Late 2025, a homeowner named — let's call her Maria — hired me about a construction dispute.
The facts were straightforward. She had contracted with a licensed California contractor for a fixed-price pool and landscaping project: $87,000, phased payments tied to milestones. The contractor completed roughly one-third of the work, collected two of four scheduled payments totaling $52,000, then went quiet. No returned calls. No on-site crew. No explanation.
From a legal standpoint, this was a textbook California construction dispute. The CSLB gives homeowners powerful tools. The contractor's $25,000 surety bond was a partial recovery path. And based on what Maria described, the factual record strongly supported both breach of contract and a potential fraud claim under California Penal Code §532.
This case should have settled within sixty days. It did not.
By the time Maria came to me, she had been exchanging written communications with the contractor for four months — and nearly every document in her file had been drafted with the help of an AI chatbot. So had most of the contractor's responses.
What followed was one of the most instructive cases I have handled in recent years — not because of the legal complexity, but because of what the AI-assisted communications had done to the dispute before I ever touched it.
What the AI Letters Actually Said
Maria had used ChatGPT to draft three demand letters and two follow-up emails. I read them carefully. They were well-written. Grammatically clean, formally structured, and impressive in tone. They cited California Civil Code, referenced the CSLB process accurately, and threatened the right remedies. If you evaluated them purely as documents, you would have concluded that Maria had a competent attorney drafting her communications.
The problem was the content — not the grammar.
Letter 1 demanded the contractor return to the worksite within 72 hours or face a lawsuit for breach of contract and fraud. It stated, as fact, that the contractor had "fraudulently induced" Maria to enter the contract with no intention of completing the work.
The fraud allegation was premature and unsupported at that stage. Fraud requires proving intent — that the contractor never intended to perform. Maria didn't have that evidence when Letter 1 was sent. What she had was a contractor who stopped showing up. That may be breach of contract. It may become fraud if you develop the right evidence. But declaring fraud in the first communication, before any investigation, is a strategic mistake — it hardens positions and invites a defamation counterclaim if the allegation turns out to be wrong.
Letter 2, sent after no response, escalated. It stated that Maria had "retained legal counsel" — she had not — and that her attorney had advised her to "preserve all evidence of criminal conduct." This was a lie, drafted by a chatbot that assumed escalation was the right move. The contractor's attorney would later use this letter to argue that Maria had acted in bad faith from the outset.
Letter 3 made a settlement demand of $87,000 — the full contract price — rather than the $52,000 actually paid. This overclaim, though presumably generated by an AI trying to "maximize" the demand, immediately undermined Maria's credibility. If she had only paid $52,000, demanding $87,000 without explanation signals either a math error or deliberate overreach.
What the Contractor's AI Letters Said
The contractor, it turned out, was doing the same thing. His three response letters were also AI-generated — I could tell from both the structure and several phrases characteristic of LLM output — and they made their own set of strategic errors.
His first response denied everything with a specificity that was actually self-defeating. He described the precise scope of work completed, itemized materials purchased, and named three subcontractors involved — all in a letter clearly drafted without attorney review. He had handed me a roadmap to his defense before we even filed.
His second letter threatened to countersue for defamation based on Maria's fraud allegation. This was probably the right instinct, but the letter also implied he had stopped work because Maria had "breached" by making payments late — a new factual claim he had never raised in any prior communication, which would be nearly impossible to introduce successfully at this stage.
His third letter attached a lien waiver form he claimed Maria had signed, releasing him from liability for the first phase. The form was real. But it had been buried in a packet of documents Maria signed at project kickoff, and she had no idea it was there.
The Cost of Getting There
By the time I was retained, the AI correspondence had accomplished the following:
- A fraud allegation was in writing — unsupported, premature, and lodged against a licensed contractor who had a partial case for legitimate payment disputes.
- Maria had falsely represented having legal counsel. That misrepresentation would need explaining in a subsequent declaration.
- The settlement demand was $87,000 on a $52,000 payment dispute. Getting from $87,000 back to a credible $52,000 demand — while explaining why we started at $87,000 — required a letter that basically said we were correcting a prior error.
- The contractor had made admissions about scope and subcontractors that were useful — but I also had to deal with four months of letters characterizing Maria's communications as harassing and bad-faith.
This case ultimately settled for $38,000 — less than the $52,000 paid — after eight months and approximately $14,000 in attorney fees on Maria's side. It should have settled in sixty days for something close to $48,000.
The AI letters cost Maria approximately $24,000 in combined loss of recovery and legal fees.
What a Human Would Have Caught
I don't write this to condemn AI tools. I use them myself — for research, for drafting, for structuring arguments. They are genuinely useful.
What I write it to say is that there is a specific category of error that AI drafting tools reliably make, and that category is strategic context.
A language model is very good at constructing a demand letter that sounds like a demand letter. It is not equipped to evaluate whether the evidence supports the legal claims being made, whether the timing of an allegation will hurt you procedurally, whether the settlement demand is calibrated to the actual recovery picture, whether a specific phrase will be used against you later, or whether the emotional posture of the letter will close off resolution or open it.
These are judgment calls that require knowing the full facts, knowing California procedure, and having handled enough disputes to recognize what moves each party toward and away from resolution. That knowledge is not in the training data. It is in the room.
What to Do Instead
If you are in a contractor dispute in California and want to protect your position before retaining counsel:
Document everything, send nothing. Write down every fact you can remember. Photograph all work. Save every text message, email, and voicemail. Make a timeline. Do not send anything until you have consulted a California construction attorney about what to say and when.
File the CSLB complaint early. The California Contractors State License Board has real authority and real teeth. A complaint filed early establishes a contemporaneous record and often brings contractors back to the table faster than a demand letter. It is free.
Understand the bond. If your contractor is licensed, they carry a $25,000 surety bond. That bond is a specific, bounded recovery path — file the claim as part of your overall strategy, not as an afterthought.
Let your attorney draft the first demand letter. Not because you cannot write — but because the first letter sets the frame for everything that follows.
If you need representation for a California construction dispute, my firm handles these matters statewide. You can reach Bay Legal, PC at BayLegal.com or (650) 668-8000. For information about your rights as a homeowner, visit BadContractorLaw.com.
Jayson R. Elliott is a California attorney and Managing Director of Bay Legal, PC. The case described in this essay is a composite of similar matters; all identifying details have been changed. This article does not constitute legal advice. Attorney advertising: prior results do not guarantee similar outcomes.