Analyze Financial Disclosures for Inconsistencies
Turn an opposing party's financial disclosure and supporting records into an organized summary, a prioritized inconsistency list, and draft discovery requests and deposition questions.
When to use this
You represent a client in a divorce or support matter and the other side has produced a financial affidavit plus bank statements, pay stubs, or tax returns. You need to organize the numbers, spot where they don't add up, and decide what to chase in discovery. Claude does the first-pass extraction and cross-checking so you spend your time investigating leads instead of building spreadsheets by hand.
Opposing financial disclosures arrive as a pile of PDFs: a sworn affidavit, a few months of bank statements, last year's tax return, maybe some pay stubs. The story they tell is rarely consistent, but finding the gaps by hand is slow, eye-straining work. Claude can read all of it at once and lay the figures side by side.
This workflow walks one document set through a single Claude conversation: extract income, assets, debts, and expenses into a summary normalized to monthly and annual figures; cross-check the numbers across documents to surface inconsistencies, omissions, and red flags; then turn those findings into a prioritized discovery list and deposition questions. Each step builds on the last, so the context stays intact.
Treat everything Claude produces as a set of leads, not proof. An AI flag that reported income doesn't support these expenses is a reason to send an interrogatory or issue a subpoena, not a fact you put in front of the judge. Verify every figure against the source document; the analysis and the argument are yours.
The Workflow
Set up the analysis and orient Claude
Start a new conversation and give Claude the documents first, each wrapped in its own <document> tag with an id, then the case context. Upload the disclosure and supporting records (affidavit, bank statements, tax returns, pay stubs) as files, or paste the text. Keep all the documents in this one conversation so later steps can reference them by id.
Prompt<document id="AFF" type="sworn financial affidavit" date="[DATE]"> [UPLOAD OR PASTE] </document> <document id="BANK-1" type="[BANK] checking statements" period="[DATE RANGE]"> [UPLOAD OR PASTE] </document> <document id="TAX-[YEAR]" type="federal tax return with schedules"> [UPLOAD OR PASTE] </document> <document id="PAY" type="pay stubs" period="[DATE RANGE]"> [UPLOAD OR PASTE] </document> I am a family-law attorney representing [CLIENT ROLE, e.g., the wife] in a [divorce / child support / spousal support] matter in [STATE/COUNTY]. The documents above are the opposing party's financial disclosure and supporting records. My goal is to find where the disclosure does not reconcile to the records so I can target discovery. Before analyzing, confirm what you received: list each document by id with its date or date range, and note anything unreadable, any missing pages (e.g., statement page 2 of 4 absent), and anything ambiguous. Do not analyze yet; inventory what you have and ask me any clarifying questions.
What you get: A document-by-document inventory keyed to your ids, with dates and date ranges, a note of any unreadable or missing pages, and a few clarifying questions about scope or context.
Extract income, assets, debts, and expenses into a clean summary
Have Claude pull the raw numbers out of the documents and organize them into a structured financial picture, normalized so weekly, biweekly, monthly, and annual figures can be compared. Asking for a source citation on every figure, and showing the conversion math, is what makes this verifiable later.
PromptUsing only the documents above, build a structured financial summary of the opposing party with four sections: (1) INCOME: wages, self-employment, bonuses, investment, and any other income; (2) ASSETS: accounts, real estate, vehicles, retirement, business interests, with stated values; (3) DEBTS/LIABILITIES: each obligation with balance and monthly payment; (4) MONTHLY EXPENSES: itemized as claimed. Normalize every income and expense figure to both MONTHLY and ANNUAL amounts before presenting it, and show the math (e.g., "$1,950 biweekly x 26 / 12 = $4,225/mo"). State the period of the source figure so I can check the conversion. For every figure, cite the document id and the specific line, statement, or page it came from. If a number appears only on the affidavit with no supporting record, label it "self-reported, unverified." If you have to estimate or infer, say so explicitly. Present each section as a table.
What you get: Four labeled tables (income, assets, debts, expenses) with monthly and annual figures, the conversion math shown, a source citation on each figure, and clear "self-reported, unverified" or "estimated" tags where support is missing.
Cross-check for inconsistencies, omissions, and red flags
Now point Claude at the contradictions. The goal is to compare what the affidavit claims against what the bank and tax records show, using the normalized figures, and to flag the classic signals of underreporting: lifestyle that outruns stated income, accounts that appear in statements but not on the affidavit, large unexplained transfers.
PromptNow cross-check the documents against each other using the normalized monthly and annual figures from the previous step, and flag problems. Look for: (1) INCONSISTENCIES: figures that conflict across documents (e.g., affidavit income vs. tax return vs. pay stubs vs. deposits into the bank account); (2) OMISSIONS: accounts, income sources, or assets that appear in the bank or tax records but are missing from the affidavit; (3) RED FLAGS: monthly expenses or spending that exceed reported income, large or recurring unexplained transfers or withdrawals, payments to unknown third parties, signs of an undisclosed account or business. For each item: give it a short label, quote or state the specific conflicting figures with their document ids and pages, show the comparison math, and rate it High / Medium / Low concern. State a plausible innocent explanation where one exists, and note where the apparent gap could be caused by a missing page or period mismatch. Do not overstate.
What you get: A categorized findings list (inconsistencies, omissions, red flags), each with the conflicting figures, their sources, the comparison math, a concern rating, and a note where an innocent explanation or missing record is plausible.
Build a prioritized discovery and deposition list
Convert the findings into action. Claude drafts the follow-up document requests, interrogatories, and deposition questions, ordered by how much each issue matters to the case.
PromptTurn the findings into a prioritized action plan. Produce three lists: (1) DISCOVERY REQUESTS: specific documents or records to request to resolve each open question (e.g., full statements for an account that only partially appears, business records, credit card statements), ordered by priority; (2) INTERROGATORIES: written questions targeting the omissions and inconsistencies; (3) DEPOSITION QUESTIONS: a focused line of questioning for the top 3–5 issues, phrased the way I would ask them, including follow-ups if the witness denies or deflects, and noting the document id I should have ready as an exhibit for each line. Tie each item back to the specific finding it addresses. Flag any issue serious enough that I should consider a forensic accountant or a subpoena to a financial institution. Do not cite procedural rules; write [APPLICABLE DISCOVERY RULE] where a rule would govern.
What you get: Three prioritized lists (discovery requests, interrogatories, and deposition questions with exhibit references), each tied to a specific finding, plus a note on any issue that may warrant a forensic accountant or subpoena.
Verify every figure against the source before you rely on it
Do not file, serve, or cross-examine from Claude's output as-is. Open each source document and confirm every number, date, conversion, and citation the summary relies on: OCR misreads, transposed digits, and a statement Claude skimmed are all possible. Confirm the inconsistencies are real and not artifacts of a missing page or a misread balance. The disclosures may also be incomplete in ways Claude can't see; the leads here point you toward what to request next. The analysis, the legal judgment, and the work product are yours.
What you get: A verified, source-checked picture you can act on: confirmed discrepancies, corrected figures, and a discovery plan you've vetted, ready to convert into formal requests and deposition preparation.
Example Output
Illustrative example — names, facts, and figures are fictional.
INCONSISTENCY & RED-FLAG SUMMARY: Opposing Party [Party B] (fictional)
(Illustrative sample. Every figure must be verified against the source document.)
HIGH CONCERN
1. Income understated vs. deposits
- AFF p.2: $4,200/mo gross wages ($50,400/yr)
- BANK-1 Jan–Jun: deposits total $47,100 / 6 = $7,850/mo ($94,200/yr)
- PAY: $1,950 biweekly x 26 / 12 = $4,225/mo
- Gap of ~$3,625/mo not explained by pay stubs. Possible innocent explanation: transfers between own accounts. Lead, not proof.
2. Undisclosed account
- Two $2,000 transfers to "Acct ...4471" (BANK-1, Mar p.3 & May p.2)
- No account ending 4471 listed on AFF. Possible undisclosed account.
MEDIUM CONCERN
3. Expenses exceed reported income
- Claimed expenses $6,900/mo (AFF p.4) vs. net income ~$3,400/mo (PAY).
4. Business interest omitted
- TAX-2024 Schedule C ("[Business Name] Consulting", gross $41,000/yr = ~$3,417/mo)
- No business asset or income listed on AFF.
LOW CONCERN
5. Vehicle value: AFF lists $8,000; may be understated. Verify with a valuation.
TOP DISCOVERY ASKS: full statements for acct ...4471; 24 mos BANK-1 statements;
all consulting business books and records; credit card statements for lifestyle check.Tips
- •Upload the actual statements and tax returns, not just the affidavit. Claude can only catch a hidden account or understated income if it can see the records the affidavit is supposed to match.
- •Give Claude more months, not fewer. Six to twelve months of statements reveal recurring transfers and seasonal income that a single month hides.
- •Insist on normalized figures with the math shown. Comparing a biweekly pay stub to a monthly affidavit line by eye is where false discrepancies come from.
- •Make Claude cite a document id and page for every figure. If a number has no document behind it, that gap is itself a discovery target, and the citations are what let you verify the output fast.
- •Push back when a flag feels thin. Ask what innocent explanation there could be, so you walk into a deposition with the discrepancy and the likely defense already in mind.
A note on confidentiality
These documents are the opposing party's financial records: full account numbers, Social Security numbers on tax returns, employer details, and spending that may reveal medical or personal information. They may also be subject to a protective order or confidentiality stipulation in your case; check its terms before uploading them anywhere. Redact SSNs and mask account numbers to the last four digits before upload, and keep the analysis on a commercial plan (Team or Enterprise) whose terms exclude your content from model training. On Pro or Max, disable model training in Privacy settings; a commercial plan's contract terms are the stronger answer for client work. Your duties of confidentiality and competence apply (Model Rule 1.6; ABA Formal Opinion 512). Every output is a draft for attorney review: verify each figure against the source before relying on it.
Frequently Asked Questions
Can I trust Claude's inconsistency flags well enough to use them in court?
No. Treat them as leads, not evidence. Claude's output is a first-pass map of where the numbers may not add up, generated from documents that can have OCR errors, missing pages, or innocent explanations. Verify every flagged figure against the source document, then convert confirmed discrepancies into proper discovery, sworn testimony, or expert analysis you can put before the court.
What documents should I give Claude for the strongest analysis?
Pair the sworn financial affidavit with the records it's supposed to reconcile to: several months of bank and credit card statements, recent pay stubs, and the last year or two of tax returns including all schedules. The cross-checking only works when Claude can compare claimed figures against underlying records, so the more complete the supporting set, the more meaningful the flags.
Is it appropriate to put an opposing party's financial records into an AI tool?
It can be, with the right setup. Check any protective order first, use a commercial plan (Team or Enterprise) whose terms exclude content from training, or at minimum disable model training on Pro or Max, and redact SSNs and full account numbers before uploading. Your duties of confidentiality and competence still apply (ABA Formal Opinion 512 and Model Rule 1.6), and you remain responsible for verifying and using the output appropriately.
Prompts used in this workflow
Get New Workflows Like This Every Week
Join the free Claude for Lawyers newsletter — practical workflows, prompts, and guides for everyday practice.