The average solo attorney who tries AI does it the wrong way: they subscribe to three tools in the same month, try to automate everything at once, lose track of which tool does what, and after six weeks give up and go back to doing everything manually. The problem is not the tools — it is the approach.
A working AI workflow is built the same way a well-run solo practice is built: one reliable system at a time, added in the order that creates the most value, and maintained with enough discipline that it actually gets used.
This is the playbook: how to start, which tasks to tackle first, what the ethics checkpoints look like in practice, and how to expand the workflow as the first tools become habits.
Start with one task, not everything
The mistake is trying to use AI for everything at once before you understand it well for anything. The right question for month one is not "how do I use AI in my practice" — it is "what is the one task that costs me the most time, follows a predictable pattern, and would be faster if I had a first draft?"
For most solos, the answer is one of four things:
- Routine client correspondence — the status updates, follow-ups, scheduling emails, and reminders that eat an hour every day and follow the same structure every time
- First drafts of recurring documents — the NDA, the engagement letter, the standard motion template — documents that start the same way every matter
- Research preparation — turning a factual situation into an organized list of legal issues and research questions before you start reading cases
- Administrative drafting — agendas, summaries of calls, task lists from meeting notes
Pick the one task on this list that consumes the most time in your practice. That is where you start.
Month one: establish one reliable workflow
Choose a single general AI assistant and learn it. ChatGPT, Claude, or Gemini — pick one, get the business or professional tier (for the confidentiality protections), and use it exclusively for the first month. Do not evaluate alternatives yet. The first month is about building the skill of prompting well, not about comparing tools.
Write a model prompt for your chosen task. Read the prompt engineering guide and write a prompt template for the one task you chose. The template should include: the role context, the situation structure, the format requirements, and the prohibitions (what not to include). Refine it over the first two weeks until it reliably produces a draft worth reviewing.
Build the review habit. Every AI output gets reviewed before it is used. This is non-negotiable — it is the professional obligation under Model Rules 1.1, 5.1, and 5.3, and it is the behavior that keeps you competent in the tool. The goal is to make review fast and systematic, not to skip it. For routine correspondence: read the draft, check accuracy, adjust tone, send. This should take two to five minutes.
Track the time. Measure how long the task took before AI and after AI. This is motivational data, but it also tells you whether the tool is earning its subscription cost.
At the end of month one, you should have one task that takes less time than it used to and a prompt template that reliably starts that workflow.
Month two: add the confidentiality infrastructure
If month one was about a task that did not require client-specific information (structure of an NDA, format of a status update without names), month two is about establishing the confidentiality framework that allows you to use AI on real client files.
Understand which tier you are on. Log in to whichever tool you chose and find the data handling terms. Specifically: does the plan you are on train on your inputs? Is there a data processing agreement (DPA)? For business and enterprise tiers of major tools, the answer is generally yes on both counts — but verify the current documentation. AI data handling terms are an active area of product development and change.
Adopt the anonymization practice for prompts. Replace client names and identifying matter details with role descriptions ("the client," "the opposing party," "the contract in dispute") in every prompt, regardless of which tier you are on. This is belt-and-suspenders practice: it does not depend on any vendor's current terms and it protects against the tool being the wrong tier for some reason. The output quality is identical.
Set a firm-wide rule for AI use. If you have any staff — a paralegal, a legal assistant, an associate — establish explicitly which tools are approved, which tier is required, and that every output is reviewed before use. The supervision duty under Rules 5.1 and 5.3 applies to their AI use as much as yours. A simple written policy (a one-page note in the office procedures) makes this enforceable.
Info
Month three: expand to a second task
By month three, one task is running smoothly and you have a review habit established. Now add a second.
Choose based on where the next-largest time sink is, but also consider task adjacency. Tasks that are similar in structure to month one (also drafting, or also organizing information) are easier to add because you are extending a skill you already have. Tasks in a completely different category (research prep, contract review) require a new learning curve.
Good second tasks for different practice profiles:
Litigation-heavy practice: deposition preparation and first-pass transcript summary. See the full workflow in AI for deposition summarization.
Transactional practice: contract clause drafting and first-pass review for unusual provisions. The prompt-based approach works well for contracts where the key job is flagging what to look at more closely. For volume contract work, a dedicated tool like Spellbook is worth adding at this stage.
Client-acquisition focus: intake workflow — building the chatbot or intake form that captures leads 24/7. This one has an infrastructure component (the intake platform) beyond the prompt workflow. See AI client intake for law firms.
All practices: email drafting workflow from the communication drafting guide. If this was not month one, it is worth adding now.
Building the toolkit over time
The model that emerges over six to twelve months of deliberate AI adoption:
| Layer | Tool type | When to add | The job |
|---|---|---|---|
| Foundation | General assistant (ChatGPT/Claude/Gemini) | Month 1 | Drafting, summarizing, prompting, admin |
| Intake | Intake platform or configured chatbot | Month 2-3 | Lead capture, triage, pre-consultation info |
| Research (if needed) | Legal research tool (CoCounsel/Lexis+ AI) | When research volume justifies it | Verify authority on real legal databases |
| Contracts (if transactional) | Spellbook or similar | When contract volume justifies it | Clause drafting and flagging inside Word |
| Practice management AI | Clio Duo, MyCase IQ, etc. | If already on that platform | Matter summaries, drafting from case context |
The right sequence is exactly this order. A general assistant used well eliminates 80% of the routine drafting overhead before you spend a dollar on anything specialized. Add specialized tools only when the job is high-volume enough to justify them.
The ongoing ethics obligation
ABA Formal Opinion 512 frames competence with AI as an ongoing duty, not a one-time assessment. Staying competent means:
- Knowing what the tool you use can and cannot do
- Knowing what tier you are on and what confidentiality protections it provides
- Keeping up with significant changes in the tool's capabilities and terms
- Communicating material AI use to clients where appropriate — the state-by-state guidance is in state bar AI guidance
- Billing AI-assisted work reasonably — you cannot charge for hours the tool saved you
These are not burdensome additions to a practice — they are the same professional habits that apply to any tool or technology you adopt. They just need to be applied deliberately to AI.
The failure mode to avoid
The failure mode that ends AI adoption in small practices is not a bad tool — it is trying to use AI for everything before mastering it for anything.
The practices getting real value from AI in 2026 are typically running one general assistant they know well, plus at most one or two specialized tools aimed at a specific high-volume workflow. They review everything. They have a confidentiality framework that is simple enough to actually follow. And they have been doing it long enough that the prompting skill is second nature.
That is the target state. Month one is just the first step.
For the full map of legal AI tools organized by job, see the best AI tools for lawyers. For the prompt skills that make every tool perform better, see prompt engineering for lawyers.
Related reading: The best AI tools for lawyers | Prompt engineering for lawyers | AI for law firms