Skip to content
PlagiatScanner

AI transparency · Template · KIT-03

Create an AI tools register: structure and example

Direct answer: A useful AI tools register does more than list product names. For every tool actually used, give the provider, displayed version and access date, purpose, affected part of the work, input category, extent of adoption, human checks and location of supporting evidence. Adapt the fillable register below to your assessment. If your institution supplies a mandatory form, use that form rather than replacing it with this model.

A fillable register with decision-relevant fields

Use one row for each tool and distinct kind of activity. If a product generated search terms and later proposed analysis code, two rows are clearer because the inputs, controls and effect on the work differ. Keep placeholders until you have reconciled the table with your working records.

FieldEntry
Register number[T-01]
Tool and provider[product name; provider]
Version and access[displayed model/version, or “not displayed”; date or period]
Activity and purpose[for example, developing search terms for Chapter 2]
Part affected[chapter, figure, file, function or dataset]
Input category[original prose, public source, code or data; do not reproduce unnecessary content]
Output and adoption[kind of response; rejected, partly used or used after revision]
Human verification[original-source review, test, comparison, independent calculation or expert editing]
Supporting record[log ID, appendix item, version comparison or protected location]
Applicable instruction[assessment regulation, brief or departmental form and its date]

Completed T-01 line: “[Product] by [provider], displayed version [value/not displayed], accessed on [date] to develop search terms for Chapter 2. I searched the suggestions independently in [database] and cited only original publications I opened and checked. Evidence: activity R-014 and redacted extract A-3.”

A register is not a bibliography or a declaration

A bibliography identifies cited publications, datasets and other source material. A tools register identifies technology and its role in producing the work. A declaration often summarises scope and responsibility in prose. Entering a generative system in the register does not make its response a scholarly source. Facts, quotations and publication details must still lead to inspectable original material.

The methods section can require more detail than the register. If generated code affected analysis, describe data flow, modifications and tests with the method, while the register provides an overview and points to that passage. Stable identifiers avoid repeating entire explanations across the dissertation. They also help an examiner connect a tool entry, a log row and a permitted appendix record.

No universal rule says that every assessment must contain a separate AI tools register. Institutions may provide a specific table, require an academic-integrity declaration, or set discipline-specific documentation. Review the current brief, regulations and departmental guidance. This page offers a transparent structure where such a register is permitted; it does not displace the authoritative local document.

Condense a working log without erasing distinctions

A detailed AI use log may contain many interactions per product. Turn those entries into coherent activity categories only after the work is complete. Ten sessions developing synonyms for one search can potentially become one row if the period, subject, selection process and evidence remain clear. A later language-editing task deserves another row because its input, intervention and verification are different.

Take version details from contemporaneous notes. If no stable model label was shown, enter “not displayed” rather than a number reconstructed from memory. A date range may suit repeated identical work when the local form allows it. The aim is an honest technical description, not an illusion of accuracy.

The input-category field should signal sensitivity without pasting confidential material into the register. “Pseudonymised interview passages processed under the approved protocol” is more useful than reproducing an interview in a prompt column. Whether that processing was permissible depends on consent, the service, institutional approval and the protection required. A register records the decision; it cannot itself authorise the input.

Place and cross-reference the register deliberately

Use the location prescribed by the assessment: perhaps a section after the bibliography, an appendix item or a form accompanying the declaration. Do not insert it into the bibliography without checking. Scholarly sources and production tools answer different evidential questions even when a local style guide uses its own labels.

Assign each row a short identifier. A methods passage can then state: “The code assistance recorded as T-03 was limited to preprocessing; Appendix B reports changes and validation.” Give the prompt extract or export the same label. Readers can follow the connection without searching through a folder of product names and screenshots.

Design the submitted table for reading. Use complete but concise descriptions, define abbreviations and allow enough space for checks. A multi-page register is preferable to a crowded landscape sheet with illegible type. Lengthy validation detail can move to a numbered note if the connection remains unambiguous.

Do not turn the register into a disclosure risk

Never place credentials, identifiable raw data or confidential full prompts in a tools register. The table does not prove that a conversation is complete or that an activity was permitted. Avoid blanket statements about a provider’s storage, sharing or security unless they have been established for the relevant service, account and settings.

Vague minimising language is equally unhelpful. “Only minor assistance” does not define a contribution; “three shortening options for two original sentences, then independently edited” does. A substantial use does not need dramatic wording either. Function, scope, verification and adoption are the relevant categories.

The German Research Foundation maintains that researchers remain responsible for content and form when generative models are used. European living guidelines connect responsible use with transparency and reproducibility. Those principles support a traceable register, but they do not impose this exact table on every student in every jurisdiction.

Audit the final register in eight passes

  1. Compare listed products with working files and contemporaneous records.
  2. Separate materially different uses of the same product.
  3. Include version information only when it was shown or otherwise evidenced.
  4. Map every activity to an identifiable part of the work.
  5. Distinguish generation, adoption, human modification and verification.
  6. Test that each evidence ID points to an available, permitted record.
  7. Remove or redact protected material using the agreed process.
  8. Compare the format, placement and language with local instructions.

Read each row from an examiner’s perspective. If the system’s effect on research or writing is still unclear, replace labels such as “support” with an observable action. The AI transparency hub connects the related records. The foundation page on academic writing and evidence explains why tools, sources and authorship require distinct treatment. Use the companion guide to describe AI use in the methods section when a register entry needs a methodological explanation.

Official foundations

Sources reviewed 4 September 2026. Local assessment, privacy and data-governance requirements remain controlling.