

Competence and human oversight ask whether someone responsible can understand the tool’s limits, inspect its contribution, and respond when something appears wrong. Ask who reviewed the output, what they could check, whether they could change it, and how to raise a concern. APA’s summary discusses human review, competence, and training. It does not promise an individual appeal process for every report, so ask what correction path is actually available rather than assuming one exists.

Continuous improvement concerns what happens after a system is introduced: whether its performance and effects are monitored and whether problems lead to changes. A reader may need to ask about this at the provider level. A clear response can identify a safeguard, a gap, or a question the provider cannot answer. Each result helps explain how the report was made; none settles whether the score supports its substantive conclusion. Human review can catch some process or output problems, but a reviewer’s presence does not itself show that evidence supports the interpretation or its proposed use.

Sources: Responsible Use of AI in Assessment: A Practical Overview

Why can a transparent process still leave the score claim uncertain?

Transparency tells you how a report was produced; validity asks whether evidence and reasoning support the meaning assigned to its score and the use proposed. A provider can disclose an automated step, score consistently, and have a person review the output, yet those facts alone do not show that a personality conclusion describes this reader or supports a prediction about work. Process quality and score interpretation are connected, but not interchangeable. The 2023 study “Defining Test-Score Interpretation, Use, and Claims: Delphi Study for the Validity Argument” makes the distinction practical. It treats interpretation as assigning meaning to a score, use as its purpose, and claims as assertions connecting evidence to those interpretations and uses. Researchers used three questionnaire rounds to clarify expert views. Thirteen experts participated; nine completed the first and second rounds, seven the third, and five completed all three. The US university-affiliated panel worked largely in mathematics or STEM education and educational psychology. This limited panel is neither a representative survey of testing professionals nor a study of any personality report. Its contribution is conceptual: it suggests asking what a report’s sentence means and what evidence could support it. Consider an illustration, not a test result. A respondent’s answers are scored; the report describes the score as a tendency to prefer planning; then a reader or employer infers that the person will perform well in a role. Each step adds an inference. Evidence supporting the scoring procedure does not automatically establish that the description captures a tendency across relevant situations. Evidence for that description does not by itself show that the tendency predicts performance in a named job. The move from a pattern to an outcome requires evidence suited to that added claim, the people and setting involved, and the proposed decision. This is why “validated” is incomplete unless the provider identifies the interpretation and use at issue. The 2014 Standards for Educational and Psychological Testing frame validity around the degree to which evidence and theory support score interpretations for proposed uses. The Delphi paper also ties claims to interpretation and use, while noting that scholars do not fully agree about whether consequences belong inside validity itself. Some treat consequences as a separate ethical or use question; others include consequences when judging an interpretation for its purpose. The disagreement affects terminology, but it does not make consequences irrelevant. Readers still need to know what conclusion is drawn and what action it may inform. An earlier conceptual account, “Validity of Test Interpretation and Use,” published by ETS in 1990, describes validity as a matter of degree that later findings can strengthen or challenge. It highlights score meaning, relevance, utility, and the implications of using scores as a basis for action. This is not evidence about AI or a current personality instrument. It reinforces a useful restraint: evidence does not give a report a permanent, all-purpose stamp. A report may support a modest reflective prompt while leaving a stronger claim unsupported. Keep two questions separate when reading an AI disclosure. Was the tool’s role explained, and can a person question or correct the process? What evidence supports this score’s meaning for this population and proposed use? A satisfactory answer to the first is valuable, but cannot stand in for the second. Ask the provider to name the precise conclusion and purpose, then look for evidence that follows that inference instead of relying on polished wording, consistent scoring, or a human signature alone.

Sources: Defining Test-Score Interpretation, Use, and Claims: Delphi Study for the Validity Argument; Validity of Test Interpretation and Use; Standards for Educational and Psychological Testing (2014 Edition)

What evidence would support the report's actual conclusion?

Evidence for a personality report should match the sentence a reader is being asked to believe. “Your answers suggest that you prefer planned work” is a claim about a tendency under a particular measure. “You will perform well in this role” adds a claim about behavior and outcome in a setting. The second does not automatically follow from the first. The question is not whether a test is called “validated,” but what evidence supports this interpretation for this proposed use. The Standards for Educational and Psychological Testing, a 2014 joint publication of the American Educational Research Association, American Psychological Association, and National Council on Measurement in Education, frames validity as the degree to which evidence and theory support interpretations for proposed uses. Its publisher record establishes the document’s identity and scope, not the validity of a particular personality instrument. “Defining Test-Score Interpretation, Use, and Claims: Delphi Study for the Validity Argument,” a 2023 conceptual study, explains why claims must be explicit before evidence can be judged. Its three-round Delphi process drew on 13 university-affiliated experts in the United States, with seven completing the final round. This specialized panel can clarify concepts; it cannot validate a commercial report or establish broad consensus among measurement professionals. Match the report’s claim to the evidence it needs: - For a self-reflection tendency, ask what construct the instrument measures, how its questions represent it, how responses become a score, and what limits are disclosed. Evidence about item content and respondents’ understanding may matter. Clear explanation makes the claim easier to inspect, not universally true of the reader. - For comparison with other people, ask who the norm group was, when it was assessed, and whether it fits the reader and setting. A norm group is the population against which a score is compared. Without that information, a percentile or “above average” label lacks a clear reference. Even a suitable comparison describes relative standing on the measure, not a fixed identity or guarantee about future conduct. - For prediction or selection, ask whether evidence addresses the specific outcome, population, and context of use. The Delphi study describes relations between scores and other variables, including future performance, as one possible source of evidence. But a tendency score cannot by itself establish job fit, performance, or a hiring recommendation. Reliability is separate. It concerns score consistency or precision under specified conditions; it does not show that an interpretation is correct or a consequential use justified. Similar scores over time can coexist with a claim broader than the evidence permits. An exploratory report might still offer a useful prompt for reflection without supporting employment decisions. “Validity of Test Interpretation and Use,” a 1990 ETS research report, makes a related point: validity is a matter of degree, may be strengthened or challenged by later findings, and concerns score meaning, relevance, utility, action, and consequences. This conceptual report is not current validation evidence for a personality test. It helps explain why one badge or general assurance cannot settle the question. Ask the provider to name the population, exact score interpretation, and intended use that the evidence supports. If a conclusion moves from reflection to a consequential work decision, the evidence should support that same step; otherwise, keep the conclusion narrower.

Sources: Defining Test-Score Interpretation, Use, and Claims: Delphi Study for the Validity Argument; Validity of Test Interpretation and Use; Standards for Educational and Psychological Testing (2014 Edition)

What does earlier computer-generated report research add—and not add?

A 2000 review found earlier computer-generated reports promising as adjuncts to professional judgment and called for more research before broader use. It is useful historical context, not an evaluation of modern AI systems or a reason to assume every automated personality report is inaccurate.

The review, “Validity and utility of computer-based test interpretation,” considered whether computer-administered tests were comparable with paper-and-pencil versions, and surveyed computerized methods in psychiatric screening, personality assessment, neuropsychology, and personnel psychology. Its abstract says computer-generated reports appeared valuable as additions to, rather than substitutes for, clinical judgment, while further studies were needed to support wider use. This measured conclusion neither endorses every automated interpretation nor says computer involvement makes an assessment unsound. The review is available through [PubMed](https://pubmed.ncbi.nlm.nih.gov/10752359/).

The paper’s breadth is also a limit. It reviews different applications, not a controlled comparison of one personality instrument in one population. Its abstract gives no pooled accuracy estimate and does not establish that every report type performed alike. It cannot answer whether a current report’s wording is accurate, whether scores mean the same thing across groups, or whether they support a decision about a specific role. Published in 2000, it predates contemporary generative systems. “Computer-generated report” should not be silently updated to mean a modern language model; this review does not test that difference.

The review helps test two tempting conclusions: that automation necessarily makes a report untrustworthy, and that computer-produced interpretation is dependable because it is systematic or sounds authoritative. It supports neither shortcut. “Adjunct” describes a supporting role; it does not certify the content. Professional judgment can add context, but a human signature alone cannot show that a score interpretation is supported. These are implications for reading the finding, not separate results measured by the review.

The comparison with APA’s June 2026 “Responsible Use of AI in Assessment: A Practical Overview” clarifies what this older research can contribute. The review addresses an earlier evidence base about computer-based assessment and narrative interpretation. APA’s overview is an informational practice resource about responsible AI use, not a study of a named personality report. Neither supplies instrument-specific evidence for a score claim. The provider would need to identify what the instrument measures, which population supports the interpretation, and whether the evidence fits the proposed use. The older review cautions against automatic trust or rejection; the current overview prompts questions about process. Neither turns polished narrative into a justified conclusion.

For a reader deciding what to do, computer involvement is a reason to ask what the system contributed, not a verdict about accuracy. Examine the conclusion and the evidence offered for its intended purpose. The historical review supports careful evaluation; it does not settle the case for any current tool.

Sources: Validity and utility of computer-based test interpretation; Responsible Use of AI in Assessment: A Practical Overview

What should I do before acting on the report?

Start with the sentence that could change what you do. Ask the provider what it means, what scores it rests on and whom it describes. Then ask whether AI contributed to scoring, interpretation, or narrative. Ask what it contributed, who reviewed it, and how to raise an error. APA’s Responsible Use of AI in Assessment: A Practical Overview names transparency, consent, privacy, oversight, competence, and monitoring as practical concerns. These questions clarify the process; they do not establish that the score’s meaning is supported.

Keep the next step tied to the report’s claim. The 2023 Delphi study Defining Test-Score Interpretation, Use, and Claims distinguishes the meaning assigned to a score from the action proposed on its basis. Its expert process clarifies concepts; it does not validate a particular personality instrument. A statement about a tendency for self-reflection and a claim that the same score predicts success in a particular role are different inferences. Ask what was measured, for whom the interpretation is supported, and whether evidence matches the intended use. Compare the report with observations across situations, allowing for context and change.

A clear account of AI use is neither proof of accuracy nor proof of error; a human signature cannot substitute for evidence. The verdict could change if formal standards are revised or evidence specific to the instrument, population, claim, and use supports a different conclusion. For structured prompts about work decisions and collaboration, the live Work Pattern Report at /assessment offers low-stakes reflection. It has no norms, cutoff, or type, and is not a hiring score or job recommendation. Use it to frame observations, not choose a career.

Sources: Responsible Use of AI in Assessment: A Practical Overview; Defining Test-Score Interpretation, Use, and Claims: Delphi Study for the Validity Argument

Questions readers ask