Why Generic AI Gets Indian Personal Finance Wrong
Plenty of young Indian investors now ask a general-purpose AI (ChatGPT, Claude, Gemini, Perplexity) to help them plan their SIPs, choose a mutual fund, or estimate retirement corpus. The answers sound confident and look tidy. For Indian personal finance, they are often wrong in ways that matter.
This isn't a takedown of AI. I think AI can be genuinely useful for planning, and I run a small AI planner at finplann.com/financial-planner. The distinction is between foundation models used cold on India questions and purpose-built tools with Indian tax, regulation, and market data plumbed in. Those are two very different products, even if their chat interfaces look identical.
Five Things Generic ChatGPT Gets Wrong About Indian Finance
1. It quotes outdated LTCG rates
Ask a general-purpose model in 2026: "What's the LTCG tax rate on equity mutual funds in India?" Often you'll get "10% above ₹1 lakh". This is the pre-Budget-2024 rate. The current rate has been 12.5% above ₹1.25 lakh since July 23, 2024. The training data for many models still has the old regime as the majority representation, so the LLM statistically favours the wrong answer.
This isn't a rare edge case. The LTCG rate is one of the most common Indian tax questions, and getting it wrong throws off every calculation that follows.
2. It hallucinates fund names
Ask for "top 3 flexi-cap mutual funds in India" and you'll frequently get a plausible-sounding list that includes funds that don't exist, or funds that were merged years ago. "Parag Parikh Flexi Cap Direct Growth" is a real fund. "HDFC Ultra Flexi Cap Direct" is not, but a model will offer it with confidence.
Foundation models don't distinguish between real fund names in their training corpus and plausible-sounding constructions. Without live AMFI data to check against, a model will invent fund names as a matter of course.
3. It doesn't know Indian regulatory constraints
Section 123 (formerly 80C) limits, ELSS 3-year lock-in, NPS Tier 1 vs Tier 2 differences, SGB 8-year maturity, PPF 15-year plus 5-year extensions, the SEBI overseas investment industry cap that periodically freezes international MFs. A generic model knows the words for all of these, but not how they interact.
Example: ask "should I move money from my NRE FD to an Indian mutual fund now?" and a generic model will give a portfolio-management answer. It won't flag that if you're a US-resident NRI, the resulting Indian MF becomes a PFIC and requires annual Form 8621 filings. That is a tax-compliance issue first, and the model misses it entirely.
4. It gets currency and inflation wrong
Retirement corpus calculations require inflation assumptions. Generic AI will use "3% inflation" from US-centric training data. Indian long-run inflation has averaged 5-6%. A model that plans a 30-year retirement at 3% inflation will badly under-estimate the corpus needed.
Similar issue with currency assumptions when planning cross-border retirement (India-US NRI cases). The model doesn't know INR has structurally weakened at 3-4%/year against USD historically.
5. It confuses accounts that Indians can't have and vice-versa
Ask about "your Roth IRA in India" and generic AI won't flag that Roth IRAs are US-only. Ask about "your PPF in the US" and similar confusion. Or worse: it will offer Roth IRA guidance to an Indian resident who cannot legally contribute to one from India.
The 5 Test Questions to Vet Any AI Advisor
Before trusting any AI tool with your Indian finances, ask these five questions. A tool that answers all five correctly is grounded in Indian data. A tool that gets any of them wrong is running on general knowledge and should not be used for planning.
Test 1: "What's the current LTCG tax rate on equity mutual funds held for more than 12 months?"
Correct answer: 12.5% on gains above ₹1.25 lakh in a year, for any sale on or after July 23, 2024 (Budget 2024; now Section 198 of the Income-tax Act 2025). The rate depends on the sale date, not the purchase date: the old 10% above ₹1 lakh applied only to sales before July 23, 2024.
Test 2: "How long is ELSS mutual fund locked-in, and can I do a SIP into ELSS?"
Correct answer: Each individual investment in ELSS is locked in for 3 years from the date of that investment. A SIP into ELSS means each monthly instalment has its own 3-year lock. Your first-year SIP contributions become redeemable 36 months after each contribution.
Test 3: "What's the difference between NPS Tier 1 and Tier 2?"
Correct answer: Tier 1 is the retirement account. For non-government subscribers, normal exit after 15 years or at age 60 (whichever is earlier), at least 20% of the corpus must buy an annuity and up to 80% can be taken as lump sum, of which 60% of the corpus is tax-free (PFRDA exit rules, December 2025; government-sector subscribers still need 40% annuity). Tier 2 is a voluntary open-ended savings account with no lock-in and no tax benefit for private-sector employees; withdrawals are taxed at applicable rates. Government employees get separate 80C benefit on Tier 2 contributions.
Test 4: "How are Sovereign Gold Bonds taxed on maturity vs interim sale?"
Correct answer: If held to maturity (8 years), capital gains on redemption are fully exempt, but from 1 April 2026 only for investors who subscribed at issue and held continuously until maturity (Section 70(1)(x), Income-tax Act 2025, as amended by Finance Act 2026); secondary-market buyers pay capital-gains tax even at maturity. If sold on the exchange before maturity, LTCG at 12.5% applies after 12-month holding. The 2.5% annual interest is taxable at slab rate throughout.
Test 5: "What triggers TCS on foreign remittances, and at what rate?"
Correct answer: TCS (Section 206C(1G), now Section 394 of the Income-tax Act, 2025) applies to LRS remittances above ₹10 lakh in a FY. From FY 2026-27 the rate is 2% for education and medical (nil if funded by an education loan) and 20% for other purposes such as investments and gifts; overseas tour packages attract a flat 2% from the first rupee. TCS is a refundable credit against your ITR, not a permanent tax. See our TCS guide.
Bonus test: ask "should I use the new tax regime or the old for Tax Year 2026-27?" and see whether the response references the current standard deduction (₹75K for salaried under new regime), the ₹12L rebate threshold under new regime (residents only), and the point at which old regime beats new (never up to ₹12.75L salary; roughly ₹6–8.5L of total deductions including the ₹50K standard deduction at ₹15–25L+). An answer that skips these isn't working from current rules.
What a Purpose-Built AI Financial Planner Actually Needs
Reasonable requirements for an "AI financial planner" for Indian investors:
- Live AMFI + NSE/BSE data feeds for actual fund NAVs, prices, and category classifications.
- Current Income Tax Act sections and rules: LTCG, plus slabs, rebates, deductions and the regime comparison.
- SEBI and RBI regulatory awareness: mutual fund categorisation, LRS limits, SEBI investment caps, insurance IRDAI rules.
- Currency and inflation defaults appropriate to India: 5-6% CPI, 3-4% INR/USD drag.
- Compliance flags for NRIs, especially US-resident NRIs (PFIC), and other DTAA cases.
- Explicit calibration for retirement corpus, education planning, insurance sizing.
Without these, an "AI financial planner" is a chat window on top of a general model. It will answer confidently and often get the details wrong.
Where FinPlann's AI Planner Fits
The FinPlann AI planner is one of several tools designed for Indian personal finance specifically. It works from live AMFI fund data, current SEBI and tax rules, and defaults set for India rather than the US. FinChat is the conversational layer on top. Full disclosure: I built these. The same tests apply to any other Indian-specific tool you compare.
What's different from ChatGPT for the same question:
- Tax rates reference current 2026 rules, not 2019 training data.
- Fund suggestions pull from actual AMFI-registered schemes, not hallucinated names.
- Retirement corpus math uses a 6% inflation default and India-appropriate returns (10% during accumulation, 7% after retirement).
- NRI questions flag DTAA and PFIC angles automatically.
When You Should NOT Use Any AI (Yet)
Being honest about limits:
- Complex NRI cases with US-India tax interactions, business income, and estate planning. These need a CA who specialises in cross-border tax. AI helps with framing but not filing.
- Business ownership tax planning (LLP vs Pvt Ltd choice, ESOP taxation timing, dividend distribution strategy). The rules are deep enough that a CA earns her fee.
- Anything involving property purchase or sale with LTCG in the 8-figure range. A mistake here costs far more than the time AI saves.
How to Get Better Answers From General AI Tools
If you'd rather use a general AI tool as a starting point, three ways to make its answers more reliable:
- ChatGPT / Claude / Gemini with grounding. Copy the exact current tax rules into your prompt before asking. Prompts like "Given the current Indian LTCG rate of 12.5% above ₹1.25L per FY per Budget 2024, help me plan..." make made-up answers much less likely.
- Perplexity for factual queries. Its citation model surfaces sources; you can verify each claim.
- A spreadsheet for the maths. Use a spreadsheet, not the chatbot, for corpus maths.