How to Use AI for ITR Filing: A CA's Workflow
CA Prateek Agarwal ·
Income-tax return season is a volume game. A single firm can file hundreds of returns in a few weeks, and most of the work — pulling the AIS and Form 26AS, reading a Form 16, decoding a capital-gains statement, comparing the two regimes — is data-heavy and repetitive. This is exactly where AI for ITR filing earns its place: it does the gathering, extraction, and first-cut computation, and the CA reviews, applies judgement, and owns the final position filed under the client's authentication.
This piece lays out a step-by-step workflow for using AI across the ITR cycle, where it helps, where it does not, and how to use it without putting client data at risk.
What AI actually does in ITR filing
AI in ITR work automates the steps a preparer would otherwise do by hand: reconciling the Annual Information Statement (AIS) and Form 26AS against the client's own documents, extracting numbers from a Form 16 or a broker's capital-gains report, running an old-versus-new-regime comparison, and drafting the computation for review. It does not decide the legal position, and it does not file — the return goes up under the client's authentication and the CA's professional responsibility.
The reason it matters is the same one that drives every return season: the work is high-volume, deadline-driven, and unforgiving of small errors. A TDS entry that sits in the AIS but never makes it into the return is a refund left on the table; an interest income the client forgot to mention but that the department already sees in the AIS is a notice waiting to happen. Machine-assisted reconciliation catches both faster than a manual pass through a stack of PDFs.
Before automating anything, one rule has to be settled: never paste a client's PAN, income figures, or source documents into a public, general-purpose LLM. Those are confidential and, under the DPDP Act, 2023, personal data you are responsible for. Use domain tools built for Indian tax that handle data inside a controlled environment — see DPDP Act and AI Tools: Handling Client Data for the longer treatment. With that settled, here is the workflow.
Step 1 — Gather and reconcile the data
The return starts not with the client's documents but with what the department already knows. The AIS and the derived Taxpayer Information Summary (TIS) consolidate reported salary, interest, dividends, securities transactions, and high-value transactions; Form 26AS carries the TDS/TCS and advance-tax credits. The first job is to pull all three and reconcile them against the client's own records.
This is where AI replaces the worst part of the job — eyeballing a long AIS against bank statements and broker notes:
- Reconcile AIS/TIS against client documents. The tool lines up each AIS line item against the salary slip, interest certificate, or capital-gains statement and classifies it as matched, present in AIS but missing in the client's documents, or present in the documents but missing from AIS.
- Catch the silent omissions. Savings-bank and FD interest, dividend income, and small TDS entries are the items clients routinely forget; if they sit in the AIS and not in the draft return, that is a mismatch the department will flag.
- Flag the AIS errors. The AIS is not gospel — duplicate reporting, sale consideration shown gross of the buyer's TDS, and wrongly tagged transactions are common. AI surfaces these as exceptions; the CA decides whether to accept the AIS figure, submit feedback on the AIS, or take a different position with documentation.
ClearTax is built around importing the prefill and Form 26AS data and reconciling it for Indian filers. The point is to enter Step 2 with a clean, reconciled picture of income, not a pile of unmatched PDFs.
Step 2 — Extract numbers from the documents
Once you know what to match against, the documents themselves have to be turned into figures. This is document extraction, and it is the second clear win:
- Form 16 — Part A (TDS deducted and deposited) and Part B (the salary break-up, exemptions, and deductions) can be read straight off the PDF into the computation. The check that matters: the TDS in Form 16 Part A should tie to Form 26AS, and any gap is the first thing to resolve.
- Capital-gains statements — broker and mutual-fund statements come in wildly different formats. AI extraction pulls the buy/sell dates, consideration, and cost, then helps bucket each transaction into short-term and long-term. For listed equity and equity mutual funds, watch the grandfathering of cost as on 31 January 2018 for long-held holdings — a place tools sometimes get the cost base wrong, so it needs a CA's eye.
- Interest and other certificates — bank interest certificates, FD statements, and home-loan interest certificates feed the income and the relevant deductions.
Extraction is fast but not infallible. Treat every extracted figure as a draft until it has been tied to a source — the Form 16 to 26AS, the capital-gains total to the broker statement, the interest to the certificate. For where extraction tends to go wrong on Indian documents, see What AI Gets Wrong About Indian Tax.
Step 3 — Computation and old-vs-new regime comparison
With clean, reconciled, extracted data, the computation is largely mechanical — and AI does it quickly, including the comparison that every salaried and small-business client now asks about.
The new tax regime is the default, with the old regime available as an opt-in. The old regime keeps the familiar deductions and exemptions — Section 80C, 80D, HRA, home-loan interest under Section 24(b) and the rest — while the new regime offers lower slab rates but strips most of those out. Which one is cheaper depends entirely on the client's deduction profile, and the honest answer is found by computing both.
This is where AI saves real time: it runs the liability under both regimes side by side and shows the difference for that specific client, instead of the preparer building two computations by hand. A few cautions:
- Verify the regime rules against the relevant assessment year. Slab rates, the rebate threshold, the standard deduction, and surcharge treatment change, and a tool trained on an older year will quietly use stale numbers. Check the output against the current law for the assessment year you are filing — and if you are unsure of a figure, look it up rather than trust the draft.
- The comparison is a starting point, not advice. A client locked into a home loan or running long-term 80C commitments may be better served by the old regime even in a year the raw comparison says otherwise. That is a judgement call.
- Set-off and carry-forward of losses — capital losses, house-property loss, and business loss carry their own rules and time limits. AI can apply them, but the CA confirms the set-off order and the carry-forward.
For research on a genuinely uncertain position — the treatment of a particular receipt, a contested deduction — a research assistant is more useful than a calculator. VIDUR and Taxmann.ai answer Indian tax-law questions with citations to the underlying provisions and case law, which is what you want when you have to defend a position rather than just total a column.
Step 4 — Draft and cross-check
With the regime chosen, AI assembles the computation and maps it to the correct ITR form. The right form follows the income profile: ITR-1 (Sahaj) for simple resident salaried cases within the eligibility limits, ITR-2 where there is capital gains or more than one house property, ITR-3 for business or professional income, and ITR-4 (Sugam) for presumptive income under Sections 44AD/44ADA. Picking the wrong form is a defective-return problem, so this mapping is worth confirming.
The cross-checks before anything goes near the portal:
- The income totals tie back to the reconciled AIS and the source documents from Steps 1 and 2.
- The TDS and advance-tax credits claimed match Form 26AS — claiming credit not reflected there invites a mismatch.
- Chosen deductions are supported by documents on file, not just asserted.
- The regime selection in the draft matches the comparison decision, and where a business assessee is switching regimes the Form 10-IEA requirement is handled.
Aalekh covers ITR filing alongside practice management, which helps when the same draft has to move through preparer, reviewer, and partner without losing the audit trail. For drafting a reply if a notice follows, TaxBotGPT drafts responses grounded in Indian tax law — useful, but the reply is still the CA's to verify and sign.
Step 5 — Review, e-verify, and file
The last step is the one AI does not do. Filing on the income-tax portal happens under the client's authentication, and the return is only valid once it is e-verified — by Aadhaar OTP, net-banking, or a digital signature where applicable — within the prescribed window after submission. Miss the verification and the return is treated as not filed.
So the review before submission is the CA's, not the software's:
- Read the full computation, not just the bottom line. A clean refund figure can sit on top of a wrong head-of-income classification.
- Confirm the high-judgement items — capital-gains grandfathering, regime choice, loss set-offs, and any position that departs from the AIS.
- Keep the working papers. The reconciliation, the extracted documents, the regime comparison, and the sign-off should be retained so that if a notice or scrutiny comes, you can show exactly how each figure was arrived at and who approved it.
Then the client e-verifies, and the return is filed. The AI prepared it; the CA stands behind it.
Where a human is still required
- Judgement calls — head-of-income classification, the treatment of an unusual receipt, residential status under the Section 6 day-count rules, and contested deductions need professional reasoning.
- Regime and planning advice — the cheaper regime this year is not always the right long-term call, and only a CA weighs that against the client's circumstances.
- Sign-off and verification — the return is filed under the client's authentication and the CA's responsibility; AI does not assume that.
- Defending the position — if the return draws scrutiny, the CA argues it. AI can draft and research, but the position taken is the professional's.
Frequently asked questions
Can AI file my income-tax return automatically?
AI can gather data, reconcile the AIS and 26AS, extract figures from documents, compute the liability, and draft the return — but filing and e-verification happen under the taxpayer's own authentication on the income-tax portal. In practice the CA reviews the draft and the client e-verifies; the AI does the preparation, not the legal submission.
Is it safe to use AI for ITR filing with client data?
Only with the right tool. Do not paste a client's PAN, income figures, or source documents into a public, general-purpose chatbot — that is confidential personal data you are accountable for under the DPDP Act, 2023. Use domain tools built for Indian tax that process data in a controlled environment, and keep your working papers.
Can AI tell me whether the old or new regime is better?
It can compute the liability under both regimes for a specific client and show the difference, which is a strong starting point. But the comparison uses the deductions on record and assumes the slab rules for the assessment year are current — verify those, and remember that long-term commitments like a home loan may make the old regime the better call even when the raw numbers are close.
Will AI replace tax preparers and CAs?
No. AI removes the repetitive gathering, extraction, and computation, but classification, regime advice, defending a position, and sign-off stay with the professional. See Will AI Replace Chartered Accountants in India? for the longer view.
The takeaway
AI does not change what an income-tax return requires — it changes how much of it the CA does by hand. Gathering and reconciling the AIS and 26AS, reading a Form 16 or a capital-gains statement, and comparing the two regimes are increasingly handled by software built for the Indian regime, leaving the classification, the judgement, the regime advice, and the sign-off as the work that needs a Chartered Accountant. Use the tools to clear the volume, keep the client's data inside a trusted environment, and review every figure before it is filed — the return goes up under your name. Browse the software directory to see the current options.
Related software
TaxBotGPT
AI tax assistant trained on Indian tax law for cited answers and notice drafting
VIDUR
AI assistant for Indian tax, corporate and regulatory law research and drafting
Taxmann.ai
AI legal and tax research and drafting assistant backed by Taxmann content
Aalekh
AI practice management and GST/ITR filing platform for CA firms