AI Bookkeeping for Micro-Business: 15 Prompts That Handle Primary Docs, VAT, and Tax Filings Without a Full-Time Accountant

Running a micro-business or being a self-employed entrepreneur (ИП) in 2026 means juggling two jobs: your actual business and the paperwork behind it. Primary documents (первичка), VAT reconciliation, advance payments under the simplified tax system (УСН), and the ever-present threat of a ФНС requirement letter — it all piles up. Hiring a full-time accountant is expensive; outsourcing costs money per operation. But a large chunk of this work is pattern recognition, reconciliation, and drafting — exactly what a modern LLM does well when you give it a precise prompt.

This is not a "AI will replace your accountant" article. It is a practical collection of 15 prompts that cover the routine: parsing invoices and acts, matching amounts across documents, calculating tax burden under different regimes, drafting responses to tax authorities, and preparing declarations. Each prompt includes the task it solves, the exact wording, and a realistic example of the result.

A word of caution up front: an LLM is a drafting and analysis assistant, not a source of legal truth. All final numbers must be verified against source documents and current legislation — the Tax Code of the Russian Federation (НК РФ), Federal Law No. 402-FZ "On Accounting," and current ФНС clarifications. Use these prompts to accelerate, not to outsource responsibility.

Why This Works for Micro-Business Specifically

Micro-businesses rarely have exotic transactions. Their document flow is repetitive: supplier invoices (счета-фактуры), acts of completed work (акты выполненных работ), payment orders (платёжки), and bank statements. This repetitiveness is what makes prompt-driven automation effective.

The table above reflects typical workflows reported by small accounting practices, not a controlled study — treat the numbers as directional.

How to Structure Prompts for Accounting

The single biggest failure mode is asking a vague question and getting a vague answer. A reliable accounting prompt has four parts:

1. Role and constraint — "You are an assistant to an accountant working under the Russian tax system."
2. Input format — tell the model exactly what you are pasting (invoice text, CSV, bank statement).
3. Output format — demand a table, JSON, or a checklist. Ambiguity here produces prose you cannot use.
4. Verification instruction — "flag anything you are unsure about instead of guessing."

That last instruction is critical. For accounting, a hallucinated number is worse than a missing one.

Prompt 1 — Turning Raw Invoice Text into a Structured Table

Task: Extract line items from a pasted invoice into a table you can drop into Excel.

You are an accounting assistant. Below is the raw text of a supplier invoice.
Extract every line item into a Markdown table with columns:

| Item name | Qty | Unit price | Amount excl. VAT | VAT rate | VAT amount | Total incl. VAT |
Rules:
- Do not invent values. If a field is missing, write "NOT FOUND".
- Preserve the exact item names from the source.
- At the end, state the invoice total and check whether it equals the sum of line items.
Invoice text:
<paste here>

Example result: The model returns a clean table plus a one-line reconciliation: "Sum of line items = 118,000.00 ₽, matches stated total. VAT at 20% = 19,666.67 ₽." You copy the table into Excel and move on.

Prompt 2 — Reconciling Acts Against Payment Orders

Task: Match a list of acts of completed work to bank payments and find gaps.

You are reconciling acts of completed work against bank payments.
I will give you two lists: (A) acts with number, date, counterparty, amount;
(B) payments with date, counterparty, amount.
Match them. Output a table:

| Act | Counterparty | Act amount | Matched payment | Difference | Status |
Status is one of: MATCHED, PARTIAL, UNMATCHED.
Do not assume a match if amounts differ. Explain each non-match in one line.
List A:
<paste>
List B:
<paste>

Example result: "Act #47, ООО Ромашка, 60,000 ₽ — UNMATCHED, no payment found in the period. Act #48 — PARTIAL, paid 40,000 of 60,000 ₽." This is exactly the list you send to the client for follow-up.

Prompt 3 — Checking a Counterparty Before You Sign

Task: Draft a due-diligence checklist and risk note for a new supplier.

You are a risk assistant for a micro-business.
I am about to sign a contract with <counterparty name>, INN <number>.
Produce a due-diligence checklist I should verify manually, covering:
- tax registration status and whether the entity is active,
- signs of a "one-day firm" (фирма-однодневка),
- contract clauses that expose me to VAT recovery risk,
- what documents to request before the first payment.
Do not state facts about this company — I will verify them myself.
Only give me the checklist and the reasoning behind each item.

Example result: A 10-point checklist including "request the charter and the director's passport data," "verify the INN against the ФНС service 'Сведения о государственной регистрации ЮЛ/ИП'," and "check whether the contract specifies VAT and the counterparty's tax regime."

Prompt 4 — Calculating Tax Burden: УСН vs. Патент

Task: Compare two tax regimes for a given revenue profile.

You are a tax planning assistant for a self-employed entrepreneur in Russia.
Given: annual revenue <X> ₽, expenses <Y> ₽, region <region>, activity <type>.
Compare two regimes: УСН "доходы" (6%) and УСН "доходы минус расходы" (15%).
For each, show: tax base, tax amount, mandatory insurance contributions, and total burden.
Present as a table. State every assumption you make explicitly.
Remind me that regional rates may differ and that I must verify with my tax office.

Example result: A comparison table showing, for revenue 3,000,000 ₽ and expenses 2,000,000 ₽, that "доходы минус расходы" yields a lower burden in this scenario — with a clear note that the 15% rate is the default and regions may set different rates under Article 346.20 of the Tax Code.

Prompt 5 — Building a VAT Declaration Checklist

Task: Generate a pre-filing checklist so nothing is forgotten.

You are a VAT reporting assistant.
I am preparing a VAT declaration (НДС) for quarter <Q> <year>.
Give me a checklist of steps to complete before I file, covering:
- sales book and purchase book reconciliation,
- checking that all invoices received are reflected,
- verifying VAT on advances,
- confirming the declaration format is the current one approved by the ФНС.
Output as a numbered checklist with a one-line explanation per item.
Do not invent form codes — tell me to confirm the current format on nalog.gov.ru.

Example result: A 12-step checklist ending with "confirm the current declaration format and the electronic filing requirement on the ФНС website."

Prompt 6 — Drafting a Response to a ФНС Requirement

Task: Turn a dry requirement letter into a structured draft response.

You are an assistant helping draft a response to a tax authority requirement (требование).
Below is the text of the requirement.
Produce:
1. A plain-language summary of what the ФНС is asking for.
2. A list of documents I need to gather.
3. A draft cover letter in formal Russian business style, referencing the requirement number and date.
4. A deadline reminder: requirements typically specify a response window — state it if present, otherwise flag it.
Do not fabricate facts about my business. Leave placeholders like [сумма] where I must insert data.
Requirement text:
<paste>

Example result: A summary ("ФНС requests explanations on a VAT discrepancy between my sales book and my counterparty's purchase book for Q2"), a document list, and a draft letter with placeholders ready to fill.

Prompt 7 — Explaining a Discrepancy Between Two Books

Task: Diagnose why your sales and purchase data do not match.

You are a VAT reconciliation assistant.
I have a discrepancy: my sales book shows <amount> ₽ for <counterparty>,
but their purchase book shows <amount2> ₽.
List the most common reasons for such a discrepancy in order of likelihood,
and for each, tell me what to check in my documents.
Do not assume which one is correct — give me a diagnostic order.

Example result: Reasons ordered as: invoice not reflected by counterparty, wrong period, amount typo, VAT rate mismatch, corrected invoice (корректировочный счёт-фактура) not accounted for — each with a concrete check step.

Prompt 8 — Categorizing Bank Statement Transactions

Task: Auto-classify transactions for bookkeeping.

You are a bookkeeping assistant.
Below is a CSV of bank transactions: date, counterparty, description, amount, direction.
Assign each transaction a category from this fixed list:
[Revenue, Supplier payment, Salary, Taxes, Rent, Utilities, Bank fees, Owner withdrawal, Other].
Output the original CSV with one added column: Category.
If you are unsure, use "Other" and add a note in a separate column.
Do not merge or split rows.
CSV:
<paste>

Example result: The CSV returns with a Category column populated; anything ambiguous is flagged as "Other" with a note like "possible advance from client — confirm."

Prompt 9 — Generating a Payment Schedule from Contract Terms

Task: Turn contract payment terms into a dated schedule.

You are a cash-flow assistant.
Contract terms: total <amount>, payment stages <describe>, start date <date>.
Build a payment schedule table:

| Stage | Trigger | Amount | Expected date | Cumulative received |
State assumptions about working days and holidays explicitly.
Do not guess legal payment deadlines — only use the terms I give you.

Example result: A schedule with three stages and cumulative totals, plus a note that dates assume no shifts for public holidays and should be adjusted.

Prompt 10 — Drafting an Accounting Policy Summary

Task: Produce a short, readable accounting policy (учётная политика) summary for an ИП.

You are an accounting assistant.
Draft a short accounting policy summary for a self-employed entrepreneur.
Cover: tax regime, method of recognizing income and expenses, document storage period,
who is responsible for signing primary documents.
Keep it under one page. Use placeholders [ ] for values I must fill in.
Remind me that the full policy must comply with current legislation and be approved by me.

Example result: A one-page draft with placeholders, ending with a reminder to align it with Federal Law No. 402-FZ and current ФНС guidance.

Prompt 11 — Extracting Data from a Photo of a Receipt

Task: Turn a photographed receipt into structured data (when you use an OCR-capable model).

You are an OCR + accounting assistant.
From the image or text below, extract:

| Date | Seller | INN | Items | Total | VAT |
If the image is unclear, list what you could not read instead of guessing.
Receipt:
<image or pasted OCR text>

Example result: A table with the seller's INN and total, plus a note: "VAT line illegible — verify against the original."

Prompt 12 — Summarizing a Quarter for the Owner

Task: Turn raw numbers into a plain-language owner's report.

You are a financial assistant to a micro-business owner.
Given this quarter's data: revenue, expenses by category, taxes paid, cash balance.
Write a short summary (max 200 words) covering:
- how revenue and expenses moved versus the previous quarter,
- the largest expense category,
- one thing the owner should watch next quarter.
Use plain language. Do not give investment advice.
Data:
<paste>

Example result: "Revenue grew modestly while supplier costs rose faster, compressing margin. The largest category was supplier payments. Watch the cash buffer before the next VAT payment."

Prompt 13 — Preparing Questions for Your Accountant or Tax Office

Task: Turn your confusion into a precise question list.

You are an assistant helping a business owner prepare questions.
Here is my situation: <describe>.
Draft 5 precise questions to ask my accountant or tax inspector,
ordered from most to least urgent.
Each question should be specific enough to get a yes/no or a number as an answer.
Do not answer the questions yourself.

Example result: "Can I deduct this expense under УСН 'доходы минус расходы' if the document is dated in the next quarter?" — a question that actually gets a useful answer.

Prompt 14 — Checking a Contract Clause for Tax Risk

Task: Flag clauses that could create tax problems.

You are a contract-review assistant.
Below is a contract clause.
Identify any language that could create tax risk (e.g., unclear VAT treatment,
missing payment terms, vague acceptance criteria).
For each issue, explain the risk in one sentence and suggest a rewrite.
Do not provide legal advice — recommend consulting a lawyer for final wording.
Clause:
<paste>

Example result: "The clause says 'price includes all taxes' but does not specify the VAT rate — risk of dispute. Suggested rewrite: state the rate explicitly or 'VAT not applicable.'"

Prompt 15 — Building a Year-End Filing Calendar

Task: Create a deadline calendar you can actually follow.

You are a compliance calendar assistant for a micro-business.
My regimes: <list, e.g., УСН, ИП with employees>.
Build a table of recurring reporting and payment deadlines for the year:

| Obligation | Frequency | Typical deadline | Notes |
Do not hard-code exact dates — state the rule (e.g., "by the 25th of the month following the quarter")
and tell me to verify exact dates on nalog.gov.ru, since deadlines for some obligations shift.

Example result: A table covering quarterly advance payments, annual declarations, and insurance contributions, each with the rule-based deadline and a verification note.

Where These Prompts Break Down

No prompt collection is honest without its limits. Three things you must always do yourself:

• Verify every number against the source document. The model can misread a digit in a blurry invoice.
• Check the law. Tax rates, thresholds, and deadlines change. Article 346.20 of the Tax Code and the reporting formats on nalog.gov.ru are your ground truth.
• Keep an audit trail. Save the source documents and the model's output together, so you can explain any figure if asked.

A Realistic Workflow

A workable routine for a micro-business looks like this: at the end of each week, paste invoices and acts into Prompts 1 and 2; run Prompt 8 on the bank statement; once a quarter, run Prompts 4 and 5 before payments and filings; and keep Prompts 6 and 13 ready for when the tax office writes. The prompts do not replace judgment — they remove the copy-paste-and-retype layer that eats your evenings.

Final Thoughts

The point of these 15 prompts is not to eliminate your accountant. It is to eliminate the drudgery that makes accounting feel impossible to do yourself. Parsing invoices, matching acts, drafting responses, and comparing tax regimes are all tasks where a well-structured prompt turns two hours into twenty minutes. Start with one prompt — the invoice parser (Prompt 1) is the easiest win — and build from there. And whenever a number matters, verify it twice: once against the document, once against the law.

If you want to go deeper into prompt engineering for business workflows, the ASI Biont blog has related collections on automation and no-code AI agents you can pair with these accounting prompts.