AI implementation, by an attorney who runs it on his own site

Both. I'm a California attorney (Bar #279869) licensed since 2011 with a J.D. from Boston University School of Law, and I've been a daily user of generative AI since the GPT-3 API was a private beta. I wrote and ship every line of the tool code on Terms.Law myself, I draft the system prompts, and I review every paid work product. The legal judgment is mine. The AI helps me move faster on the parts that are mechanical.

No. The tool proxy on Terms.Law does not persist input or output payloads. Anthropic processes the prompt through the Messages API and returns a response, and the response is rendered into the browser. When a user opts in by capturing an email for a lead notification, only the email plus a short results summary go to my inbox and a Telegram backup, never the raw prompt body. Free-text inputs are kept ephemeral by design.

Claude Opus 4.7 via the Anthropic Messages API is the workhorse for every interactive tool on the site, plus the calculator narrative layer. The chatbox runs an A/B between Opus and a legacy GPT-5.4 deployment so I can measure conversion lift directly. I picked Opus because it stays grounded on multi-step California-specific reasoning where cheaper tiers hallucinate citations. The differentiation matters when the output is going on a demand letter.

Yes. The eight tools on this page cover plaintiff litigation, contract review, AI ethics, and pro se filing. Replicating the pattern for employment, IP, tax, healthcare, family, real estate, or transactional work is mostly a matter of swapping the system prompt and the tier-routing logic. Flat-fee audits start at $2,500. Full custom builds run $3,500 to $5,000 depending on scope. I quote a fixed fee in writing before any work begins.

Three layers. First, vendor diligence: I only route confidential work through providers with contractual non-training commitments. Second, redaction: every system prompt instructs the model to treat input as patterns and facts rather than identities, and the audit package includes a redaction checklist for staff. Third, workflow design: the most sensitive material never touches a third-party API at all, and I document which task category gets which tool. CRPC 1.6 compliance is engineered in, not bolted on.

Yes. Every $2,500 audit package includes a one-hour live training session, recorded so new hires can watch later. The session covers the firm's specific AI Use Policy, the named-vendor matrix, the redaction protocol, the supervision and review checklist, and the engagement-letter disclosure language. For the $3,500 to $5,000 implementation package, I add a second training that walks the team through the custom workflow. Both formats are written communication only, no calls during scoping.

I work with any combination of Claude, ChatGPT, Gemini, CoCounsel, Harvey, Lexis+ AI, Westlaw AI, and Microsoft Copilot. The audit produces a vendor matrix that maps each vendor to a data classification and a permitted task list, so the policy survives a vendor swap. I have no vendor referral fees, no white-label resale, and no incentive to push any specific tool. The right stack is whatever fits your data, budget, and existing licenses.

The $2,500 audit runs one to two weeks from kickoff email to delivered policy. The $3,500 to $5,000 implementation runs four to twelve weeks depending on workflow complexity, number of tools to deploy, and team-training scope. Both timelines assume responsive email turnaround from your side. I write a scope and timeline document at the start of the engagement and bill against it, so there are no surprise overruns.

Solos through small firms in the two-to-fifty-attorney range, plus in-house GCs running lean legal departments, plus accountants and professional-services firms whose confidentiality calculus mirrors law firm ethics. The connecting thread is written work product where the underlying data is sensitive and the cost of an AI mistake is real. I do not work with consumer-facing chatbot startups or with anyone building unsupervised legal-advice products.

Conversion math. The tools attract informational traffic that would otherwise bounce off a static service page, and a meaningful share of users who finish a tool session end up emailing me to hire me for the actual work. A single $575 demand letter conversion pays for roughly 11,500 Opus tool calls at current API pricing. The break-even per acquired client is tiny. The tools also prove competence on a live page, which is harder to fake than a testimonials carousel.