CORAA
Tool guide · Perplexity

Perplexity for audit.

The citation-first answer engine. Where it fits in audit research — recent ICAI / MCA / RBI / SEBI updates, case-law lookups, fast triangulation of facts. When citations matter and when they don’t.

Updated 28 May 2026 · ~8 min read

1 · What Perplexity is, for an auditor

Perplexity is a search-grounded answer engine. Instead of producing a free-form generative response from training data alone, it browses the web in real time, finds relevant sources, and produces an answer with inline citations to each source. For an auditor, this matters: the most common hallucination patterns of ChatGPT and Claude (made-up section numbers, fictional cases, stale rates) are exactly what Perplexity’s architecture is designed to mitigate.

That’s the upside. The downside: Perplexity is a research tool, not a drafting tool. The answers are shorter and tighter than ChatGPT or Claude, which is great for fact-finding and terrible for first-draft engagement letters.

2 · What Perplexity is good at, in audit work

2.1 · Recent regulatory updates

“What did the MCA notify about CSR Form CSR-2 in March 2026?”, “Has the GSTR-9C threshold changed for FY 2025-26?”, “What’s the latest RBI Master Direction on concurrent audit?” — Perplexity is the right first stop. The citation links take you directly to the gazette notification or the regulator’s circular.

2.2 · Case law and disciplinary orders

“NFRA orders against bank branch auditors in 2025 and 2026”, “ICAI disciplinary proceedings involving Section 13(3) violations”, “ITAT rulings on Section 44AB applicability for presumptive-tax opt-out” — Perplexity aggregates across orders, summarises, cites. The summary is a starting point; you read the actual order before relying on it.

2.3 · Cross-checking what ChatGPT or Claude told you

When a generative model gives you a confident answer with a citation, paste the claim into Perplexity to verify. If Perplexity can’t find a source for it, that’s a strong hallucination signal. Treating Perplexity as your default fact-checker tightens the rest of your workflow.

2.4 · Quick standards lookups

“What does SA 540 say about disclosure of significant assumptions?”, “Which Ind AS governs leases of low-value assets?”, “What’s the threshold under Section 188 for material related-party transactions?” — Perplexity finds the relevant text and cites it. Faster than navigating the official website.

2.5 · Sector and industry primers

Before a planning meeting on a new client in an unfamiliar sector (jewellery, real estate, EdTech, hospital), Perplexity Pro Search produces a structured industry brief with sources. Useful for risk- identification under SA 315 — you scan it, decide what’s relevant, ignore the rest.

3 · What Perplexity is not good at

  • Drafting long-form audit documents. Perplexity’s outputs are shaped for question-answering, not for producing an engagement letter or an SA 260 communication. Use Claude or ChatGPT for drafting; come back to Perplexity for fact verification.
  • Multi-step reasoning over your own documents. Perplexity can read uploaded files (Pro tier), but the citation-first architecture is built around web sources. For multi-document reasoning over a stack of board minutes and contracts, Claude or ChatGPT’s long context will serve better.
  • Anything that needs a maintained context. Perplexity treats each query as relatively stateless. ChatGPT Memory and Claude Projects let you build up engagement context; Perplexity doesn’t.
  • Code execution. No equivalent to Code Interpreter. For data work on a CSV, use ChatGPT Plus or Claude with the analysis tool.
  • Source quality varies. Perplexity will sometimes cite an opinion blog post alongside an official notification. The citation alone doesn’t guarantee authority — read the source.

4 · Specific features worth knowing

  • Pro Search. Deeper, multi-step reasoning. Worth using for any non-trivial regulatory question — it surfaces a much wider pool of sources.
  • Focus modes. Restrict the search to specific source types — Academic, Reddit, YouTube, Wolfram Alpha. The Academic focus is useful for international audit-research papers; the default Web focus is what you’ll use most.
  • Spaces. Workspace areas where you can keep related queries and uploaded files together. A Space per ongoing research thread (e.g. one for “CARO 2020 updates”, one for “ICAI AI initiatives”) keeps context organised.
  • File upload (Pro). You can upload PDFs and ask grounded questions against them. For a regulator’s 80-page consultation paper, this is faster than reading top-to-bottom.
  • Threads with follow-ups. Each answer suggests two or three follow-up queries. Click these to drill down on a regulatory thread quickly.
  • API access. Available, though most CAs won’t need it. Useful for audit-tooling vendors building citation-grounded research into their products.

5 · Use by audit phase

  • Acceptance and planning. Strong fit. Industry briefings, recent regulatory shifts affecting the client’s sector, NFRA orders affecting peer firms.
  • Risk assessment. Usable. SA 315 risk universe is generic enough that ChatGPT/Claude do this without needing live web access.
  • Sampling, substantive testing. Not the right tool. Methodology lives in the standards; pattern detection needs audit-grade tools.
  • CARO drafting, communications. Not the right tool — drafting is Claude/ChatGPT’s strength.
  • Concluding work. Use for verification — does the conclusion you’ve drafted cite the right paragraph of the right SA? Paste the claim into Perplexity, see if the citation holds up.
  • Ongoing professional reading. Strong fit. Set up Spaces for each topic you follow (AI in audit, ICAI announcements, NFRA orders, MCA notifications) and run weekly check-in queries.

6 · Prompts that play to Perplexity’s strengths

Recent regulatory development scan
What are the material regulatory developments affecting Indian statutory audit in the last 90 days? Cover changes from ICAI (including AASB), NFRA (orders against auditors), MCA (Companies Act amendments, CARO clarifications), CBDT (Section 44AB or transfer-pricing), and SEBI (LODR / audit-committee guidance).

For each, give me:
- The notification / order / circular reference
- A two-line summary
- The link to the official source

Stay in the last 90 days. Indian regulators only.
Verify a citation
I have a draft audit observation that cites "SA 540 (Revised), paragraph A47" for the principle that the auditor's risk assessment of accounting estimates should include the susceptibility of those estimates to bias.

Verify: does SA 540 (Revised) paragraph A47 actually say this? Quote the exact text. If not, which paragraph does say something similar? Cite the source you used to verify.
NFRA disciplinary patterns
Summarise the most common professional-conduct findings in NFRA orders against statutory auditors during 2024-2026. Group by:

1. SA-citation patterns (which SA was cited as breached most often)
2. Industry of the audited entity
3. Severity of the order (debarment vs monetary penalty vs reprimand)

Cite the orders. Skip cases that are still pending — only closed orders.

7 · Tier comparison and data handling

  • Perplexity Free. Five Pro Searches per day, unlimited basic searches. Citations are the same quality at both tiers.
  • Perplexity Pro. Around USD 20 / month. Unlimited Pro Search, file uploads, Spaces, more model choices (you can pick the underlying LLM — GPT-4o, Claude, Sonar).
  • Data handling. Perplexity’s default privacy posture is that conversations may be used to improve their models unless you opt out under Settings → Account → AI Data Retention. For any substantive use, toggle this off. Treat all queries as if a third party could read them — keep identifiable client data out.
  • Enterprise tier. Perplexity Enterprise has contracted no-training. Less common in CA firms than ChatGPT Enterprise or Claude for Work, but it exists.

8 · Common gotchas

  • Citations don’t guarantee authority. Perplexity will cite a CA-firm blog post next to an MCA notification. Always check which source the answer rests on, and click through to the original.
  • Generative summaries can drift from sources. The cited text says one thing; the model summary states a slightly stronger claim. Read the original quote when the matter is material.
  • Indian-context regulatory results sometimes mix US/UK references. Especially on ESG, BRSR, IFRS-vs-Ind-AS questions. Specify “Indian context” or “Ind AS, not IFRS” in the prompt.
  • Same answer shifts with focus mode. A query in Academic focus produces different sources from the same query in Web focus. For regulatory work, default to Web.

9 · Where to start, this week

  • Run a Pro Search on “Indian audit-firm regulatory developments last 30 days”. Read the top three sources end to end. You’ll absorb more in 30 minutes than scrolling LinkedIn for a week.
  • Pick one citation in a current working paper. Verify it through Perplexity. Build the habit.
  • Create a Space called “ICAI & AI” — run a weekly check-in query (“what’s new on ai.icai.org this week”) and let the thread accumulate context over time.

For the broader principles — what to keep off prompts, how to document AI use under SA 230 — the Practical Guide is the place to start.

How to read this guide

CORAA does not endorse any specific AI tool. This guide describes how Indian CAs use the named product in audit work — what tends to work, what tends not to, and the practical considerations around client data. It is not an integration guide, an affiliate page, or a recommendation. You decide which tool fits your engagement.

Whichever tool you choose, the principles in the Practical Guide still apply: AI assists, the auditor decides. Keep identifiable client data off prompts that go to consumer tiers. Document AI use under SA 230. Verify every citation.

For official AI credentials and CPE-eligible programmes, refer to ICAI’s AI portal — AICA Level 1, AURA, and the AI Innovation Summit. CORAA AI Lab is a free practice environment, not a regulator substitute.

For real client data — built differently
CORAA is the AI engine built for Indian audit — India-hosted, DPDPA 2023 compliant, no training reuse.
Seven hubs covering the whole engagement: Setup → Scrutiny → Reconciliation → Procedures → Working Papers → Findings → Reporting. 164 scrutiny rules across 13 modules. SA 230, 240, 320, 450, 510, 530, 570, 700 baked into the workflow.
India-hosted
Every byte on Indian soil. Azure India regions for all production workloads.
DPDPA 2023 compliant
Data-fiduciary obligations built in by default. Not bolted on after the fact.
ISO 27001:2022 · SOC 2 Type II
Certified and attested. The security baselines enterprise audit committees actually ask for.
No training reuse
Client data never enters any foundation model. Contractual, not aspirational.
See CORAA in 20 minutesExplore the AI ModulesVisit the Trust Centre