Partners in Evaluation
Access to Justice was built by Vals AI with support from NextLadder Ventures.
Background
Most low-income Americans face their civil legal problems without a lawyer. According to the Legal Services Corporation’s 2022 Justice Gap study, low-income Americans did not get any or enough legal help for 92% of their civil legal problems, and nearly half (46%) of those who didn’t seek help cited cost as a reason. AI tools could help close this gap, but few benchmarks measure how well models handle the kinds of problems low-income people actually face, such as housing, employment, and family law.
This benchmark tests whether AI models can act as a reliable first point of contact for someone with a legal problem, the role an intake worker or a court self-help center plays. For this benchmark, we simulated conversations based on real legal-aid queries from legal aid organizations and criteria written by legal aid experts. Our experts outlined that a good first response asks about the person’s situation, gives accurate general legal information, and steers the person to the right local help when needed, such as a legal aid organization, a court self-help center or an attorney.
Key Takeaways
- No model clears the bar for a reliable first point of contact. We set that bar at 80% of expert rubric criteria; even the top-performing model, Qwen 3.8 Max, reaches 76.4%.
- Models explain the law but often fail to identify specific local help. Every model meets 78–87% of criteria for general explanations of the law, but only 30–50% for specific resources such as a local legal aid program or court self-help center. Instead, 21 of 23 models name a national directory, such as LawHelp.org, the ABA’s Find Legal Help or 211, in 70% or more of their final answers.
- More expensive models did not consistently score higher. Models like Claude Sonnet 5.5 and GPT-5.6 Luna, which people likely encounter when using products like Claude or ChatGPT without a paid account, performed about the same as their providers’ premium models. Only xAI’s premium model showed a clear edge.
Results
Overall Model Performance
Qwen 3.8 Max leads at 76.4% of expert-defined criteria met over all scenarios, 1.4 points ahead of Claude Sonnet 5.5 (74.9%), a gap within the margin of error. Meta Wonder Full follows at 74.7%. Most of the models meet between 67.1% and 73.8% of the rubric criteria.
Free-Tier vs. Premium Models
Many people seeking legal help will use whichever model a chatbot offers for free. In a March 2026 Epoch AI–Ipsos survey of 2,021 U.S. adults, 84% or more had no paid subscription to ChatGPT, Gemini, or Claude.
To approximate what these users get, we grouped the models from major chatbot providers by whether we believe each one powers the provider’s free tier or its paid plans. This grouping is our best estimate, not a confirmed mapping: providers don’t always disclose which model a chatbot uses, they can change it without notice, and their apps may use different settings than the API. We tested every model through its API with the same setup, so these results approximate what a chatbot user would see rather than reproduce it. Open-weight models are left out of this comparison.
| Provider | Free-tier | Premium |
|---|---|---|
| OpenAI | GPT-5.6 Luna (65.6%) | GPT-6.1 Sol (65.8%) |
| Anthropic | Claude Sonnet 5.5 (74.9%) | Claude Opus 5.5 (73.8%) |
| xAI | Grok 4.6 (71.3%) | Grok 4.7 (73.7%) |
| Gemini 3.8 Flash (67.1%) | Gemini 4 Argon (68.5%) |
The free-tier models average 69.7%, compared with 70.4% for the premium ones, a gap of 0.7 points. Within the same provider, the premium model isn’t consistently better: Anthropic’s free-tier Claude Sonnet 5.5 scores 1.2 points higher than Claude Opus 5.5, while the premium model leads by 2.4 points at xAI, 1.4 points at Google and 0.3 points at OpenAI.
Overall accuracy (%) of each provider's free-tier and premium model
Models’ Performance
Performance by Topic
Each scenario is tagged with one legal topic. The complete list of topics and what each covers is in the Methodology section below.
Because the scenarios start from real questions that people brought to legal aid practitioners, the topic mix reflects what those practitioners see rather than an even split across areas of law. Family law is the largest topic (110 of 252 scenarios), followed by housing (65). The chart below pools every rubric criterion across all 23 models for the scenarios in each topic.
Criteria satisfied (%), all models pooled
Models do best on housing and landlord-tenant questions (75.4% across all models). Family law, the largest topic, lands in the middle at 71.2%, but scores vary meaningfully within it.
Criteria satisfied (%), all models pooled, subfields with 3+ scenarios
Visitation (61.3%) and kinship caregiving with child protective services (63.0%) score lowest, and guardianship, divorce finances, family-member adoptions and domestic violence fall between 66% and 68%. Our experts point out that most of family law isn’t something a model can answer by reading a statute or a case: the right answer depends on the people involved, and there is rarely a one-size-fits-most answer.
Public benefits (5 scenarios) and the catch-all civil category (11 scenarios) are the hardest topics for models, though both are small. Our Public Benefits Bench evaluates in more depth how well models help people apply for and receive SNAP benefits.
Average criteria satisfied (%) by topic for the three highest- and three lowest-scoring models
The best- and worst-performing models have similar topic profiles, indicating that the failure modes of LLMs are relatively consistent. Across all 23 models, public benefits or civil is the lowest-scoring topic for 21, and housing is among the three highest-scoring topics for all 23.
There is one notable exception: Muse Spark 1.3 scores only 41.5% on criminal and records questions, because its provider’s content filter blocked it from answering five of them on every retry (questions about gun rights, expunging or sealing firearm records, and a trafficking conviction), and those blocked scenarios count as zero. Excluding them, it scores 71.5% on the other 11 criminal and records scenarios, in line with the 69.6% the other models average on the same scenarios. Muse Spark 1.3 Max was blocked on the same five and scores 73.2% on the rest.
Where Models Fall Short: Local Help
Models do best on general explanations of the law and the person’s options (78–87% of criteria met) and on deadlines and timing (65–81%). They do worst on pointing the person to specific help:
- Specific resources (a named court self-help center, a local legal aid program, a particular form): This is one of the two weakest categories for every model, at 30–50% of criteria met. Models often fall back on national directories instead: 21 of the 23 models name a national directory such as LawHelp.org, the ABA’s Find Legal Help or 211 in 70% or more of their final answers.
- Our experts note that accurate local resources, especially a local legal services organization or a courthouse volunteer program in the court where the person needs to file, are far more useful than national directories.
- For example, in the sample query highlighted below, the rubric expects a referral to Southeast Louisiana Legal Services, the legal aid program for the user’s region:
- Claude Sonnet 5.5 names it directly and meets the criterion: “Southeast Louisiana Legal Services, (504) 529-1000 or (877) 521-6242, is free if you’re income-eligible.”
- MiniMax M3.1 Flash instead lists directories, including the national Legal Services Corporation locator, the state bar’s referral service and a statewide legal-aid directory, and leaves the parent to work out which program serves them.
- Referral to an attorney (when the rubric says the person should be told a lawyer can help, or pointed to a lawyer referral service): This ranges from 42% (Gemini 4 Argon) to 73% (MiMo V2.6 Pro) of criteria met. Higher-scoring models tend to refer more often (correlation 0.58).
Every model could search the web and open pages. Across the 23 models, the runs issued 60,400 searches and tried to open 22,598 pages, of which 16,849 (75%) loaded. We count a source as read only if the model opened the page and it loaded.
Share of pages the model opened and read, by type of site. Search results the model never opened are not counted.
| Model | Pages read / scenario | Source mix |
|---|---|---|
| GPT-6.1 Sol | 12.1 | |
| Claude Sonnet 5.5 | 7.3 | |
| Kimi K3.1 Pro | 5.7 | |
| Claude Opus 5.5 | 5.1 | |
| Wonder Full | 3.8 | |
| MiniMax-M3.1-Flash | 3.7 | |
| GPT-5.6 Luna | 3.1 | |
| DeepSeek V4.1 Flash | 2.9 | |
| Wonder | 2.8 | |
| Qwen 3.8 Max | 2.4 | |
| DeepSeek V4.1 Flash 1017 | 2.4 | |
| Grok 4.7 | 2.2 | |
| GLM 5.3 | 1.9 | |
| Muse Spark 1.3 Max | 1.8 | |
| Grok 4.6 | 1.8 | |
| Gemini 4 Argon | 1.7 | |
| MiMo V2.6 Flash | 1.4 | |
| MiMo V2.6 Pro | 1.1 | |
| Muse Spark 1.3 | 0.9 | |
| GLM 5.3 Flash | 0.9 | |
| Kimi K3 | 0.9 | |
| MiniMax-M3 | 0.7 | |
| Gemini 3.8 Flash | 0.1 |
Most-read sites, all models
| law.lis.virginia.gov | Statutes, rules & case law | 2,842 |
| texas.public.law | Statutes, rules & case law | 1,164 |
| mgaleg.maryland.gov | Statutes, rules & case law | 953 |
| texaslawhelp.org | Legal aid & nonprofit self-help | 583 |
| womenslaw.org | Legal aid & nonprofit self-help | 552 |
| law.cornell.edu | Statutes, rules & case law | 489 |
| mdcourts.gov | Courts & court self-help | 468 |
| peoples-law.org | Legal aid & nonprofit self-help | 378 |
| secure.ssa.gov | Government agencies | 346 |
| legis.la.gov | Statutes, rules & case law | 335 |
| vacourts.gov | Courts & court self-help | 332 |
| statutes.capitol.texas.gov | Statutes, rules & case law | 295 |
| newyork.public.law | Statutes, rules & case law | 248 |
| lawserver.com | Law firms, commercial & other | 239 |
| courtrules.net | Statutes, rules & case law | 234 |
| lasc.org | Courts & court self-help | 212 |
| selfhelp.vacourts.gov | Courts & court self-help | 192 |
| courts.ca.gov | Courts & court self-help | 145 |
| consumerfinance.gov | Government agencies | 129 |
| irs.gov | Government agencies | 116 |
| rulesofcivilprocedure.com | Law firms, commercial & other | 88 |
| california.public.law | Statutes, rules & case law | 84 |
| probonodeskmanual.loyno.edu | Legal aid & nonprofit self-help | 84 |
| eeoc.gov | Government agencies | 82 |
| dcfs.louisiana.gov | Government agencies | 79 |
What models read is mostly primary law: statutes, court rules and case law make up 42% of pages read. Every other category makes up 15% or less, including legal aid and nonprofit self-help sites at 10%.
We find that models read enough law to explain it, but rarely open the local self-help, legal aid or court pages that would let them name the specific program or office a person should go to next. Models also have to include the local resource in the answer.
Our experts see this as a serious gap and a big opportunity for improvement. Legal information without a steer toward local help can leave a person worse off. They may think all they need to do is fill out a form, when sometimes that’s the worst thing they could do. Statutes and case law are the right sources for a lawyer, but a member of the public needs a local next step, and some states have guides for unrepresented clients that models could point to instead. Pointing people to specific court forms is riskier, because forms change often and court websites are frequently out of date.
Eliciting Context vs. Delivering Legal Information
Each scenario’s rubric has two kinds of criteria. Elicitation criteria (the rubric’s “Must find out” items, about 24% of criteria) cover the context a careful advocate must find out about the person’s situation. Legal information criteria cover what a good response should deliver. In this report, legal information means general information about the law, the person’s options and where to get help: the kind of information a non-lawyer first point of contact can give. It is not legal advice, which only a licensed attorney can give about a person’s specific situation.
For example, from the sample query below about a small claims court case over an unpaid verbal agreement:
- Elicitation criteria example: “Must find out: Was there a written contract regarding the debt?”
- Legal information criteria example: “Must explain that since there was no written contract, to win the case the client will need at least one witness to testify about their knowledge of the contract.”
Criteria satisfied (%) by criterion type, sorted by overall score
| Model | ■ Gathering context | ■ Legal information | Overall |
|---|---|---|---|
| Qwen 3.8 Max | 75.7% | 76.6% | 76.4% |
| Claude Sonnet 5.5 | 75.8% | 74.7% | 74.9% |
| Wonder Full | 79.6% | 73.1% | 74.7% |
| Claude Opus 5.5 | 69.4% | 75.1% | 73.8% |
| Grok 4.7 | 72.0% | 74.2% | 73.7% |
| GLM 5.3 | 71.7% | 74.3% | 73.7% |
| MiniMax-M3.1-Flash | 76.3% | 72.4% | 73.3% |
| DeepSeek V4.1 Flash 1017 | 70.7% | 73.8% | 73.1% |
| Wonder | 78.9% | 71.1% | 73.0% |
| DeepSeek V4.1 Flash | 69.6% | 73.9% | 72.9% |
| Muse Spark 1.3 | 76.3% | 70.9% | 72.2% |
| Muse Spark 1.3 Max | 75.0% | 71.3% | 72.2% |
| Grok 4.6 | 65.9% | 73.0% | 71.3% |
| GLM 5.3 Flash | 69.2% | 71.7% | 71.1% |
| MiMo V2.6 Flash | 73.4% | 69.9% | 70.7% |
| MiMo V2.6 Pro | 66.8% | 71.0% | 70.0% |
| MiniMax-M3 | 70.2% | 69.6% | 69.8% |
| Kimi K3.1 Pro | 74.8% | 67.8% | 69.5% |
| Gemini 4 Argon | 64.0% | 69.9% | 68.5% |
| Kimi K3 | 61.1% | 70.4% | 68.2% |
| Gemini 3.8 Flash | 66.1% | 67.3% | 67.1% |
| GPT-6.1 Sol | 68.1% | 65.1% | 65.8% |
| GPT-5.6 Luna | 69.7% | 64.3% | 65.6% |
On legal information, most models land between 67% and 77%, and the two OpenAI models score about 65%.
Part of this comes down to how models interact in the conversation. Each model can have up to five exchanges with the simulated client. Conversations average about 3.5 exchanges, and some hit the limit. Most models give a full answer in their first reply and only ask questions afterward, if at all.
Averages per scenario, sorted by overall score
| Model | Overall | Exchanges | First reply asks a question | Final answer words | Searches | Pages opened | Pages read |
|---|---|---|---|---|---|---|---|
| Qwen 3.8 Max | 76.4% | 3.2 | 23% | 3,899 | 7.5 | 3.5 | 2.4 |
| Claude Sonnet 5.5 | 74.9% | 3.3 | 10% | 1,641 | 13.0 | 9.3 | 7.3 |
| Wonder Full | 74.7% | 3.4 | 16% | 2,932 | 15.6 | 4.9 | 3.8 |
| Claude Opus 5.5 | 73.8% | 3.3 | 8% | 1,827 | 12.1 | 6.2 | 5.1 |
| Grok 4.7 | 73.7% | 3.6 | 2% | 1,653 | 16.5 | 2.9 | 2.2 |
| GLM 5.3 | 73.7% | 3.4 | 2% | 1,901 | 7.2 | 2.8 | 1.9 |
| MiniMax-M3.1-Flash | 73.3% | 3.4 | 14% | 3,433 | 8.2 | 5.1 | 3.7 |
| DeepSeek V4.1 Flash 1017 | 73.1% | 3.4 | 10% | 2,054 | 11.2 | 3.6 | 2.4 |
| Wonder | 73.0% | 3.5 | 7% | 2,410 | 14.7 | 3.7 | 2.8 |
| DeepSeek V4.1 Flash | 72.9% | 3.5 | 6% | 1,655 | 9.9 | 3.9 | 2.9 |
| Muse Spark 1.3 | 72.2% | 3.4 | 19% | 1,806 | 6.5 | 1.2 | 0.9 |
| Muse Spark 1.3 Max | 72.2% | 3.4 | 19% | 1,989 | 6.6 | 2.2 | 1.8 |
| Grok 4.6 | 71.3% | 3.6 | 2% | 1,454 | 10.4 | 2.3 | 1.8 |
| GLM 5.3 Flash | 71.1% | 3.5 | 4% | 1,467 | 4.8 | 1.4 | 0.9 |
| MiMo V2.6 Flash | 70.7% | 3.7 | 7% | 1,768 | 7.2 | 2.0 | 1.4 |
| MiMo V2.6 Pro | 70.0% | 3.4 | 6% | 1,650 | 6.2 | 1.6 | 1.1 |
| MiniMax-M3 | 69.8% | 3.7 | 13% | 2,943 | 6.4 | 0.9 | 0.7 |
| Kimi K3.1 Pro | 69.5% | 3.5 | 17% | 1,869 | 26.0 | 7.9 | 5.7 |
| Gemini 4 Argon | 68.5% | 3.4 | 2% | 2,122 | 4.7 | 2.1 | 1.7 |
| Kimi K3 | 68.2% | 3.4 | 7% | 989 | 4.1 | 1.3 | 0.9 |
| Gemini 3.8 Flash | 67.1% | 3.8 | 1% | 1,552 | 2.1 | 0.1 | 0.1 |
| GPT-6.1 Sol | 65.8% | 3.8 | 88% | 1,352 | 13.0 | 16.5 | 12.1 |
| GPT-5.6 Luna | 65.6% | 3.6 | 3% | 1,753 | 25.6 | 4.1 | 3.1 |
Asking more doesn’t by itself mean asking the right thing. GPT-6.1 Sol ends its first reply with a question to the user 88% of the time, far more than Qwen 3.8 Max, which is next at 23%. Yet Sol’s elicitation score is only 68.1%, 18th of 23. Each elicitation criterion names one specific fact, such as when the tenant asked for the repair, or what the son was convicted of. Sol’s questions often go elsewhere: to the person’s state, to where a child is now, or to a checklist of facts to gather before an intake appointment. In the cases we read, the specific fact the rubric needed went unasked.
When models don’t ask, they sometimes assume key details instead, like the person’s state or whether there’s already a court order, and build their answers on those guesses. About three-quarters of criteria are legal information, so legal-information performance tracks the overall score more closely than elicitation does (correlation 0.93 vs. 0.62). Longer answers also tend to score higher (correlation 0.50 between final-answer length and score), and Qwen 3.8 Max writes the longest final answers, about 3,900 words on average. We aim to conduct further analyses determining a length-adjusted scoring and ranking as a future step for this project.
Our experts see a similar pattern in practice: when self-represented litigants use AI to draft court filings, the documents tend to run very long, and the extra length doesn’t necessarily make them more accurate.
Methodology
Dataset Construction
The scenarios start from real questions people bring to legal aid organizations. Two legal practitioners supplied real-world queries. Six legal aid experts then each adapted the queries to one assigned state and determined what a reliable first point of contact should ask about and tell the person, framing it as general legal information rather than legal advice.
| State | Scenarios |
|---|---|
| California | 44 |
| Louisiana | 44 |
| Virginia | 43 |
| New York | 41 |
| Maryland | 40 |
| Texas | 40 |
For each scenario, the expert wrote:
- the person’s opening question;
- background facts about the person’s situation, which the person shares only in conversation;
- a rubric of what context a good response must elicit and what information it must deliver (15.4 independent criteria on average);
- a difficulty rating, based on how complex the legal question is.
Scenarios are categorized by topic, subfield of law, case stage, and whether the question involves state law, federal law or both. The full dataset has 504 scenarios, split 50/50: models are evaluated on a 252-scenario test split used for published results, and the other 252 scenarios are kept private as a validation split.
The complete list of topics is below:
- Housing & Landlord-Tenant: evictions, leases, required repairs and utilities.
- Family Law: divorce, property division between spouses, child custody, child welfare and adoption, and civil domestic violence protections.
- Estate & Probate: wills, trusts and inheritance, and guardianship of incapacitated adults and minors.
- Employment: illegal discrimination, wage and hour, and labor law.
- Consumer & Debt: credit reports, billing and utility contract disputes, and debt collection.
- Criminal & Records: criminal law and procedure, and record expungement.
- Public Benefits: government assistance programs such as SNAP (“food stamps”), Medicaid, SSI/SSDI and Housing Choice Vouchers.
- Civil & Other: matters that don’t fit the other topics, such as a name change or how to file a small claims case.
The dataset captures state-level variation in the law that legal aid clients face. The states were chosen by comparing their legal frameworks, including statutes, adoption of uniform laws, procedural rules and common legal principles across major practice areas. The set covers the main legal approaches and includes outlier states that may need separate treatment, to capture as much legal diversity as possible without writing scenarios for all 50 states. The state law overview below summarizes the legal regimes represented across those states.
How each benchmark state approaches five major fields of law, compared with the approach most U.S. states take.
Majority U.S. approachMain alternativeState-specific
| State | Family Law | Criminal Law | Employment | Landlord-Tenant | Civil Procedure |
|---|---|---|---|---|---|
| California | Community Property | California statutory | Employee-protective | CA-specific | Unique |
| Louisiana | Community Property | Civil law | At-Will | Civil-law influenced | Unique |
| Maryland | Equitable | MPC influenced | At-Will | URLTA | FRCP-style |
| New York | Equitable | MPC influenced | At-Will | URLTA-inspired | Unique |
| Texas | Community Property | Traditional/Common Law | At-Will; Right-to-Work | TX Property Code | Unique |
| Virginia | Equitable | Traditional | At-Will | VRLTA | FRCP-style |
Evaluation Design
Overall, the evaluation consists of a model under test (target model) having a conversation with a simulated user, and then an automated judge grading the model’s performance against the expert-written rubric criteria. More details about each model follow.
- Target model, the model being evaluated. It acts as a first point of contact for legal help, can search the web and read pages, and can ask the person follow-up questions before giving its final answer.
- Simulated client (GPT-5). It opens with the expert’s question word for word, then answers only what the target directly asks, using the scenario’s background facts. It never volunteers other facts or corrects the target’s assumptions.
- Judge (Decider v1.2). It grades the full conversation against the expert rubric, scoring elicitation and legal information separately.
A model’s score is the share of rubric criteria it meets, pooled across all 252 scenarios. “Must find out” criteria count toward the score: a model gets credit for finding out a fact only if it asked about it. If the person’s opening question already gave the fact, or the client volunteered it, that criterion is set aside rather than counted for or against the model.
A sample conversation with each model, the criteria it was graded on, and the judge’s verdict on each is shown below.
Short answer
It depends entirely on what the money was for. Small claims court is a money-only court with no power over family-law matters, so:
- If the money is child support (or something a support order covers, like agreed-upon shares of daycare, medical, or school costs) → No. Small claims almost never has jurisdiction over support. You enforce it in the family/divorce court that issued the order, or through your state's child support enforcement agency.
- If it's an ordinary debt unrelated to support (they borrowed money from you, promised to reimburse you for a car repair, owed you rent, ran up a joint credit card and promised to pay it) → Usually yes, as a breach-of-contract/debt claim, if it's under your state's dollar limit.
Scenario 1: There's already a court order for support
Go back to the court that issued it. File a motion to enforce and/or motion for contempt (names vary by state — in Texas it's a "Motion to Enforce Child Support," in California an "Order to Show Cause"/FL-series forms). You can also open a case with your state child support agency (the federal Title IV-D agency), which is free or very low-cost and does not depend on your income — for example, Ohio's county Child Support Enforcement Agency is free to apply to (jfs.ohio.gov). Search "[your state] child support enforcement agency."
The agency route is usually far stronger than any lawsuit you could file yourself, because it can use tools a small claims judge cannot:
- Income withholding / wage garnishment sent directly to the employer
- Interception of federal and state tax refunds
- Suspension of driver's, professional, and recreational (hunting/fishing) licenses
- Liens on real estate and seizure of bank accounts
- Credit bureau reporting
- Denial or revocation of a U.S. passport once arrears reach $2,500
- Contempt, including fines and jail for willful non-payment
- Locating a parent who has moved, including across state lines (agencies are required to cooperate interstate)
Also important: arrears generally accrue interest, in many states a judge cannot waive interest or retroactively reduce past-due support, and in several states there is effectively no statute of limitations on past-due support — though a support judgment may need to be renewed after 10–20 years depending on state law.
Scenario 2: No court order — just a private promise
- If the promise was essentially "I'll pay you $X a month for the kids": a private side agreement between parents is generally not enforceable as child support. You need to file a petition in family court to establish a support order (and, if you were never married, to establish parentage first). Small claims can't set up ongoing support, and filing there risks dismissal and a wasted filing fee.
- If the promise was a separate, ordinary debt: small claims is typically the right venue. Bring the written agreement, texts/emails where they promised to pay, bank records showing partial payments, receipts, and a copy of any demand letter you sent.
Scenario 3: A separation agreement or written settlement
- If it was signed by a judge / incorporated into your divorce or custody judgment, enforce it in family court (contempt, or a motion to reduce the amount owed to a money judgment).
- If it's a private written agreement never filed with the court, some states will let you sue on it as a contract in small claims; others require anything arising from a domestic relationship to stay in family court. Ask the court's self-help center or clerk before you file.
If small claims is the right route, check these
- Dollar limit — verify on your state judicial branch's website. Examples from official sources: California is $12,500 for individuals and $6,250 for businesses (courts.ca.gov self-help guide); Colorado is $7,500 and its statute stresses that small claims "has only that jurisdiction specifically conferred upon it by law" (C.R.S. § 13-6-403, full text); Texas justice courts go up to $20,000; New York is $10,000 in NYC, $5,000 in city courts, $3,000 in town/village courts. Ignore the "limit charts" floating around commercial websites — several are out of date.
- Statute of limitations for written vs. oral contracts (commonly 3–6 years, varies by state).
- Venue — usually the county where the defendant lives or where the agreement was made/breached.
- Collecting is on you. As California's official guide puts it: "You have to collect the money if you win your case. The court does not do it for you." A judgment against a broke parent can be hard to turn into cash — which is another reason the state agency (with garnishment and license-suspension powers) is often the better tool.
- Lawyers — most states don't allow attorneys to appear with you in small claims, though you can consult one beforehand.
Free help
- Your state child support enforcement agency (free/low-cost application)
- Your local court's self-help center or law library — they'll tell you whether your claim belongs in small claims or family court
- Legal aid: lsc.gov/find-legal-aid for free civil legal help by ZIP code
- WomensLaw.org — plain-language, state-by-state pages on child support enforcement and court processes
Two things would let me give you specifics: (1) which state you're in, and (2) what the money was for — court-ordered support, agreed add-on expenses, or a separate debt — and whether your agreement was ever signed by a judge.
This is general legal information, not legal advice. If the amount is significant or there's a custody dispute tangled up with the money, a short consultation with a family-law attorney (many offer free or low-cost initial consults) is worth it.
One safety note: if there's any history of domestic violence, using the state agency or an attorney as your point of contact — rather than dealing with the other parent directly — is safer, and most states have address-confidentiality programs so your location isn't disclosed in court filings.
Bottom line for Louisiana: yes, you can — but the hard part will be proving the promise
Because this is a debt/loan claim, not child support, it's an ordinary civil matter. Louisiana's small claims courts handle exactly this kind of case. The catch is Louisiana's unusual rule for oral contracts.
1. Which court, and the $5,000 cap
Small claims in Louisiana lives in one of three places, depending on your parish:
- City Court (Small Claims Division) — e.g., Baton Rouge City Court, First City Court of New Orleans (421 Loyola Ave., Room 201), Second City Court (Algiers/west bank)
- Parish Court (Jefferson Parish)
- Justice of the Peace Court — JPs exist in only about 10 regions of the state and have been abolished in wards where a city court exists
The limit is $5,000 in the "amount in dispute," which does not include interest, court costs, attorney fees, or penalties (per the Law Library of Louisiana's JP Court guide). JP and city courts have concurrent jurisdiction with district court up to $5,000, so you have a choice.
If you're owed more than $5,000: you can either sue for $5,000 in small claims, or file the full amount in the city court's regular civil docket (commonly cited as up to $25,000) or in district court. Ask the clerk which makes sense — the regular docket is slower and more formal.
⚠️ How you word the petition matters. JP and city courts are barred from hearing divorce, annulment, separation, custody, alimony/support, successions, and title-to-real-estate cases. Your claim isn't one of those — but if you plead it as "money he owes me for the kids" or "support he promised," a judge may treat it as a family-law matter and dismiss it. Plead it as a loan / money lent / breach of an oral contract.
2. The big one: La. Civil Code art. 1846
Louisiana has a proof rule most states don't. La. C.C. art. 1846 says an oral contract for a value not over $500 may be proved by competent evidence, but:
"If the price or value is in excess of five hundred dollars, the contract must be proved by at least one witness and other corroborating circumstances."
The Louisiana Supreme Court explained how this works in Read v. Willwoods Community, 165 So. 3d 883 (La. 2015): your own testimony can count as the "one witness," but the corroborating circumstances must come from a source other than you, and they need only be general — not independent proof of every detail of the agreement.
So, gather anything from outside yourself:
- Texts, emails, voicemails, DMs where they acknowledged owing you or promised to pay ("I'll get you next week," "I know I owe you $2,000") — this is your best evidence
- Money-transfer records: Zelle/Cash App/Venmo/PayPal memo lines, bank statements, wire confirmations, cancelled checks, ATM withdrawal receipts
- Any partial repayments they made — extremely powerful; it's corroboration from their own conduct and it destroys a "that was a gift" defense
- Receipts/invoices showing what you paid for and why
- A third-party witness — anyone who was present or heard them promise to pay you back
- A written demand letter and their reply
Expect two defenses: "it was a gift" and "that was for the child, not a loan." Partial payments and written acknowledgments rebut both.
3. Deadline: 3 years (prescription)
Under La. C.C. art. 3494, "an action on money lent" and "an action on an open account" prescribe in three years. Other contract theories carry a 10-year period under art. 3499, but do not rely on that — a loan claim is almost certainly 3 years. It generally runs from when payment was due or when you demanded it. A partial payment or written acknowledgment can interrupt or restart prescription; a lawyer can tell you where you stand.
4. How the case actually runs
Under La. Code Civ. Proc. art. 4921.1:
- You file a petition with the clerk; the justice of the peace or clerk can set it for trial on filing
- The citation states the date, time, and place of trial
- The first trial date is no less than 10 and no more than 45 days after service of the citation
- The defendant doesn't have to file an answer unless the court orders it
- If they're served and don't show up, the judge can enter a default judgment
Venue: generally the parish where the defendant is domiciled (La. C.C.P. art. 42), or where the obligation arose/is to be performed. City courts and JP courts also have territorial boundaries (city limits/wards), so filing in the wrong one gets you dismissed. Call the clerk and give them the other parent's address before you file.
Costs: filing fees run roughly $30–$75 depending on parish, plus a separate fee for service by the constable or sheriff. If you can't afford it, ask about proceeding in forma pauperis (fee waiver).
Lawyers: unlike most states, Louisiana does not bar attorneys from small claims, so you can bring one — or just pay for an hour of advice beforehand.
5. Winning isn't collecting — a Louisiana wrinkle
Louisiana is one of the few states where wages are largely exempt from seizure for ordinary private debts (La. R.S. 13:3881, with exceptions mainly for support, taxes, and federal student loans). Sources conflict on how far the federal 25% cap still applies here, so don't count on wage garnishment — ask the clerk or a lawyer. Your realistic tools after judgment:
- Garnishment of their bank account
- Seizure and sale of property
- Recording the judgment in the parish conveyance/mortgage records to create a lien on real estate they own
- Judgment debtor examination to find assets
- Legal interest accrues on the judgment
6. Do this first (cheap and it builds your evidence)
Send a written demand letter by certified mail, return receipt requested: state the amount, when and how the money was given, when they promised to repay, a deadline (e.g., 10–14 days), and that you'll file in small claims otherwise. Keep a copy. Their reply — or even the green receipt showing they received it — becomes a corroborating circumstance under art. 1846.
7. Free and low-cost help in Louisiana
- Law Library of Louisiana / LEAP — free plain-language guides at lasc.libguides.com, including Justice of the Peace and small claims topics; LEAP is also staffed through many public libraries
- Southeast Louisiana Legal Services (slls.org) or Acadiana Legal Services (acadianalegal.org), depending on your parish — free civil legal aid
- LSBA Modest Means Directory — reduced-cost attorneys
- Your city/parish court clerk's office — most publish "How to Use Small Claims" instructions and a fee schedule (Baton Rouge City Court's is a good example)
- Court listings and addresses: louisianacourts.org/small-claims (unofficial, but useful for finding your court)
To narrow this down, tell me: (1) how much you're owed, (2) roughly when they promised to pay and when payment was due, (3) what evidence you have — any texts, transfer records, or partial payments? — and (4) your parish. With those I can point you at the specific court and tell you whether the 3-year clock is still open.
This is general legal information about Louisiana procedure, not legal advice. Given the art. 1846 proof requirement, a short consult with a Louisiana attorney (or a free legal-aid intake) before you file is genuinely worth it.
Which court: one of the two New Orleans City Courts
Orleans Parish has no Justice of the Peace courts (JPs were abolished in wards where a city court exists). Small claims lives in the Small Claims Division of one of two city courts, and which one depends on where the other parent lives:
| Court | Serves | Address | Phone / Hours |
|---|---|---|---|
| First City Court of New Orleans | East Bank of Orleans Parish | 421 Loyola Ave., Room 201, New Orleans, LA 70112 | 504-407-0400 · Mon–Fri 9:00–4:00 |
| Second City Court of New Orleans | West Bank / Algiers (15th Ward) | 225 Morgan St., New Orleans, LA 70114 | 504-407-0435 · Mon–Fri 9:00–4:00 |
- Both handle small claims up to $5,000 (exclusive of interest, court costs, attorney fees, and penalties), and regular civil suits above that (commonly cited as up to $25,000) — so if you later pin the amount above $5,000, you stay in the same building on the civil docket.
- Venue is generally the parish where the defendant is domiciled (La. C.C.P. art. 42), plus the court's own territorial boundary. East Bank address → First City Court. Algiers → Second City Court. If they live in Jefferson, St. Bernard, or another parish, you generally must file there — call the clerk before driving downtown.
- Do not file this in Civil District Court for the Parish of Orleans (400 Royal St.). It's the wrong tool for a $500–$5,000 debt: slower, more expensive, formal pleading rules.
- Call 504-407-0400 (First) or 504-407-0435 (Second) and ask for: their small claims petition form, the current filing fee and service fee, and whether they require anything before filing.
How to word it so it isn't treated as child support
City courts have no jurisdiction over divorce, custody, alimony, or support. If your petition reads like a support case, you risk dismissal and a lost filing fee. The fix is to describe a loan transaction, not a parenting relationship.
Use these words: money lent · advance · personal loan · oral contract · repayment · reimbursement · indebted · breach of contract · legal interest from judicial demand
Never use these words: child support · support · arrears · back support · "for the kids" · "for our child's expenses" · daycare/medical/school costs · custody · visitation · family · divorce · separation
Sample petition (adapt it; the clerk may have its own form)
PETITION FOR MONEY OWED
(MONEY LENT / BREACH OF ORAL CONTRACT)
Plaintiff, [your full legal name], domiciled at [your address],
Orleans Parish, Louisiana, respectfully alleges:
PARTIES AND VENUE
1. Plaintiff is an adult resident of Orleans Parish, Louisiana.
2. Defendant, [their full legal name], is an adult domiciled at
[their address], within the territorial jurisdiction of this Court.
3. Venue is proper in this Court because Defendant is domiciled
within its jurisdiction, and because the contract was made and was
to be performed in Orleans Parish.
STATEMENT OF FACTS
4. On or about [month/year], at Defendant's request, Plaintiff
advanced to Defendant the sum of $[amount].
5. At the time of the advance, Defendant orally promised and agreed
to repay Plaintiff that full sum [on or about (date) / upon demand].
6. Plaintiff delivered the funds by [cash withdrawn from (bank) on
(date) / Zelle transfer dated (date) / cashier's check no. ___ /
paying Defendant's bill to (creditor) directly].
7. Defendant has repaid $[0 or amount], and has failed and refused to
repay the balance despite Plaintiff's demands on [dates].
8. Defendant is therefore indebted to Plaintiff in the sum of
$[amount], plus legal interest from date of judicial demand.
CAUSE OF ACTION
9. Plaintiff's claim is an action on money lent and for breach of an
oral contract. See La. C.C. arts. 1846, 1906, 2900 et seq.;
La. C.C. art. 3494(3).
PRAYER FOR RELIEF
WHEREFORE Plaintiff demands judgment against Defendant in the full
amount of $[amount], with legal interest from date of judicial
demand, together with all costs of these proceedings.
Respectfully submitted,
____________________ [signature, printed name, address, phone]
[Sworn/notarized signature block if the clerk requires it]
One honesty check before you file: if the money was really handed over for the child's needs and the promise was framed as "I'll chip in for the kids," a judge may recharacterize it as support — and then the answer is a family-court petition, not small claims. If it was genuinely a loan or a reimbursement for something you paid on their behalf (their rent, their car, their credit card, their phone bill), you're on solid ground. Describe the transaction, not the purpose.
Your real problem is proof — do this before filing
Since the amount is over $500, La. C.C. art. 1846 requires the oral contract to be proved by "at least one witness and other corroborating circumstances." The Louisiana Supreme Court held in Read v. Willwoods Community, 165 So. 3d 883 (La. 2015) that your own testimony can be the "one witness," but the corroboration must come from a source other than you — and it need only be general, not independent proof of every detail.
Right now you say you have nothing. So build it, in this order:
Step 1 — Mine your own records (free, fastest).
- Bank statements/online banking for the period: cash withdrawals matching the amount, transfers, bill payments made for them
- Zelle, Venmo, Cash App, PayPal, Apple Pay histories — export them; memo lines are excellent corroboration
- Search your texts for:
$, "pay you back," "owe," "loan," "borrow," "next week," "I'll get you" - Email, Facebook Messenger, Instagram DMs, WhatsApp
- Voicemails and call logs
- Any receipt, bill, or invoice you paid on their behalf
Step 2 — Send one text designed to get an admission. This is the highest-value step you can take, because it does two things at once:
- It creates the corroborating circumstance from a source other than you that art. 1846 requires
- Under La. C.C. art. 3464, prescription is interrupted when the obligor acknowledges the right of the person against whom it had begun to run — so an acknowledgment can restart the 3-year clock you're worried about
Keep it short, calm, specific about the amount, and with no mention of the child and no threat to sue:
"Hi [name] — I'm getting my records straight. You borrowed $[amount] from me around [month/year] and said you'd pay me back. Can you confirm the amount and tell me when you can pay it?"
If they reply "yeah I know, I'll pay when I can" or "it was $1,200 not $1,500" — screenshot it immediately with their name/number and the date visible, then back it up outside your phone (email it to yourself, save to cloud). Even a partial admission of the amount is powerful.
Step 3 — Send a written demand letter by certified mail, return receipt requested (~$8–$10 at USPS). State the amount, when it was advanced, that repayment was promised, and a 10-day deadline. Keep the green receipt. Their reply — or their silence plus the receipt — is more corroboration.
Step 4 — Identify a witness: anyone present when the promise was made, or who heard the other parent say they owed you. A person who heard them promise is far stronger than a friend you merely told about it.
Only sue for what you can prove. If you believe it's $3,000 but can only corroborate $1,200, plead $1,200 — or plead the larger figure and be ready for the judge to award less.
The date problem
You don't know when it was promised or when it was due. That matters two ways:
- Prescription. "An action on money lent" prescribes in 3 years (La. C.C. art. 3494(3)). Reconstruct the date from bank records, texts, photos, or life anchors ("right after I started the job in spring 2023"), and plead "on or about [month/year]." If it's clearly more than 3 years old with no acknowledgment and no partial payment, the claim is likely prescribed and filing just costs you money — the Step 2 acknowledgment is your way out. Get a lawyer's read before relying on it.
- Pleading. You must allege a date and that payment was due. If no repayment date was ever set, the obligation is payable on demand — plead "upon demand" and allege the date you first demanded repayment.
If the defendant files a peremptory exception of prescription, they normally bear the burden of proving it — but if your own petition shows a time-barred claim on its face, that burden shifts. Another reason not to plead a date you can't support.
What to expect, and what it costs
- You file a petition with the clerk; the citation states the trial date, time, and place (La. C.C.P. art. 4921.1)
- The first trial date is not less than 10 and not more than 45 days after service of the citation
- The defendant need not file an answer unless the court orders it
- If they're served and don't appear, the judge may enter a default judgment
- Filing fee is roughly $50–$100 plus a separate service fee for the constable/sheriff — get the exact schedule from the clerk. Ask about proceeding in forma pauperis (fee waiver) if you can't afford it
- Lawyers are allowed in Louisiana small claims, unlike most states
- Bring: your filed petition, three copies of every piece of evidence (yours, the judge's, the defendant's), a one-page dated timeline, your witness(es), and a calculator showing the exact figure
- Appeals go to the district court
Collecting in Orleans Parish: Louisiana largely exempts wages from seizure for ordinary private debts (La. R.S. 13:3881 — exceptions mainly for support, taxes, and federal student loans), so don't count on garnishing their paycheck. Your realistic tools are garnishing a bank account, seizing property, or recording the judgment in the Orleans Parish Clerk of Court's mortgage/conveyance records to create a lien on any real estate they own. Legal interest runs from judicial demand. If they have no bank account and no assets, a judgment may be uncollectible — worth knowing before you spend the filing fee.
Free help in New Orleans
- First City Court clerk, 421 Loyola Ave., Room 201 — small claims instructions, forms, current fee schedule · 504-407-0400
- Law Library of Louisiana — Louisiana Supreme Court building (Chief Justice Pascal F. Calogero Jr. Courthouse), 400 Royal St., 2nd Floor, New Orleans · 504-310-2400, toll-free in LA 800-820-3038 · Mon–Fri 8:30–4:30. Free and open to the public; librarians will pull art. 1846, art. 3464, and art. 3494 for you and point you to forms. They also run the LEAP guides at lasc.libguides.com
- Southeast Louisiana Legal Services (SLLS) — New Orleans office, 1340 Poydras St., Ste. 600 · 504-529-1000, toll-free 1-877-521-6242 · slls.org. Free civil legal aid if you're income-eligible; serves Orleans Parish
- New Orleans Public Library Virtual Civil Legal Clinic — free attorney appointments by phone/video on the 2nd and 4th Wednesday of each month; sign up through any NOPL branch (nolalibrary.org)
- LSBA Modest Means Directory — reduced-fee Louisiana attorneys
My suggested sequence: (1) pull your bank/app/text records this week; (2) send the Step 2 text and screenshot any reply; (3) send the certified demand letter; (4) call the correct city court clerk to confirm venue, form, and fees; (5) then file. Steps 1–3 cost almost nothing and are what will actually decide whether you win.
This is general information about Louisiana procedure, not legal advice. Because art. 1846 makes proof the whole ballgame here, a free SLLS intake or a Law Library visit before you file is well worth your time.
End of conversation