The table below sets disclosure expectations across three levels of AI use. Pay attention to any legislative, regulatory, contractual, licensing, or publisher requirements around AI use disclosure, which tend to apply more to public agencies, which take precedence over the guidance below (Kostyniuk 2025, app. C).
| Type of AI use | Disclosure expectation |
|---|---|
| Assistive features in tools already in everyday use, including embedded assistants such as Microsoft Copilot (spell-check, grammar suggestions, autocomplete, reference formatting) | Avoid disclosing. This is routine assistance that does not substantively affect findings or outcomes. Anecdotal evidence suggests that any mention of AI, regardless of its actual role or accuracy, can reflexively lower a reviewer's perceived credibility of a work product and erode the trust that disclosure is meant to build. Over-disclosure of this kind punishes transparency rather than rewarding it, and ultimately discourages the kind of disclosure this guidance is meant to encourage. |
| Material generative use that drafts, rewrites, or synthesizes content, or that shapes methodology, analysis, findings, or conclusions, especially where it directly informs decision-making | Presume disclosure. Where the recipient's expectations are known, align with them. Where they are not, either seek them out or disclose the material use directly, defaulting to disclosure if no answer comes. As a starting point for that disclosure, consider sharing your organization's Responsible Use of AI policy (if you have one) with a statement that your AI use is material, giving the recipient the opportunity to ask for specifics. Re-send the statement whenever needed to keep the recipient's understanding of your AI use current. |
| AI use in safety-of-life and other critical functions, including its use therein as part of code, algorithm, or model development | Disclose. The disclosure must cover the specifics of the use, stating what AI did, where, and how it was verified, rather than a general statement that AI was involved. This covers signal controllers, adaptive control applications, vehicle safety functions, and other operational technology, wherever AI contributes to a system that can injure people or affect safe network operation if it behaves incorrectly. The same bar applies whether AI touched the code, the calibration, the assumptions, or the documented design criteria. |
Have the conversation early. The best protection is to be upfront with the recipient on their expectations around AI use and disclosure, and be responsive to those. The purpose is to maintain or build trust and encourage responsible AI use. An upfront conversation or contract clause gives you confidence that you are calibrating disclosure correctly and helps avoid potential surprises. This conversation can serve as the disclosure, if it satisfies the recipient’s needs and expectations.
Be fully accountable to the work, and keep that separate from the disclosure question. Consistent with responsible charge and the "QA/QC Procedures" section, every deliverable needs to be human-vetted, not an AI product requiring a professional to review it, correct it, and takes responsibility. But reliability and disclosure answer different questions. Responsibility asks whether the work is correct. Transparency asks whether the recipient has an independent interest in knowing how it was made, regardless of how high quality the work is. The rationale is in part so that the recipient can appropriately tailor how carefully they review the material. Accountability is a precondition for ethical AI use, not a substitute for disclosure where the tiers above call for it.
A useful gut check. If you would be uncomfortable disclosing use of AI, reflect on why that is. The discomfort may signal that the AI use itself is inappropriate, in which case adjust your approach. Or the disclosure itself may feel awkward. If that’s because you don’t know how it would be received, that’s a sign to have that conversation upfront. If it’s because you’re not familiar with industry norms, then defer to the guidelines above and apply them in a manner that would build or maintain professional trust with the recipient and/or would encourage responsible AI use.
Comparison versus Transportation Research Board (TRB)’s standard. Our guidance builds on and takes further the general approach from the National Society of Professional Engineers, which is that: “Ethical principles favor transparency when AI plays a substantial role in generating work products. To uphold ethical standards, engineers integrating AI into their practice should … consider disclosing AI involvement when it plays a significant role in the final product.” (NSPE 2024)
That said, our guidance is narrower than Transportation Research Board’s standard which requires authors to disclose any use of generative AI (TRB n.d.). The reason is that engineering practice has evolved to include overlapping systems to ensure quality and trust, namely the principles of responsible charge, professional licensure, QA/QC review, and applicable code and standards. Thus, the work generally carries external checks that do not depend on knowing whether AI was involved, whereas research work does not. For research work, neither reviewer nor reader can independently re-derive a novel finding without replication and so disclosure of AI use is essential for calibrating how much scrutiny a claim needs, since the claim itself can be tested only against plausibility and the author's word.
Proprietary and security restrictions. Where proprietary or security restrictions limit what can be said publicly, the record of what AI produced and how it was verified should be in the project file, software documentation, or client agreement.
Graphics and figures. Treat a figure the way you treat text: what matters is whether it is accurate, not whether a person or a tool produced it. A reviewer checks a figure against the underlying data and the design intent regardless of origin, and an AI-generated graphic is not inherently lower value if it survives that check. AI-generated or AI-assisted visuals can be disclosed in the figure caption, stating plainly what the image is and is not, for example a concept rendering rather than a design drawing. See the figure caption sample under "Example language."
Calibrate the detail. This kit provides samples to help you calibrate the level of detail on a disclosure, when disclosing specific use. Too vague, and it can read as though something is being concealed; too granular, and it risks implying the team lacked the expertise to do the work itself. An acknowledgment that notes AI's role alongside the human team's contributions answers both concerns.
What goes in a disclosure statement
When specific disclosure is warranted, the level of detail matters. Centers for Disease Control and Prevention (CDC) guidance for scientific work can be a useful structure, which builds a disclosure from five elements: content affected + action taken + AI tool + purpose of AI use + human oversight, as described below (CDC 2026):
The five elements of a disclosure statement
Example language
These disclosure examples are starting points, and not standards. Note that not all five of the above elements are included, based on their relevance to the situation. Adapt these samples to your organization's policy and to any client or publisher requirements for the work product.
Agency planning document or staff report detailing review of a project: Placed in the introduction pages.
Portions of this document were drafted with the assistance of generative AI tools approved for [agency] use. Staff reviewed and verified all content, and [agency] is responsible for the analysis and recommendations presented. AI tools were not used to make or recommend any determination affecting an individual application, permit, or entitlement.
Stamped or sealed engineering document: Placed with the seal block.
AI-assisted tools were used in preparing portions of this document, as documented in the project file, which can be found at [location]. The undersigned professional has reviewed the complete document, independently verified all standard references, specifications, and numerical values against primary sources, applying the reasonable standard of care ordinarily exercised by members of the profession practicing under similar conditions. Use of AI-assisted tools does not alter that standard of care. The undersigned professional is responsible for the accuracy and completeness of the document.
Proposal or statement of qualifications: Placed in the proposal and/or contract.
Contractor may use Generative Artificial Intelligence (AI) in developing Deliverables through commercially available business or enterprise versions that do not use Contractor’s or client’s inputs or outputs to train the AI provider’s models. Contractor shall remain responsible for reviewing and verifying all Deliverables and for complying with the confidentiality and data-security requirements of this Agreement.
Consultant report with material contribution from GenAI: To add in the methodology section for work where AI contributed to technical content, analysis, or synthesis.
[Tool name and version] was used to summarize the literature reviewed in Chapter 2 and to draft the initial text of Sections 3.1 through 3.4. All sources were independently retrieved and verified, and all numerical values, standard references, and citations were checked against primary sources. [Firm name] is responsible for the methodology, analysis, findings, and recommendations in this report.
Public-facing communication and translated material:
This material was translated from English using [tool name] and reviewed by [staff, contractor, or reviewer]. Please contact [contact] for any questions on this document.
Slide deck: Placed on the closing notes-and-sources slide.
Sources: [list]. Generative AI tools ([tool name]) were used to assist with drafting narrative content and formatting. Slide 12 graphic generated using [tool name]. All analysis and conclusions are those of [firm or agency name].
Figure or visualization: Placed in the caption.
Figure 4. Conceptual rendering of the proposed streetscape. Image generated using [tool name and version] and vetted by report authors; not a design drawing, not for construction, and not to scale.
Code, script, or analysis tool: Placed in the header of the script.
Portions of this script were generated with [tool name and version]. All logic has been reviewed and outputs validated against [known dataset, hand calculation, or benchmark case] by [name or role].
Compliance with legislative requirements and software licensing: Placed in the methodology section or the project record.
AI-assisted tools were used in preparing this deliverable in accordance with [firm or agency] AI policy, applicable legislative and regulatory requirements, and the license terms of each AI tool and data source used. No licensed standards, specifications, or third-party copyrighted material were entered into an AI tool except where the applicable license permits that use. All AI-assisted content was reviewed and verified by the project team, exercising a reasonable standard of care, and [firm or agency name] is responsible for the work presented.
Work produced with a proprietary or restricted-access AI tool: Placed where the deliverable describes its methodology or limitations.
This deliverable was prepared using [tool name], operated within [firm or agency] controlled environment. The tool's configuration, source data, and prompts are subject to [confidentiality, security, or licensing] restrictions and are not published here; they are documented in the project record and available to [client or agency] on request. [Firm or agency name] has reviewed and verified all content and is responsible for the analysis, findings, and recommendations presented.
Sources cited on this page are listed in full on the References page.