Demand letters, answered

What belongs in a demand letter, how the damages figure is actually derived, how limitation deadlines differ by state, and what a reinstatement quote or an FDCPA validation notice has to say. Every figure below comes from a real letter produced by this platform, published in full on the demo page.

01

Demand letters, generally

What the document is, what belongs in it, and how long it takes to produce.

How long does it take to write a demand letter?

On DemandAI Pro, minutes. Two letters captured from the production pipeline on 4 August 2026 took 72 seconds (a Virginia rear-end personal-injury demand) and 31 seconds (a mortgage reinstatement demand) end to end, and both timings include the independent audit pass that re-checks every number. Those are two captured runs, not a benchmarked average — the time scales with how much source material you upload.

Drafted by hand, the same document is a multi-hour task, and almost none of those hours are spent writing prose. They go on assembling a dated treatment chronology from scattered records, totalling itemised bills, computing lost wages from a pay rate and a date range, and then proofreading arithmetic that a single transposed digit can undo.

Both captured letters are published in full, with their figures and audit results, on the demo page. No signup is required to read them.

What should a demand letter include?

Eight things: an identifying header carrying the claim number, date of loss, insured and claimant; a short introduction stating who you represent and that this is a formal demand; a statement of facts; a liability section arguing duty, breach and causation; the injuries with a dated treatment chronology and the charge for each encounter; an itemised damages breakdown that separates economic from non-economic losses; the demand figure with its arithmetic shown rather than asserted; and a response deadline with an enclosures list.

  • Show the arithmetic. An adjuster who can follow every addition from the bills to the demand figure has nothing to push back on. A bare total invites a counter that starts from nothing.
  • Date every encounter. Gaps in treatment are the first thing a carrier looks for; an unbroken chronology from the date of loss forward pre-empts the argument that the injury came from somewhere else.
  • Separate economic from non-economic. They are argued differently — one from receipts, the other from severity and prognosis.
  • Name what is missing. Where the source documents are silent, a bracketed placeholder is stronger than a guess. The sample personal-injury letter leaves the property-damage figure as [Attorney to insert documented property-damage amount] because no such figure appeared in the uploaded records.
  • Set a deadline. The sample letter demands a good-faith response within 30 days and states the limitation deadline behind it.

What is the difference between economic and non-economic damages?

Economic damages are the losses that come with a receipt: medical bills, lost wages, property damage. Non-economic damages are the losses that do not: pain, suffering, inconvenience, loss of enjoyment of life. In the sample letter on our demo page, economic damages were $29,230 — $15,370 in medical specials plus $13,860 in lost wages — and non-economic damages were $76,850, producing a $106,080 demand.

EconomicNon-economic
Proved byBills, invoices, pay recordsRecords, prognosis, testimony
Typical itemsMedical specials, lost wages, property damagePain, suffering, inconvenience, loss of enjoyment
How it is computedAddition of documented amountsUsually a multiple of medical specials
In the sample letter$29,230$76,850

Do AI-drafted demand letters get accepted by insurers?

We do not publish an acceptance rate, because we have not measured one — and you should be sceptical of any vendor that quotes you one without a methodology. What a carrier evaluates is the evidence, not the drafting tool: whether liability is documented, whether the treatment records support the injuries claimed, and whether the arithmetic holds up when an adjuster recomputes it. The letter is produced for a lawyer to review, sign and send; the platform never sends anything.

The failure mode worth guarding against is not a clumsy sentence — it is a fabricated citation. Every generation carries an instruction forbidding the model from inventing case names, reporter citations, docket numbers or statute sections. Where controlling authority belongs, it writes a placeholder instead:

[Counsel to cite controlling authority]

A placeholder is an inconvenience. A citation to a case that does not exist is a professional-conduct problem. The platform is built to produce the first.

02

Money and calculations

How the numbers in a settlement demand are actually derived, and who checks them.

How is a personal injury settlement calculated?

The common pre-suit method has two halves: total the economic damages from the documents, then estimate non-economic damages by multiplying the medical specials by a factor reflecting injury severity, and add the two together. In the sample letter on our demo page, $15,370 of medical specials at a 5x multiplier gave $76,850 in non-economic damages, which added to $29,230 in economic damages for a $106,080 demand. The multiplier is a negotiating convention, not a rule of law — no statute sets it.

The full chain from the sample letter

StepCalculationResult
Medical specials4,280 + 2,150 + 4,640 + 650 + 3,200 + 450$15,370
— of which physical therapy32 sessions x $145$4,640
Lost wages, full absence40 hrs x $38.50 = $1,540/wk x 6 wks$9,240
Lost wages, reduced schedule20 hrs x $38.50 = $770/wk x 6 wks$4,620
Economic damages15,370 + 13,860$29,230
Non-economic damages15,370 x 5$76,850
Demand29,230 + 76,850$106,080

What moves the multiplier

The sample letter argues for 5x rather than a lower factor on specific, documented grounds: an MRI-confirmed L5-S1 disc herniation with nerve-root impingement, 32 physical-therapy sessions that failed to resolve it, an invasive epidural steroid injection, and a treating orthopedist's prognosis of chronic intermittent symptoms with possible future intervention. It distinguishes those facts from the 3x (minor) and 4x (moderate) soft-tissue cases explicitly.

Objective imaging plus failed conservative care is what separates a high multiplier from an aspirational one. A multiplier asserted without that record is the first thing an adjuster discounts.

What is the independent audit pass?

It is a second, separately instructed generation step. Once the draft exists, the platform makes a fresh call carrying the original source documents and the draft, governed by its own auditor instructions and a strict 14-field JSON schema, and tells it to re-derive every dollar figure from the documents rather than trust what the draft claimed. It returns whether the arithmetic verified and a list of any discrepancies. If it found one, a reconcile pass runs once to correct the letter.

What “independent” means here, precisely

It is the same model in a separate, differently-instructed pass — not a second vendor's model, and we will not describe it as one. The independence comes from isolation of the task: the auditor is told to derive each figure from the source documents first, so it is checking the documents against the letter rather than checking the letter's arithmetic against itself.

What happens when the audit cannot run

The two engines behave differently, and the difference matters if you are relying on the flag:

EngineWhen figures are reported as verifiedIf the audit fails
Mortgage servicingOnly when the audit ran and affirmed every stated figure.An explicit discrepancy is injected telling the reviewer the figures were not verified.
Personal injuryThe arithmetic result is reported separately from whether the audit completed.The draft is returned unchanged and flagged as unverified.

Read the arithmetic result, not merely the presence of an audit. A completed audit and a verified figure are two different statements, and the mortgage engine is the stricter of the two.

What does the demand strength score mean?

It is a 0-100 score computed from the audit pass’s own structured output rather than from a separate opinion, so it moves with what the auditor could actually substantiate in the source documents. The labels are Strong at 75 and above, Solid from 55 to 74, Developing from 40 to 54, and Weak below 40. The sample personal-injury letter on our demo page scored 89, which is Strong.

ScoreLabelReading
75-100StrongFigures verified against the documents; damages well substantiated.
55-74SolidSupportable, with gaps a reviewer should close before sending.
40-54DevelopingMaterial documentation missing; expect the figures to move.
0-39WeakNot ready to send. Go back to the records.
03

Jurisdiction and deadlines

Which states the platform reasons about, what it encodes for each, and what it deliberately refuses to state.

Which states does the jurisdiction engine cover?

Twenty-three US jurisdictions: 22 states plus the District of Columbia — California, Texas, Florida, New York, Illinois, Pennsylvania, Ohio, Georgia, North Carolina, Michigan, New Jersey, Virginia, Washington, Arizona, Massachusetts, Tennessee, Indiana, Missouri, Maryland, Colorado, Wisconsin, Minnesota and DC. It is not all 50 states. If you enter a jurisdiction outside that list, the platform adds no jurisdiction context to the draft at all rather than guessing at rules it does not hold.

For each covered jurisdiction the platform holds the applicable negligence rule, the general personal-injury limitation period, the property-damage limitation period, whether a non-economic damages cap exists, and whether punitive damages are capped.

What it will not tell you, on purpose

It records only whether a cap exists, never a dollar amount, and the drafting prompt instructs the model not to state a cap figure. Cap amounts are amended, indexed and struck down often enough that a stale number in a letter is worse than no number. The same reasoning applies to state foreclosure timelines on the mortgage side, which are emitted as placeholders rather than asserted.

What is contributory negligence, and which states still use it?

Contributory negligence bars a claimant from recovering anything if they bear any fault at all — even one percent. Four of the jurisdictions the platform covers apply it: Virginia, Maryland, North Carolina and the District of Columbia. Every other covered jurisdiction uses a comparative rule instead, which reduces recovery by the claimant’s share of fault rather than eliminating it.

RuleEffectCovered jurisdictions
ContributoryAny claimant fault bars recovery entirely.VA, MD, NC, DC
Pure comparativeRecovery is reduced by the claimant’s fault share, even if they are mostly at fault.CA, NY, WA, AZ, MO
Modified comparative, 50% barBarred once the claimant is 50% or more at fault.GA, TN, CO
Modified comparative, 51% barBarred once the claimant is 51% or more at fault.TX, FL, IL, PA, OH, MI, NJ, MA, IN, WI, MN

The rule changes how the liability section has to be written. In a contributory jurisdiction the letter must close off any theory of claimant fault rather than merely outweigh it — which is why the sample Virginia letter emphasises that the claimant was stopped at a controlled intersection, struck from behind, and received no citation while the other driver was cited for following too closely.

How does the statute of limitations affect a demand letter?

It sets the outside date by which suit must be filed, and that date is the leverage behind the demand — a carrier reads a letter very differently when the window is closing. The platform computes a deadline from the date of loss and the jurisdiction’s limitation period, then flags it as expired, urgent at 90 days or fewer, approaching at a year or less, ok, or unknown. In the sample letter, a Virginia date of loss of 12 April 2026 produced a 12 April 2028 deadline with 616 days remaining at the time of capture.

Limitation periods vary more than most people expect. Across the covered jurisdictions the general personal-injury window runs from one year in Tennessee to six years in Minnesota — a five-fold spread that decides whether a file is urgent or routine.

Every limitation output the platform produces carries the same disclaimer, and it is there for a reason: general reference only, not legal advice. Limitation periods have exceptions — the discovery rule, minors, government-claim notice periods and wrongful death — and the exact deadline has to be confirmed against current law and the facts of the specific matter.

04

Mortgage servicing and borrower notices

The servicer-side letter types, and the consumer-protection rules that shape them.

What is a reinstatement demand?

A reinstatement demand, also called a reinstatement quote, is the itemised amount a borrower must pay to cure a default, bring the loan current and stop a pending foreclosure. It is good only through a stated date, because interest keeps accruing at a daily per-diem rate. In the sample on our demo page the itemised amounts supplied come to $9,127.76, with a $47.42 per-diem running against an unpaid principal balance of $271,438.22. That is a subtotal, not the final payoff: the letter leaves foreclosure attorney fees and the rescission recording fee as placeholders for the servicer to confirm.

The itemisation from the sample letter

ComponentCalculationAmount
Past-due principal and interest4 x $1,772.14$7,088.56
Past-due escrow (taxes, hazard insurance)4 x $412.60$1,650.40
Late charges4 x $87.20$348.80
Property inspection fees2 x $20.00$40.00
Foreclosure attorney fees and costsServicer to confirm
Recording fee to rescind the foreclosureServicer to confirm
Less suspense / unapplied balance($0.00)
Subtotal of itemised amounts supplied$9,127.76

The letter does not present $9,127.76 as the amount due. Two rows are left as placeholders because the source documents did not supply them, so the total is stated as the subtotal plus those amounts once the servicer confirms them. Leaving the gap visible rather than guessing at a number is the intended behaviour.

A partial payment does not reinstate the loan. Because attorney fees, costs and advances can continue to accrue, a quote that has passed its good-through date has to be reissued rather than paid.

What is the difference between a reinstatement quote and a payoff demand?

A reinstatement quote is the amount needed to cure the arrears and put the loan back on schedule — the borrower keeps the mortgage and keeps paying it. A payoff demand is the amount needed to extinguish the loan entirely: unpaid principal, accrued interest, escrow and fees. Reinstatement is typically a few thousand dollars, the arrears only; a payoff is the whole remaining balance.

Reinstatement quotePayoff demand
PurposeCure the default, de-accelerate, stop foreclosureRetire the loan in full
Headline figureArrears plus fees, with a per-diemPrincipal, interest, escrow and fees
Typical sizeThe missed paymentsThe remaining balance
After paymentThe mortgage continuesThe mortgage is discharged
In the sample letter$9,127.76Not applicable

What is FDCPA debt validation?

Under 15 U.S.C. § 1692g, a debt collector’s initial communication must tell the consumer they have 30 days to dispute the debt in writing, and that on a written dispute the collector will obtain verification and mail it to them. Regulation F, at 12 C.F.R. § 1006.34, adds a required itemisation of the amount owed. Letters generated as FDCPA validation notices carry the mini-Miranda — that the communication is from a debt collector attempting to collect a debt — and the 30-day validation language verbatim.

The platform assumes the FDCPA applies by default rather than making the operator opt in, on the basis that an unnecessary disclosure is a far cheaper mistake than a missing one.

The RESPA side, in the same letters

  • A loss-mitigation invitation under Regulation X, 12 C.F.R. § 1024.39.
  • A single point of contact for the borrower to reach.
  • A referral to a HUD-approved housing counsellor.
  • The 120-day rule before a foreclosure referral, with no dual tracking.

Statutory citations of this kind are the one category the model is permitted to state, because they are fixed in the drafting guidance rather than recalled. Case names, reporter citations and docket numbers remain forbidden.

Which mortgage servicing letter types can the platform draft?

Six. A default notice, being the Paragraph 22 breach letter and notice of intent to accelerate; a notice of acceleration; a reinstatement demand; a payoff demand; an FDCPA validation notice; and a general demand or RESPA Regulation X early-intervention outreach. Each produces a different headline figure, and each carries its own drafting guidance rather than being a template with the title swapped.

Letter typeHeadline figure
Default notice / Paragraph 22 breachThe cure amount
Notice of accelerationThe full accelerated balance
Reinstatement demandReinstatement total, with per-diem and good-through date
Payoff demandPrincipal, interest, escrow and fees
FDCPA validation noticeItemised amount owed, per Regulation F
General demand / early-interventionDelinquency status and loss-mitigation options

Every mortgage draft closes with a mandatory notice: DRAFT — FOR COMPLIANCE AND LEGAL REVIEW ONLY. Amounts, dates, deadlines and disclosures must be verified and approved by the servicer and its counsel before a letter is finalised, dated or sent. These are not ready-to-mail documents and are not compliance sign-off.

05

The platform

Privacy handling, the medical-code dictionary, the API and what it costs.

How is client and borrower data protected before it reaches an AI model?

Personal identifiers are replaced with opaque tokens before any text is sent to a model, and restored only on the finished letter. The detector covers ten identifier types — email address, phone number, Social Security number, person name, date of birth, street address, credit card, medical record number, driver’s licence and account number — plus any unlabelled run of eight or more digits, which catches loan, reference and claim numbers that carry no obvious label. If the scrub cannot be verified clean, the request is rejected with a 422 rather than sent.

  • The token map is never persisted. It lives in memory for the duration of a single request and is used to re-identify the finished letter, then discarded.
  • Money, dates and percentages are deliberately not tokenised. They have to survive intact, or the audit pass could not do arithmetic on them — which is the whole point of the audit.
  • The gate fails closed. A residual identifier after the scrub rejects the request. It does not warn and proceed.

Stated plainly, because the industry tends not to: detection is automated and best-effort. Name and address matching in particular are heuristics, not guarantees. This is not HIPAA de-identification and we do not describe it as such.

Does the platform read CPT and ICD-10 codes in medical records?

It recognises the codes that appear most often in personal-injury records and translates them into plain English inside the letter: 36 CPT procedure codes and 24 ICD-10 diagnosis code families. Diagnosis matching is by code prefix, so a record showing S13.4XXA resolves to a sprain of the cervical spine — whiplash. Recognised codes are injected into the drafting prompt with their meanings attached.

Coverage is curated to what personal-injury treatment actually looks like: emergency department and office visits, CT, MRI and X-ray imaging, physical therapy, chiropractic care, trigger-point, epidural and facet injections, and the common surgical repairs — meniscectomy, rotator cuff repair, microdiscectomy, cervical fusion and total knee arthroplasty.

This is not full CPT or ICD-10 coverage, not a licensed code set, and not a billing source of truth. Descriptions are general references for drafting, and ICD-10 seventh characters can change the specifics of a diagnosis.

Is there an API?

Yes. Five POST endpoints — demand letters, case value, negotiation, future medical and mortgage demand — authenticated with a dl_-prefixed API key sent as either an Authorization: Bearer header or an X-API-Key header. The specification is published as OpenAPI 3.1 at /api/v1/openapi.json. There is no SDK, no OAuth flow and no GET or list endpoints: five generation endpoints, JSON only.

Limit or behaviourValue
Rate limit50 generations per hour per key across all v1 endpoints, or the key’s own hourly, daily or monthly limit — whichever is lower
On 429A Retry-After header is returned
Max inputdocumentsText is capped at 800,000 characters
Error shapeA uniform { error, message } envelope on 400, 401, 403, 429 and 500
422Returned specifically when the PII scrub could not be verified complete
Every responseCarries a piiProtected boolean reporting whether anything was tokenised for that request

Full endpoint reference and request examples are on the API documentation page. API access is included from the Professional plan upward.

How much does DemandAI Pro cost?

The free trial runs 14 days and requires no credit card. Paid plans are $127 a month for Starter with 10 letters a month, $247 for Professional with 50 letters a month and public REST API access, and $397 for Business with 100 letters a month. Litigation financing reports are included with Professional and Business plans, and expert witness recommendations are in beta.

PlanPriceDemand letters
Trial$0 for 14 daysNo credit card required
Starter$127 / month10 per month
Professional$247 / month50 per month, plus API access
Business$397 / month100 per month

Current plan details are on the pricing page.

Can homeowners use it without a lawyer?

Yes — there is a homeowner sign-in and a homeowner-facing letter flow alongside the attorney and servicer front doors. But every output is a draft for review, not a filing and not advice. Mortgage letters close with a mandatory DRAFT — FOR COMPLIANCE AND LEGAL REVIEW ONLY notice, and every limitation-period output carries an explicit general-reference-only disclaimer. For anything contested, a deadline that is close, or a dispute already in litigation, involve a lawyer.

The design assumption throughout is that a competent human reviews the document before it goes anywhere. That is why the platform leaves bracketed placeholders where the source documents are silent, refuses to invent legal authority, and reports its audit result rather than simply asserting the letter is correct.

06

How a demand letter is drafted and audited on DemandAI Pro

The four stages every letter passes through: document intake with identifier tokenisation, drafting, an independent audit that re-derives each figure from the source documents, and a reconcile pass before counsel review.

  1. 01

    Upload the source documents

    Add the medical records, itemised bills, wage documentation and incident or police report. The platform scans them for CPT and ICD-10 codes, and replaces personal identifiers with opaque tokens before any text is sent to a model.

  2. 02

    Draft

    The model writes the letter from the tokenised documents. Jurisdiction context — the applicable negligence rule and limitation period — and plain-English translations of the recognised medical codes are injected into the prompt alongside them.

  3. 03

    Audit

    A separate pass receives the source documents together with the draft and is instructed to re-derive every dollar figure from the documents rather than trust the draft. It returns a structured verdict: whether the arithmetic verified, and a list of any discrepancies found.

  4. 04

    Reconcile, then review

    If the audit reported a discrepancy, one correction pass rewrites the affected figures. The finished letter is returned with the audit note and a 0-100 demand-strength score for counsel to review, edit and sign.

Read a real letter

Both letters referenced throughout this page — the $106,080 Virginia personal-injury demand and the $9,127.76 reinstatement quote — are published in full, with their figures, audit results and drafting times. No signup.