Fake Bank Statements on Rental Applications: A Checklist to Spot AI Fakes

Check that the documents agree with each other and with the source. Add up the statement, match the payslip's deductions to its gross pay, compare dates and employer details, then verify directly with the bank or employer, or ask for open-banking access. Never reject an applicant on an AI detector score alone.

The reason this is harder than it was: a generator can now produce a payslip or bank statement with the right fonts, logos and believable shop names in seconds. Appearance is no longer evidence. What a fake struggles to do is stay consistent across several documents and survive a call to the real source. The checklist below goes in the order I would run it, cheapest first.

Follow me on Instagram@sagnikteaches

Two safeguards apply at every step. First, apply the same checks to every applicant, so your process is fair and you can show it. Second, treat every flag as a reason to ask a question, not a verdict. Genuine documents look odd more often than you would think.

Connect on LinkedInSagnik Bhattacharya

Start with what the applicant actually sent

Before analysing a single number, look at how the file arrived. Four quick points take two minutes.

Subscribe on YouTube@sagnikteaches

• File type. A bank statement normally arrives as a PDF downloaded from online banking. A screenshot, a phone photo of a screen, or a Word or image file is weaker. Ask for the original download.
• Document properties. Open the PDF's properties. A statement should usually show a bank or system as the producer. A creation date after the statement period ends is expected, but one created today for a statement six months old, with an editor or design tool listed as the producer, deserves a question. Metadata can be stripped or faked, so it only prompts a question.
• Completeness. Ask for three consecutive months, not the best one. A fabricated set is more work to keep consistent over three months.
• Format drift. Compare the layout of month one and month three. A real statement keeps the same columns, fonts and spacing. Generated sets sometimes drift.

Arithmetic checks that a fake often fails

This is the most useful step, and it needs only a calculator or a spreadsheet. If you want to automate it, our guide to using ChatGPT with Excel shows how to build the checks into a sheet.

Bank statements

1. Opening plus credits minus debits equals closing. Do this for each month. Generators often get individual lines right and the totals wrong.
2. Each month's opening equals the previous month's closing. A break here is the single most common giveaway.
3. Running balance. If the statement shows a balance after each line, spot-check ten lines. Small slips appear where a line was edited by hand.
4. Salary pattern. The payroll credit should land on a regular date with a regular reference, and the net amount should equal the payslip's net pay.
5. Spending shape. A person earning a stated income normally shows rent, utilities, groceries and small irregular spending. A statement with a large salary and almost no everyday transactions is suspicious. So is one where every transaction is a round number.
6. Dates and weekdays. Check that dated transactions fall on days the bank was open to process them, and that the order is chronological.

Payslips

1. Gross minus deductions equals net. Re-add every line.
2. Year-to-date figures climb sensibly. Across three payslips, the year-to-date totals should rise by roughly each month's amounts.
3. Deductions are plausible for the pay. Tax and pension lines should scale with gross pay. A deduction that stays identical when gross pay changes is a flag.
4. Pay frequency and date. Pay dates should follow the employer's pattern.
5. Employer details. Company name, address, and payroll reference should match the employment letter and the employer's public details.

A $3,900 monthly salary that does not add up

Illustration, with invented figures. A landlord is checking an application for a flat at $1,300 a month, so a common rule of thumb asks for income of about three times rent, or $3,900. The applicant sends three bank statements and three payslips.

The payslips show gross pay of $4,800 and net pay of $3,912 each month. The landlord re-adds the deductions: they total $888, which leaves $3,912, so the payslips reconcile. The statements show a payroll credit of $3,912 on the 28th each month, matching the net pay.

So far consistent. Then the opening and closing balances. Month one closes at $2,140. Month two opens at $2,410. That is a $270 gap with no explanation. The landlord adds up month two's lines and finds the closing balance does not equal opening plus credits minus debits either, being off by $270 in the other direction.

Next, the year-to-date figure on the third payslip. It shows $14,400, which is three months of gross pay at $4,800. But the same payslip is labelled pay period 10, and the employment letter gives a start date more than a year ago, so ten months at $4,800 should be about $48,000. Either the start date or the year-to-date figure is wrong.

The landlord does not accuse anyone. They write: "Month two's opening balance does not match month one's closing balance, and the year-to-date figure on the latest payslip looks low for your stated start date. Could you send the statements as downloaded from online banking, and ask your employer for a payslip copy?" That message is neutral and specific. A genuine applicant resolves it quickly, perhaps because a statement export combined two accounts. A forger tends to go quiet.

Ask the source, not the document

Consistency checks catch careless fakes. Careful ones pass. The reliable step is verification with the source.

• Employer verification. With the applicant's written consent, contact the employer using details you found independently, such as the company's published switchboard or website, not the number on the document. Ask only what you need: employment status, start date, and whether the stated salary band is right.
• Bank-side verification. Some banks let a customer generate a verified statement or share one directly with a third party. Ask the applicant to do this rather than emailing a file.
• Open banking. Where available, an applicant can give a regulated provider explicit consent to share their account data with you, choosing what is shared and for how long, and can withdraw that consent in the provider's app. You receive transactions straight from the bank rather than a file someone could edit. Several tenant-screening services offer this; check what the provider shares and how long it keeps the data before you rely on it.
• Previous landlord reference. A call to a landlord you found independently confirms the tenancy history and often the rent paid on time.

The tenant-screening choice is covered more widely in AI tools for landlords: lease checks and application fraud.

What AI detectors can and cannot do

Several tools claim to detect altered or generated documents. Used well, they save time. Used badly, they create unfair rejections. My rules:

• Treat the output as a prompt to look closer, never as the decision.
• Expect both mistakes. A genuine statement exported through an unusual app can be flagged, and a well-made fake can pass.
• Do not paste applicants' full documents into a general public chatbot. They contain account numbers and personal details. Check a tool's data handling first.
• Keep a note of which signals you found. "Detector said 83 per cent" is not a reason you could explain to the applicant, whereas "opening balance does not match prior closing balance" is.

An AI assistant is useful for a different job: doing the arithmetic for you. Here is a prompt for a reconciliation helper you can run on text you have first redacted.

Below are three months of bank statement lines in CSV (date, description, debit, credit, balance). Account numbers and names are removed.
1. For each month, compute opening balance + credits - debits and compare to the stated closing balance.
2. Check that each month's opening balance equals the previous closing balance.
3. List any lines where the running balance does not follow from the previous line.
4. List the payroll-style credits with their dates and amounts.
Return a table of mismatches only, with the row number. Do not guess whether any document is genuine.

Illustrative output: "Month 2 opening 2,410.00 does not equal month 1 closing 2,140.00 (difference 270.00). Row 31 balance 1,882.40 does not follow from row 30 (expected 1,812.40)." You would then check those exact rows against the original file yourself, since the model can misread columns.

Five more fake patterns and what to ask

The salary that is too round

A payroll credit of exactly $4,000.00 each month with no variation, for a job paid by the hour. Ask for the payslips that explain overtime or shift changes.

The deposit that appears just before application

A large one-off credit a week before the statement ends, lifting the balance. Ask about the source. Gifts and savings transfers are legitimate, but a lump sum with no history needs a plain explanation.

The employer that cannot be found

The company name on the payslip returns no website, no registration entry or a different address. Check the employer's name independently before contacting any number on the document.

The mismatched name

The account holder's name or address on the statement differs from the application. Joint accounts and name changes explain this innocently, so ask.

The identical formatting across different banks

Two statements from different banks, or a statement and a payslip, share the same odd font or spacing. Real documents from different institutions look different.

A scoring sheet so one odd detail does not decide

To avoid rejecting people for one quirk, score what you find. This is a simple internal tool, and you adjust the weights.

Notice that no row is "detector says fake". The sheet rewards checks you can explain. Whatever the total, the outcome is a request for more verification, not an accusation.

A second case: the self-employed applicant

Payslips do not exist for a freelancer, so the checks shift. Illustration: a self-employed designer applies and sends six months of business account statements and a one-page income summary. The summary claims $5,200 a month.

The landlord adds up the credits across the six statements and finds $27,300, an average of $4,550. Several large credits come from one source and are all round numbers. Instead of rejecting, the landlord asks for three things: invoices behind the three biggest credits, the most recent filed tax return summary, and, if the applicant is willing, open-banking access to see the full account history. The applicant provides invoices and a return that match the credits, and the average over a longer period supports the rent. The summary was simply optimistic, not fake.

The lesson is that an inflated claim and a fabricated document are different problems. Compare the claim to the evidence, then decide how much weight the stated figure deserves. Often the right result is to assess affordability on the verified average rather than the claimed one.

When a genuine applicant looks like a fake

Some honest patterns trip the checks, and a fair process allows for them.

• Exports that merge accounts. A statement downloaded from a banking app that combines accounts can break the opening-and-closing balance link. Ask for the per-account version.
• Pay-day changes. A payroll date that moves around a weekend or public holiday is normal. Compare the amounts, not only the date.
• A new job. Low year-to-date figures are expected when someone started recently. Ask for the contract or offer letter, and check the start date against the payslip.
• Cash-heavy work. A hospitality worker with tips may have irregular credits. Ask for the employer's confirmation of average earnings.
• Privacy-minded applicants. Some people redact merchant names on statements. Redacting spending lines is fine; changing balances or credits is not. Explain which parts you need unredacted: dates, opening and closing balances, and the payroll credits.
• Applicants relocating from abroad. Statements may be in another format or language, and verification through familiar routes is harder. Allow an alternative, such as a guarantor or a larger deposit where the law permits, instead of treating unfamiliar documents as suspect.

Message templates you can adapt

Keep the tone neutral. These are illustrative and use placeholders.

Subject: A quick check on your application

Hi [first name],

Thanks for your application. Before we go further I need to verify your income. Could you send:
1. The three most recent statements as downloaded from your online banking (PDF, not screenshots).
2. Your latest three payslips.
3. Your consent for us to confirm your employment with your employer.

If you would rather share your account data directly through an authorised open-banking provider, tell me and I will send the link.

Many thanks,
[your name]

Subject: Following up on your documents

Hi [first name],

I could not match the opening balance on the second statement to the closing balance on the first. This often happens with exports that combine accounts. Could you re-download that month for the single account that receives your pay?

I will hold the application for [number] days while we sort this out.

Thanks,
[your name]

If a draft like these needs rewording, an AI assistant is fine for polishing tone, provided you remove applicant details from anything you paste. For formal letters later in the tenancy, the checks are stricter; see using ChatGPT for tenant notices and messages.

Keeping the records without hoarding data

Application documents are sensitive. A small landlord can keep this tidy with four habits.

1. Store documents in one access-controlled folder, not in email attachments and personal chat threads.
2. Record the outcome of each check in a single line per applicant: date, checks run, result, who ran them. Do not copy account numbers into notes.
3. Set a deletion date when you collect the documents. For unsuccessful applicants, delete soon after the decision unless a rule requires otherwise. Ask a qualified adviser what applies to you.
4. Delete redacted copies you made for AI tools as soon as the check is done.

That record is also your defence if a rejected applicant asks why. A line saying "balances did not reconcile; applicant asked for original downloads; none received" is clear and fair.

Three quick sums that make fakes stand out

Illustration, with invented numbers, for the quick mental maths that needs no software.

• Net-to-gross ratio. If a payslip shows gross pay of $4,800 and net of $4,500, deductions are only 6 per cent. For most employees on that pay that is too low, so ask why. A fake often forgets that deductions scale with pay.
• Rent against the real credit. Rent of $1,300 against a verified net payroll credit of $3,912 is about a third of take-home pay. Compare affordability with the credit that actually reached the account, not the gross figure printed on a payslip.
• Balance against spending. A statement showing a $9,000 average balance but no savings transfers, interest or large purchases in six months is unusual. Ask where the money sits and how it got there.

None of these proves anything alone. Together with the reconciliation steps and a call to the source, they let you reach a fair decision quickly, and they keep you away from the risky habit of letting a single detector score make the call.

Handling the conversation and the paperwork

Put your verification steps in the application form so applicants know them in advance: what you check, that you may contact their employer, and how long you keep the documents. That is fair and also puts off some fakers. Tell the applicant what you could not verify, and give them a way to supply another proof. If you decline, make the reason about the unverifiable income, which you can document, and apply the same standard to everyone. Rules on tenant screening and data protection differ widely, so take advice on what you may check and how long you may keep it. If you suspect deliberate forgery, stop corresponding on the details and ask a qualified adviser how to report it.

Finally, keep the process light enough that you will use it. The first three checks, which are file type, reconciliation and payslip arithmetic, take about fifteen minutes and catch the lazy fakes. Verification with the source takes a little longer and catches the rest. If you handle many applications, the enquiry-screening tutorial shows how to filter early so fewer documents need this treatment.

Questions landlords ask about faked tenant documents

Can an AI detector tell me if a PDF is fake?

Not reliably enough to act on alone. Detectors give probabilities and can flag genuine documents or miss good fakes. Use them as one signal beside arithmetic checks, metadata, and verification with the source. Never reject an applicant on a detector score by itself.

Is a clean-looking statement proof it is real?

No. Good generators produce tidy layouts, correct fonts and plausible merchants. Realness comes from the figures reconciling and from verification with the bank or employer, not from how the document looks.

What if the applicant refuses to share bank access?

Offer a second route, such as a verified payslip request to the employer or an alternative proof of income, and apply the same rule to every applicant. A refusal alone is not evidence of fraud, since some people have privacy concerns.

Should I accuse the applicant of forging documents?

No. Say that you could not verify the income, name the specific item, and invite another form of proof. Keep a note of what you found. If you strongly suspect fraud, take advice on reporting it rather than confronting the applicant.

Do I need consent to run these checks?

Applicants are normally asked to consent in the application form to verification of the information they give. Data-protection law may apply to the documents you hold, so state in the form what you check and why, and delete documents when you no longer need them.

Further reads

• AI Landlord Updates and Owner Statements in Minutes — Keeping owners informed once a tenant is placed.
• Making Tax Digital for Landlords: Keep Rental Income Records With AI — Rental income records once the tenancy is running.
• AI Software for Landlords: Rent, Maintenance and Reminders Compared — Software options that include tenant screening.
• Can ChatGPT Draft a House Rental Lease? What Landlords Must Check — What to check when AI drafts a tenancy agreement.
• AI Assistant for Landlords: WhatsApp Maintenance Requests and FAQs — Handling tenant messages once the tenancy starts.
• Landlord Compliance Checklist: Track Gas Safety and EICRs With AI — Tracking certificates and deadlines after move-in.
• Tenant Referencing With AI: Document Checks and Fraud Signals — A referencing checklist for letting agents: AI consistency checks, metadata red flags, verification calls, and a rule that nothing is auto-rejected.
• Can AI Draft a Tenancy Agreement After the Renters' Rights Act? — AI can draft the first version of a periodic tenancy agreement, but it will not know the new rules. A checklist of what to verify before you sign.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general fraud-prevention practice for tenant screening; open-banking consent described from an open-banking industry explainer, read November 2026; no vendor claims relied on. Confirm local screening and data-protection rules with a qualified adviser.