Gains:
- Ability to use artificial intelligence as a navigator and confirm each reference from the official text when connecting an audit or accounting question to the relevant standard (BDS, TFRS/TMS, VUK).
- The standard trick of AI is the ability to recognize a tendency to hallucinate matter and date and cross-check it with a primary source.
- Ability to understand that standard interpretation and application decisions are the responsibility of the auditor and that artificial intelligence output never replaces the official text
The auditor's job comes back to a constant question: "Is this legal?" When should an income be recognized? How is a lease accounted for? Under what condition is a provision reserved? An audit procedure is a task required by which standard? The answers to these questions are written in standards and legislation. The main reference sources of the auditor in Türkiye are: BDS (Independent Auditing Standards - regulates how the audit is carried out, compatible with ISA, published by the POA); TFRS/TMS (Turkish Financial Reporting Standards / Turkish Accounting Standards — regulates how financial statements are prepared, compatible with IFRS); and VUK (Tax Procedure Law) on the tax side. Additionally, BRSA, CMB, Treasury and Finance regulations come into play depending on the sector.
In this unit, we will cover how to use AI as a navigator in standards and regulatory research, but why you must verify every citation from the primary official source. Because this is exactly where AI is most dangerous: its tendency to invent standard numbers, clauses, paragraphs and effective dates.
Why AI is prone to hallucinations in legislation
The artificial intelligence language model generates "possible" sentences from the patterns in the texts on which it is trained. Because legal and technical texts fit a certain pattern — such as “according to IAS 36 paragraph 12…” — the model is very good at imitating that pattern; but he cannot know whether he filled the mold correctly or not. The result: a non-existent article number, a wrong paragraph, a made-up phrase "changed in 2022" appear before you as fluently as if it were real. This is called source hallucination and leads to disaster in the audit; because a conclusion based on a false reference to a standard is both technically erroneous and untenable.
So the golden rule: use AI to understand the topic and find out which standard might be relevant; but do not use any clause, paragraph or date without confirming it from the official text. AI tells you "you should look here"; You don't trust someone who says "it says exactly that".
Safe use of AI
Dangerous use of AI
“Which family of standards does this topic fall into?”
"Give me the article number and the full text, I will write it down"
"Explain this concept to me in plain language"
"Tell me if this article changes in 2023" (unconfirmed)
"Draft the decision tree/checklist"
"Cite the standard, I'll put it in quotes" (unconfirmed)
"What questions should I ask?"
"That fits my case, for sure."
Verification discipline: rule of three sources
The safe workflow in standards research is this:
- Navigate (AI). Explain the topic to the AI, ask what standard(s) might be relevant and the concept in plain language. This is a start.
- Go to primary source (official text). Look at the actual text of the standard that YZ refers to — the official BDS/TFRS text published by the KGK, the Official Gazette, the website of the relevant institution. Read the article and paragraph there.
- Implement and document (auditor). You interpret and apply the standard to your own case; In the working paper, write the attribution that you confirmed from the official source, not the attribution given by AI.
Caution: Do not place a quoted "standard quote" from the AI directly into your worksheet or report. That sentence may be made up. If you are going to quote, copy the image from the text and show the source.
Where AI really adds value: understanding and structuring
The risk of hallucinations does not make AI useless in legislation; it just limits its role. AI is truly accelerator at explaining a complex standard in plain language, listing which questions solve a topic, sketching a decision tree, and roughly mapping the relationship between different standards. For example, "what criteria are considered when determining whether a lease is a financial lease or an operating lease?" AI gives you a thinking framework to the question; you validate this framework with the official standard and apply it to your case. If the frame is right, you accelerate; If it is wrong, the official text will correct you anyway.
three mini cases
Case 1 — Correct use. An auditor was uncertain about a client's timing of revenue recognition on a long-term contract. I explain the subject to AI and ask "which standard family does this fall into, what concepts should I look at?" he asked. AI led to the revenue recognition standard and concepts such as "performance obligation" and "transfer of control". The auditor went to the official standard text, read the relevant paragraphs, applied it to the case, and cited the official source in the working paper. AI gave direction, text verified, auditor made decision.
Case 2 — Fake matter trap. A team member asked AI, "Which article contains the provisioning requirement?" he asked; YZ gave a very clear reference such as "TMS 37 paragraph 14/b, 2022 revision". The team member wrote this directly on the worksheet. During the quality review, it was revealed that such a "2022 revision" and subclause "14/b" do not exist in that form; The reference has been corrected. Lesson: clear-looking attribution is not correct attribution; Confirmation is required.
Case 3 — Copying the quote. An auditor asked AI "in full text" for a standard clause to put in the report and directly used the sentence in quotation marks. The sentence was fluent, but not exactly the same as the wording in the official text; There was a shift in meaning. The person in charge saw the difference when he compared it to the official text. Lesson: the standard quote is copied only from the official source.
Weak prompt / Strong prompt
Weak prompt:
Write down the exact article and number of the standard regarding provisioning and I will put it in my worksheet.
Problem: Asking for "resource" directly from the AI. The item number and text can be made up and enter the file without verification.
Powerful prompt:
Your role: you are a research assistant to an independent auditor. You will give me DIRECTION; I will verify the official text. Assume that you are NOT SURE about the item number, paragraph, and date, and indicate that I need to confirm them each time.Topic: Should a customer recognize a provision for a possible liability arising from a past event or disclose it as a "contingent liability" in a footnote? Task:1) Explain in plain language which family of standards this topic falls into and which basic concepts (definition of liability, probability, reliable measurement, etc.) are decisive.2) Outline a decision tree draft of the questions I need to ask in order to make the decision.3) Do not present ANY article numbers, paragraphs or dates as definitive information; Note "confirm from the official text".4) DO NOT QUOTE the standard text in quotation marks; I will get it from the official source.
This demand is strong because it anchors the AI in the role of navigator, prohibits quote and clause fabrication, and leaves the decision to the decision tree and the auditor.
Common mistakes
- Mistaking AI for a resource. Using without confirming the article number, paragraph, date.
- Copying made-up quote. Putting the AI's quoted "standard text" into the report as if it were official text.
- The "it looked clear" fallacy. The more specific the reference, the more reliable it is considered; whereas specificity hides fabrication.
- Failure to verify the effective date. Relying on statements like "it changed that year" without confirmation; The legislation is constantly updated.
- Blindly applying it to the case. Applying the general explanation of AI without questioning the specific conditions of the business.
Tip: When you ask the AI a standard question, have it include these two questions in its answer: “From what official source should I confirm this information?” and “What are the chances that this attribution is false?” This leads to confirmation for yourself and the AI.
In summary
Standards and regulatory research is the area where AI adds the most value but is most dangerous. Value: explains complex subject in plain language, directs to the relevant standard, produces decision tree and question list. Danger: invents item number, paragraph and date (source hallucination). Rule: AI is a wayfinder, not a source. Confirm each reference from the official BDS/TFRS text published by the POA and the relevant legislation; copy the quote only from the official source; The auditor makes the interpretation and application. AI output never replaces official text.
Application task
Select an accounting/auditing question (e.g. classification of a lease, whether a provision should be made). With the above powerful prompt pattern, ask the AI for (1) the relevant standards family, (2) defining concepts, (3) a decision tree outline; Prohibit fabricating items/dates. Then find the official text of the standard that the AI points to, read the relevant clause yourself, and note whether the AI's framework is correct. List the differences.
checklist
- [ ] I used the AI as a direction finder; I did not count the source for the article/paragraph/date.
- [ ] I put the "make up attribution, state that confirmation is required, do not cite" rules in the prompt.
- [ ] I found and read the official text of the standard that YZ pointed out (KGK/Official Gazette).
- [ ] I wrote only the attribution that I confirmed from the official source on the working paper.
- [ ] I verified the effective date and current version from the official source.
- [ ] I interpreted and applied the standard to my own case.
- [ ] I have only copied the quotes I used from the official text and cited the source.