AI tools can accelerate a CA's drafting, summarising, and first-pass research, but they can hallucinate section numbers, rates and case law, so every output must be verified against the bare Act and official portals before it is relied on. Use AI for speed, not as the source of truth.
Where AI genuinely helps
- Drafting — replies to notices, engagement letters, internal notes (then edit and verify).
- Summarising — long circulars, judgments and agreements.
- Data work — explaining Excel/Power Query formulas, writing reconciliation logic.
- Structuring — turning rough notes into a clear client memo.
Where AI fails (and the risk)
General AI models can confidently produce wrong section numbers, outdated rates, and non-existent case citations. For a YMYL field like tax, one wrong section can mislead a client. AI also may not know the latest amendment (for example, a Finance Act change after its training cut-off). Treat every legal fact as unverified until checked.
A verify-before-you-rely workflow
- Use AI for a first draft or to find the likely provision.
- Verify each section, rate and date against incometax.gov.in / gst.gov.in / cbic.gov.in.
- Confirm case law on an authoritative database — never cite a judgment you have not checked exists.
- Apply your professional judgment and sign off.
Prompts that work
- Give context: assessment year, regime, entity type, facts.
- Ask for section references you can verify, and request the model to flag uncertainty.
- Ask for structured output (tables, step lists) for client use.
- Never paste confidential client data into tools without a confidentiality/usage check.
Key takeaways
- Use AI for drafting, summarising and data work — for speed.
- AI can invent sections, rates and case law — verify everything.
- Confirm legal facts on official portals before relying.
- Give context, request verifiable references, protect client data.