Gains:
- Understand what the working paper should contain within the scope of BDS 230 and how to use artificial intelligence to produce a draft procedure, conclusion and justification.
- Ability to quickly create a working paper framework that makes traceable the work done, the evidence obtained and the conclusion reached, with the support of artificial intelligence
- Understanding that the documentation produced by artificial intelligence is a draft and is not valid until it is evidenced and reviewed and signed by the auditor.
There is a famous saying in auditing: "You are not considered to have done what you have not done, but what you have not done to document." No matter how well an auditor performs a test, if he does not write down that test, the evidence he obtained and the conclusion he reached on the working paper, that work will be considered "not done" in the audit quality review or in the event of a lawsuit. BDS 230 (Audit documentation standard) regulates what working papers should contain and sets a single criterion: another auditor who is experienced but not involved in the audit should be able to understand the work performed, the evidence obtained and the conclusion reached by looking at the working paper. This is called traceability.
In this unit, we will cover how to use artificial intelligence in drafting workpapers. AI is very powerful at producing well-structured, consistent and readable outlines; It reduces hours of typing work to minutes. But let's be clear from the beginning: the AI-generated text is a draft, not documentation. Documentation is evidenced, reviewed, corrected and owned by the auditor. An unsigned draft has no evidentiary value in the audit file.
Anatomy of a good worksheet
According to BDS 230 and good practice, a working paper typically contains the following elements:
element
what to write
Why is it important?
Title and ID
Customer, period, account/subject, prepared by, date
Traceability and accountability
Purpose/target
What claim does this procedure test?
Linkage of the test to the audit purpose
work done
What was done step by step, which method
repeatability
Source of evidence
Which document/system/confirmation was used, references
commitment to evidence
Scope / sample
How tested, how chosen
Competency assessment
Findings / exceptions
What was found, in numbers
transparency
Conclusion
Is the claim met, on the grounds
supervisory judgment
review trail
Who prepared it, who reviewed it?
quality control
AI builds this framework instantly and removes the burden of starting with a blank page. But two pillars are never left to him: evidence source references (you link to actual documents/tests) and conclusion (audit judgment is yours). The AI might write a sentence like “result: claim is satisfied”; But it is the auditor's job to stand behind this sentence and see that it is supported by evidence.
The right way to draft with AI
You can use AI in worksheet drafting in two modes: skeleton generator (create a blank template) and text editor (translate the raw notes you enter into neat, coherent text). The second mode is particularly safe because you provide the content; AI only improves language and structure.
Step by step:
- Collect your raw score. What you tested, what document you looked at, what you found — write it down in short notes.
- Make AI say configure. Have these notes prepared according to BDS 230 elements; But the content is yours.
- You add evidence references. Manually insert actual links such as "See: bank confirmation ref. B-12, contract ref. S-04".
- Write the result yourself or check it closely. AI may suggest conclusions, but the final sentence depends on your reasoning.
- Review and own. Read the text line by line; See if there are any exaggerations, gaps, or unsubstantiated claims; then sign.
Attention: Do not tell the AI "write a worksheet for this account" and put the output into the file as is. Since AI does not have any evidence, it can add plausible but fabricated details (non-existent document references, fabricated amounts, tests that were not actually performed). A made-up "work done" description is the most dangerous mistake in documentation.
three mini cases
Case 1 — Efficient and accurate. An auditor did the verification test of bank balances: took 8 bank confirmations, reconciled them all with the trial balance, found a difference and explained it. He gave his raw notes (which bank, which amount, which difference) to YZ and asked for a draft working paper in BDS 230 format. AI produced regular text in 4 minutes; The auditor added confirmation references, wrote the conclusion in his own words, and signed it. The 40-minute writing task was reduced to 10 minutes, the content was entirely up to the auditor.
Case 2 — The fabricated detail trap. A team member told the AI to "write an inventory valuation worksheet" without giving any raw notes. The AI produced a persuasive text containing made-up references, describing tests that had not actually been performed (net realizable value analysis, aging table) as if they had been performed. The team member put it in the file. Quality review revealed that the references did not actually exist; This created a serious documentation and ethics problem. Lesson: AI can write work that doesn't exist as if it were "done"; You must always provide the content.
Case 3 — Leaving the outcome to AI. An auditor prepared the worksheet for the receivables allowance test with AI. The evidence actually showed that the provision might have been inadequate, but the AI wrote in generic language “conclusion: the provision is reasonable” and the auditor accepted this without question. The officer realized that the evidence contradicted the conclusion. Lesson: the concluding sentence is a control judgment; Never automatically accept the default optimistic language of AI.
Weak prompt / Strong prompt
Weak prompt:
Write a worksheet for stock testing.
Problem: no content. The AI makes up the test, the amounts, and the result because it has no proof. A dangerous text emerges that describes work that has not actually been done.
Powerful prompt:
Your role: you are a documentation assistant for an independent auditor. I provide the content; You will organize it according to the BDS 230 structure and clarify the language. DO NOT ADD any tests, amounts or document references that are not in my raw notes.Topic: Testing of the claim that inventories are valued at net realizable value.My raw notes:- Scope: Out of 320 inventory items, the 40 largest by amount (78% of the total) were selected.- Method: Cost value of each item was compared with its estimated net realizable value after sales; source: [sales price list], [after-sales cost estimate].- Finding: cost in 3 items > net realizable value; A total impairment of 145,000 TL was detected; no provision has been made.- My draft conclusion: 145,000 TL missing provision for impairment; is below materiality (€600,000) but corrections will be recommended.Task:1) Organize this with the following headings: Purpose/Claim, Work Performed, Scope and Selection, Source of Evidence, Findings, Conclusion.2) Keep [square brackets] placeholders for source of evidence; I will add real references.3) Write the concluding sentence following my outline; Don't be optimistic.
This prompt is powerful because the checker gives the content, sets the "do not append" rule, leaves the references as placeholders, and links the result to the checker's draft.
Common mistakes
- Asking for a draft without content. Having the AI write a worksheet without giving raw grades; produces fake work.
- Not noticing the made-up references. Placing non-existent document/test references in the file.
- Leaving the outcome to AI. Accepting the optimistic default language without questioning.
- Not relying on evidence. Leaving an unreferenced, untraceable text.
- Sign without reviewing. Taking ownership of the draft without reading it line by line.
Tip: When you have the AI prepare the worksheet, instruct it to "add nothing that is not in my raw notes, mark the missing parts as '[MISSING: ...]'". So the AI shows you the gaps instead of making them up; You fill it with evidence.
In summary
The working paper is the memory and evidence of the audit; BDS 230 requires traceability. AI is an excellent drafting and editing tool that removes the burden of blank pages: it scaffolds, turns raw notes into neat text, ensures consistency. But the substance of the documentation—the reality of the work performed, evidence references, and concluding judgment—belongs to the auditor. If you don't provide the content, AI will make it up; If you don't connect the outcome, AI will optimize it. Rule: content from inspector, layout from AI; signature and responsibility from the auditor.
Application task
Write down the raw notes of an audit procedure you performed (or a hypothetical one) in 5-6 items: purpose, method, scope, findings, draft conclusion. With the powerful prompt pattern above, have the AI edit it in the BDS 230 structure; Establish "do not add" and "mark missing" rules. Read the output and check to see if (1) there are any fabricated details, (2) the result is true to your judgment, and (3) the evidence references remain as placeholders.
checklist
- [ ] I provided the content (raw notes); I banned AI from fabricating.
- [ ] The working paper contains BDS 230 elements (objective, task, scope, evidence, findings, conclusion).
- [ ] I linked the evidence references to the actual documents; No made-up references.
- [ ] The concluding sentence reflects my audit judgment; not optimistic.
- [ ] I read the text line by line; No unsubstantiated claims or exaggerations.
- [ ] “Would another experienced auditor understand this?” I passed the (traceability) test.
- [ ] I reviewed and adopted the draft and added the preparer/date information.