AI-Style Legal Word Salad Backfires: Self-Represented Appellant Loses De Facto Appeal and Is Ordered to Pay $44,351 Indemnity Costs

In Falvo & Melita (No 3) [2026] FedCFamC1A 154, the Full Court dismissed the appellant’s appeal from a declaration that she and the respondent had been in a de facto relationship from July 2018 to April 2023. The critical feature of the appeal was not simply that she lost. It was that her written material used complex legal language, constitutional language, procedural fairness language, bias language and appellate terminology without demonstrating legal error. Her oral submissions then revealed she did not understand or could not properly engage with the legal concepts she had deployed in writing. The Court observed that the submissions may well have been generated with the aid of artificial intelligence, and treated the resulting volume, confusion and lack of connection to appealable error as conduct productive of cost and delay. The appeal was dismissed and she was ordered to pay the respondent’s costs fixed at $44,351 within 28 days.

🧩 Facts and Issues

Facts:

The appellant, Ms Falvo, appealed from a declaration made under s 90RD of the Family Law Act 1975 (Cth) that she and the respondent, Mr Melita, were in a de facto relationship from July 2018 until April 2023. The respondent opposed the appeal. The Full Court dismissed it.

The parties had met in late 2017 or early 2018 and commenced a romantic and sexual relationship by early 2018. The appellant accepted there had been a romantic relationship but argued it was brief and that the relationship was really a business relationship, not a de facto relationship.

The primary judge had made extensive findings supporting the existence of a de facto relationship. These included that the parties spent many nights together, had a continuing sexual relationship, shared daily life, shared holidays, lived between both properties and later exclusively at the appellant’s property, were engaged to marry, publicly presented as partners, shared information with the same lawyer, and that the respondent provided substantial financial support and improvements to the appellant’s property.

The primary judge concluded that the parties lived together as a couple on a genuine domestic basis from about July 2018 until April 2023, and that the evidence compellingly supported a de facto relationship within s 4AA.

Issues:

1. Did the appellant identify appealable error in the de facto declaration?
2. Did her grounds of appeal properly engage with s 4AA and the primary judge’s reasoning?
3. Did allegations of procedural unfairness, bias, inadequate reasons, factual error and discretionary error have merit?
4. What significance did the Court give to the appellant’s use of legal terminology she could not engage with orally?
5. How did the suspected use of AI affect the Court’s view of the appeal?
6. Why were indemnity costs ordered?
7. Why was the appellant ordered to pay $44,351?

⚖️ Applicable Law – Legislation, Regulations, Rules

Family Law Act 1975 (Cth)

• s 4AA — definition and indicia of a de facto relationship.
• s 90RD — declaration as to the existence and period of a de facto relationship.
• s 90SM — property adjustment following de facto relationship finding.
• s 95 — overarching purpose: just resolution as quickly, inexpensively and efficiently as possible.
• s 96 — duty of parties and lawyers to conduct proceedings consistently with the overarching purpose.
• s 114UB — costs discretion in family law proceedings.

Judiciary Act 1903 (Cth)

• s 78B — notice of constitutional matter. The appellant filed such a notice, but the Court held it did not identify any real constitutional issue and was an abuse of process.

Federal Circuit and Family Court of Australia (Family Law) Rules 2021 (Cth)

• r 13.23 — page limit and requirements for appeal summaries of argument.
• r 12.17 — power to fix costs.
• r 13.54 — appeal-related procedural rule referred to in the judgment.

📌 Precedents Relied On

Key authorities included:

• Fairbairn v Radecki — holistic assessment of whether a de facto relationship exists under s 4AA.
• Warren v Coombes — appellate review and presumption that the primary judge’s decision is correct unless error is shown.
• Ebner v Official Trustee in Bankruptcy and Charisteas v Charisteas — apprehended bias test.
• Kioa v West, SZBEL, Ex parte Lam, and Stead — procedural fairness principles.
• Bennett and Bennett, Yarrow & Yarrow, DL v The Queen, Whisprun, and Fox v Percy — adequacy of reasons.
• Cizek & Mihov, Lee v Lee, Edwards v Noble, and Robinson Helicopter — appellate approach to factual findings and credibility findings.
• Kohan, Yunghanns, Colgate-Palmolive, and Kaba & Zemin (No 2) — indemnity costs principles and consequences of persisting with hopeless or problematic appeals.

🧠 Analysis

Issue

Why did the appeal fail, and why did the Court treat the appellant’s AI-like, legally dense written submissions and unfamiliar oral advocacy as conduct justifying an indemnity costs order of $44,351?

Rule

An appeal is not a fresh opportunity to reargue dissatisfaction with the outcome. The appellant must identify legal, factual, procedural or reasons error in the judgment below. The primary decision is presumed correct unless appealable error is demonstrated.

Self-represented litigants are given some allowance, but they remain bound by the same appellate principles and obligations as any other litigant. An appellant must state grounds clearly, particularise the alleged errors, and connect submissions to the grounds of appeal.

Artificial intelligence is not prohibited. But AI-generated or AI-assisted material can become dangerous when it produces legal language, legal labels and voluminous submissions without identifying actual appealable error. The Court’s concern is not the technology itself; the concern is that legal language without legal understanding can obscure the real issues, increase cost, delay the proceeding and burden the opposing party.

Application

1. The appellant used appeal language but did not demonstrate appeal error

The appellant’s grounds alleged legal error, factual error, procedural unfairness, apprehended bias, discretionary error, decision against the weight of evidence and unsafe findings. But the Full Court found that the written material recited broad, generalised legal complaints without engaging with the actual requirement to demonstrate legal error.

The problem was not that she was self-represented. The problem was that her grounds used legal labels as conclusions. Saying “procedural fairness”, “bias”, “unsafe findings” or “decision against the evidence” is not enough. The appellant had to show where the primary judge went wrong and why that error mattered.

She did not do so.

2. Her oral submissions exposed the problem with the written submissions

The Full Court made a pointed observation: the appellant prepared her material without legal representation, but although her documents used legal language, they did not engage with the need to demonstrate legal error. Her oral submissions then revealed a lack of familiarity with the terms she had used in writing.

That is the key fact for the article. The written submissions had the appearance of legal sophistication. But when tested orally, the appellant could not properly connect the legal terminology to the grounds of appeal or the judgment under challenge.

The Court described her oral submissions as broad statements of general principle from disparate and sometimes unrelated authorities, without any nexus to a ground of appeal, showing limited awareness of the core complaints asserted.

In plain terms: the document sounded legal, but the argument did not operate legally.

3. The Court linked that problem to possible AI use

The Full Court then made the AI observation. It said the appellant’s submissions may well have been generated with the aid of artificial intelligence. The Court considered that explanation consistent with the fact that the document used legal language and concepts with which the appellant herself could not engage orally.

That is not a finding that AI was definitely used. The Court used cautious language: “may well have been generated”.

But the practical effect was clear. The Court treated the suspected AI-style submissions as part of the explanation for why the appellant had produced large amounts of legal language that did not illuminate the real appeal issues.

This is the warning: AI can make a self-represented litigant sound legally fluent without making the appeal legally competent.

4. The Court’s criticism was not “AI bad”; it was “AI without legal understanding is dangerous”

The Court did not say AI can never be used. The criticism was more precise. The Court said that the use of AI by those without legal training may seem deceptively as though it will compensate for the user’s lack of knowledge, but that is not universally true. Large amounts of legal language with limited connection to the central issue can obscure rather than illuminate the issue.

This is a crucial distinction. AI was not condemned as a tool. It was held against the appellant because of the effect of the material she filed:

• it was voluminous;
• it was poorly ordered;
• it was legalistic without being legally useful;
• it did not identify appellate error;
• it increased cost and delay; and
• it forced the respondent’s lawyers and the Court to deal with unnecessary material.

That is why the AI point fed into the costs analysis.

5. The appellant had already been warned her appeal faced serious obstacles

The appellant had earlier filed two Applications in an Appeal on 2 June 2026. One sought review of procedural orders and was supported by a 114-page affidavit. The other sought the Court to supply five days of transcript or dispense with the transcript requirement and was supported by a 48-page affidavit.

On 12 June 2026, reasons were delivered explaining that the appellant’s grounds faced significant problems. Those reasons said some alleged legal errors were difficult to follow, factual mistakes were unparticularised, and the procedural fairness/bias complaints may have been little more than a mask for dissatisfaction with the declaration.

The Full Court found that the appellant must have understood, at least from those reasons, that her appeal faced significant obstacles.

That mattered heavily on costs. Once warned, persisting with a meritless appeal became much harder to justify.

6. The appellant’s 117-page Summary of Argument was contrary to the Rules and excessive

The appellant later filed an Application in an Appeal seeking to have a document titled “Legal Submission in Appeal” accepted as her Summary of Argument. She asked the Court to accept the first 19 pages, or allow her to compress it to 15 pages, and also sought leave to rely on pages 19–117 as background and facts.

At the appeal hearing, the Court allowed her to rely on the first 19 pages, already more than the 15 pages permitted by r 13.23, but dismissed the rest of that application. The Court noted that it is not necessary for an appellant to provide background because the background is in the Appeal Book.

This is another major conduct point. The appellant did not just file a weak appeal. She generated excessive material that failed to perform the function of an appellate summary: identifying appealable error concisely.

7. The constitutional notice was described as an abuse of process

The appellant also filed a s 78B constitutional notice alleging broad failures by judges and delegated officers to act independently, impartially and consistently with Chapter III of the Constitution. The Court found that the notice did not identify any essential constitutional ingredient and that ordinary appellate principles were sufficient to deal with the complaints. The notice was described as an abuse of process.

That was another factor showing the appellant’s legal language had become disconnected from the real appeal. Invoking the Constitution did not transform ordinary complaints about bias, procedure or outcome into a constitutional matter.

8. The appeal itself had no merit

The Court rejected the bias complaint because the appellant did not identify facts that could lead a fair-minded observer to apprehend that the primary judge might decide the case other than on its legal and factual merits. Mere dissatisfaction with the result was not apprehended bias.

The procedural fairness complaints failed because the appellant had been given an opportunity to be heard. Her real complaint was that the primary judge gave little or no weight to evidence she relied upon. But a failure to accept a party’s submissions is not a denial of procedural fairness.

The inadequate reasons complaint failed because the primary judge had engaged in the required holistic assessment under s 4AA, considering the duration of the relationship, common residence, sexual relationship, financial interdependence, property use, mutual commitment, care and support of children, and public aspects of the relationship.

The factual and credibility challenges failed because the primary judge’s reasons explained why the respondent’s evidence was preferred and why the appellant’s evidence was approached with circumspection.

The Full Court’s conclusion was blunt: no ground of appeal had merit.

9. The original de facto findings were strong

The appeal was also weak because the unchallenged findings below strongly supported the de facto declaration.

Those findings included that the parties spent most nights together, shared daily life, had a sexual relationship, shared holidays, lived at the appellant’s property, were engaged to be married, publicly described each other as partners, shared legal appointments, and that the respondent provided significant financial support including money for food, groceries, living expenses, property improvements, legal fees, and funds to help the appellant retain her property.

The primary judge had concluded that the evidence compellingly supported a de facto relationship from July 2018 to April 2023.

So the appeal did not merely suffer from poor drafting. It also ran into a strong evidentiary foundation.

10. Costs: the Court held the appellant’s conduct breached the overarching purpose

The respondent sought indemnity costs fixed at $44,351, or alternatively scale costs fixed at $34,393.34.

The Court accepted that the appellant’s conduct of the appeal was contrary to the overarching purpose in ss 95 and 96, which require parties to facilitate the just resolution of disputes as quickly, inexpensively and efficiently as possible.

The costs reasoning focused on several features:

• the appeal was wholly unsuccessful;
• the appellant had already been warned that her grounds faced substantial challenges;
• she filed a misconceived review application supported by a 114-page affidavit;
• she filed unnecessary or excessive material;
• she filed a 117-page Summary of Argument contrary to the Rules and the registrar’s order;
• she served a misconceived constitutional notice; and
• the respondent’s lawyers had to read and respond to the material.

11. The appellant’s financial hardship argument did not save her

The appellant opposed costs, saying in substance that she could not pay and that each party should pay their own costs. But her Financial Statement had refused to disclose her financial circumstances beyond ownership of an unencumbered property, and two earlier costs orders against her — $5,000 and $3,600 — remained unpaid.

The Court held that alleged impecuniosity, without more, was not enough to deprive the respondent of costs. It also rejected the appellant’s suggestion that the respondent’s costs were merely a product of his “choice to litigate” or his choice to retain lawyers.

That submission was described as incredible in substance because the respondent had to defend an appeal she chose to prosecute.

12. Why indemnity costs were ordered

Indemnity costs are exceptional and compensatory, not punitive. The Court accepted that this was an exceptional case because the appellant knew, or ought to have known, from the earlier June 2026 reasons that her appeal had no reasonable prospect of success, and that persisting would likely expose her to a costs order.

The Court relied by analogy on Kaba & Zemin (No 2), where indemnity costs were justified because the appellant ignored earlier judicial observations that the grounds faced serious obstacles.

The Full Court held that the appeal had unnecessarily caused the respondent costs and that it would be unjust for him to be left out of pocket.

The result was a fixed indemnity costs order of $44,351, payable within 28 days.

Conclusion

The appeal was dismissed because none of the appellant’s grounds had merit. Her material used legal language but failed to demonstrate legal error. Her oral submissions showed she was unfamiliar with the legal concepts used in her written documents. The Court considered that the written material may have been AI-assisted, and warned that AI-generated legal language can obscure rather than illuminate appellate issues where the user lacks the legal training to connect the material to actual error.

The AI issue was held against her not because using AI is automatically improper, but because the apparent AI-style output contributed to a broader pattern of excessive, poorly ordered, legally disconnected material that increased cost and delay.

The appellant had already been warned that her appeal faced significant obstacles. She nevertheless persisted, filed excessive material, filed a misconceived constitutional notice, and prosecuted an appeal that was wholly unsuccessful. The Court found the circumstances justified indemnity costs and ordered her to pay the respondent $44,351 within 28 days.

🧠 Take-Home Lesson

This case is a major warning for self-represented litigants using AI in family law appeals.

AI can produce legal language. It can generate headings, authorities, constitutional phrases, procedural fairness complaints, bias language and appellate terminology. But it cannot make a weak appeal strong unless the actual legal error is identified, particularised and connected to the judgment.

The danger is that AI may create the appearance of legal sophistication while hiding the absence of a real appellate point. That is exactly what concerned the Court here. The appellant’s written material sounded legal, but her oral submissions showed she could not properly engage with the concepts she had used.

The practical message is sharp: AI-generated legal language without legal understanding can cost you real money.

Here, the cost was not theoretical. The appellant lost the appeal and was ordered to pay $44,351 on an indemnity basis.

The lesson is not “never use AI”. The lesson is: use AI to clarify, organise, draft documents and test your argument — not to bury the Court and the other party in legalistic word salad, which you don't even understand or read, moreover AI is not a substitute for legal advice, you should always have a lawyer review your documents before filing as you can see here the costs can far outweigh what you might save.

Every appeal ground must answer one question: where is the actual error in the judgment?