How to Use ChatGPT and Claude for an XBRL Applicability Check: A Step-by-Step Guide for CAs
Yes, a general AI assistant can help you decide whether a company must file AOC-4 XBRL, as long as you paste Rule 3 yourself, give it anonymised facts, and make it list every assumption. Below is the exact prompt sequence, the answers you should expect, and the checks that catch the mistakes AI tends to make.
It pairs with our reference post, XBRL applicability under the Companies Act: who must file AOC-4 XBRL. Read it for the rules; this post is the workflow.
Last reviewed: 26 September 2026. Educational guide, not professional advice.
What can ChatGPT or Claude do here, and what should you never delegate?
| Good use of AI | Never delegate |
|---|---|
| Applying each limb of Rule 3 you pasted to a fact pattern | Remembering thresholds from training data |
| Producing a limb-by-limb table with reasons | Deciding a fact you have not verified (is the parent listed in India?) |
| Flagging missing facts as questions | Concluding on exemption without the company's registration category |
| Drafting a dated working-paper note | Filing, signing or certifying anything |
| Explaining Rule 3(2) in plain words | Pasting real company names, CIN, PAN or GSTIN |
Applicability looks simple, but it turns on four limbs, an exemption proviso and a carry-forward rule. That combination is exactly where an AI working from memory slips, and where an AI working from text you pasted does well.
Step 1: Set up a safe workspace
- Never paste the company name, CIN, PAN, GSTIN or director details. Call the company "Company X" and round figures.
- Use synthetic profiles for practice. On a live file, reduce the facts to the few the rule needs: listed or not, paid-up capital, turnover, accounting framework, sector, prior XBRL history.
- Store your role prompt once so you do not retype it: ChatGPT Custom Instructions or a Project, a Claude Project, or a Gemini Gem. Otherwise keep it in a text file.
Redaction wording is in our DPDP-safe prompt template library for CA firms.
Step 2: Give the role prompt
You are a compliance assistant helping a chartered accountant decide
whether a company must file financial statements in e-form AOC-4 XBRL.
Rules:
1. Use ONLY the rule text I paste in this chat. If something you need
is not in it, write "not provided" and ask me. Do not use remembered
thresholds or rule numbers.
2. Test every limb separately and show a pass/fail for each.
3. Check the exemption proviso and any carry-forward rule before
concluding.
4. List every assumption. Never assume a fact I have not stated.
5. End with: conclusion, taxonomy, limb relied on, open questions.
Step 3: Paste the governing rule yourself
Copy the text of Rule 3 of the Companies (Filing of Documents and Forms in Extensible Business Reporting Language) Rules, 2015, as amended up to the 2025 amendment, from the MCA site or another source you trust. Include Rule 3(1) with its proviso, Rule 3(2) and Rule 3(3). Provide the source, ask it to apply it.
RULE TEXT (pasted from Rule 3, XBRL Rules 2015, as amended):
[Paste Rule 3(1)(i)-(iv), the proviso to 3(1), Rule 3(1A), Rule 3(2)
and Rule 3(3) here.]
Reply with a short list of the tests and exemptions this text creates,
one line each. Do not analyse any company yet.
Compare its list with the text. If it has added a test that is not in your paste, stop and correct it.
Step 4: Run the worked cases
Four synthetic unlisted companies for FY 2025-26, each testing a different trap.
Apply the rule text above to each company. State the limb relied on,
the form (AOC-4 or AOC-4 XBRL) and, where XBRL applies, which
taxonomy annexure. Show pass/fail for every limb, list assumptions,
and quote the sentence of the rule for each conclusion.
Company W: unlisted private company, paid-up capital Rs 1 crore,
turnover Rs 60 crore this year. It filed AOC-4 XBRL two years ago
when its turnover was Rs 120 crore. Not Ind AS. Not an NBFC, bank
or insurer.
Company X: unlisted private company, paid-up capital Rs 3.5 crore,
turnover Rs 88 crore, not Ind AS, no prior XBRL filing, not an
NBFC, bank or insurer.
Company Y: wholly owned subsidiary of a company listed on an Indian
stock exchange. Paid-up capital Rs 2 crore, turnover Rs 40 crore,
prepares accounts under the Accounting Standards rules, no prior
XBRL filing, not an NBFC, bank or insurer.
Company Z: NBFC, paid-up capital Rs 50 crore, prepares Ind AS
financial statements, no prior XBRL filing.
Then challenge it: "Which of these conclusions would change if Company Y's parent were listed only on a foreign exchange?"
What a good answer looks like
Compute these yourself from the verified rules so you know the target.
| Company | Limb tests | Result |
|---|---|---|
| W | Fails size limbs this year, but has filed in XBRL before | AOC-4 XBRL under Rule 3(2), once filed always file. Taxonomy: Annexure-II (AS), as it is not Ind AS |
| X | Not listed; capital 3.5 crore is below 5 crore; turnover 88 crore is below 100 crore; not Ind AS; no prior XBRL | Plain AOC-4 |
| Y | Indian subsidiary of an India-listed company, so limb (i), regardless of its own size | AOC-4 XBRL. Taxonomy: Annexure-II (AS) |
| Z | Meets the Ind AS and size limbs, but NBFCs are exempt under the proviso | Plain AOC-4 |
Illustrative, wording varies run to run: a good AI reply reaches these four conclusions, quotes the pasted rule text, says the Company Y result depends on the parent being listed in India, and notes that Company W's answer turns on its prior XBRL filing being a filing under Rule 3(1) or the earlier XBRL rules, which you should confirm from the filing history. Also expect it to mention the signed PDF attachment required since 14 July 2025 if your pasted text includes Rule 3(1A).
Where AI goes wrong
| Failure | What it looks like | Check |
|---|---|---|
| Outdated thresholds | States an old paid-up capital or turnover limit, or drops a limb | Confirm ₹5 crore and ₹100 crore appear in your pasted text |
| Invented rule numbers | Cites "Rule 3(4)" or a proviso that is not there | Match every cited rule against the text you pasted |
| Stale power-sector exemption | Treats power companies as exempt, an old rule removed in 2017 | Search the answer for "power"; the proviso in your text will not include it |
| Silent assumptions | Assumes the company is not Ind AS-bound, or treats a foreign-listed parent as listed | Read the assumptions list against your facts |
| Forgetting Rule 3(2) | Re-tests size from scratch and lets Company W go back to plain AOC-4 | Ask: "Has this company ever filed in XBRL?" |
Five checks on every output:
- Both size limits and all four limbs are present and match your pasted text.
- The exemption proviso was tested before the conclusion.
- Rule 3(2) was considered, with a stated answer on prior filing.
- Each cited rule number exists in your paste.
- The assumptions list contains nothing you did not provide.
Step 5: Verify in 60 seconds
Enter the same profile into the free XBRL applicability checker. It returns the limb that applies, the taxonomy and a working paper you can file. If it disagrees with the AI, trust neither yet: recheck the facts you entered, then the rule text. A mismatch usually means a fact was entered differently or the AI assumed one.
Do ChatGPT, Claude and Gemini differ for this task?
Not in the method. In mechanics:
- All three let you paste rule text, and current paid plans let you attach a PDF. Pasting the specific rule is still safer than relying on the model to find it.
- Custom Instructions or Projects (ChatGPT), Projects (Claude) and Gems (Gemini) can store your Step 2 prompt. Menu names change, so check the current interface.
- None of them knows your client's filing history. That fact always comes from you.
Frequently asked questions
Can ChatGPT or Claude tell me whether a company must file AOC-4 XBRL?
It can apply Rule 3 to the facts you give it and show its working. It cannot know the company's listing status, prior filings or accounting framework unless you state them, and the conclusion remains yours.
Is it safe to paste company details into an AI tool?
Use "Company X", rounded figures and no CIN, PAN or GSTIN. Check your firm's policy and the tool's data settings.
Why paste Rule 3 instead of asking the AI what it says?
Models may recall an old version, such as one that still lists a power-sector exemption. Pasted text makes the task application, which you can verify.
What is the most common AI mistake on XBRL applicability?
Testing size only and missing the Ind AS limb, the Indian-subsidiary limb, or the once-filed-always-file rule. Require a pass/fail for every limb.
Does an NBFC preparing Ind AS statements file AOC-4 XBRL?
No. The proviso to Rule 3(1) exempts NBFCs, housing finance, banking and insurance companies, so they file the ordinary AOC-4. Make sure the AI applies the proviso before the limbs.
How should I record an AI-assisted applicability conclusion?
Keep a dated working paper with each limb, the source figures traced to the audited statements, the rule text version used, and your own conclusion. Note that AI assisted the first draft and that you verified it.
Where CORAA fits
A chat gives you a draft; the audit file needs the conclusion kept with its evidence. CORAA's Report Studio helps assemble engagement reporting so the applicability working paper sits alongside it. To see how that fits your engagements, start a free trial.
Last reviewed: 26 September 2026. Sources: Companies (Filing of Documents and Forms in XBRL) Rules, 2015, as amended to 2025, as summarised in our companion post.