AI Tools for Solo Attorneys: What to Test Before You Rely on Them

Choosing AI for a small law firm starts with one question: which job do you want help with? A tool that produces a useful first draft may not be the tool you need for legal research, intake or billing.

Before moving client information into a new system, compare tools on the same fictional task. Check what each output gets right, what it invents, what you must verify and how much review it takes. The exercise and worksheet below give you a place to start. They are a way to evaluate a tool, not a certification that any product is safe or suitable for your practice.

Choose one job, not an entire AI stack

Write down a recurring task and the result you need. For example: turn a short document packet into a source-labeled chronology and an internal summary for attorney review. That is more testable than asking whether a tool can handle your practice.

Keep different jobs separate. A satisfactory document summary does not establish that a tool can find controlling authority, answer clients, calculate a deadline or reconcile a trust account. Each needs its own inputs, checks and limits.

The Iowa State Bar Association's vendor-vetting framework starts with the job, then considers ethical obligations, data handling, accuracy and operational fit. Use those categories to structure your questions rather than treating a product list as an endorsement. Vetting AI for Attorneys

Ask about data handling before using real matters

Get answers you can keep, not just a reassuring badge. Ask where inputs and outputs go, whether they are used to train models, how long they are retained, what deletion means, who can access them and which other providers process them. Read the applicable contract and privacy terms. An unanswered question stays unanswered.

ABA Formal Opinion 512 addresses existing duties involving competence, confidentiality, client communication and fees when lawyers use generative AI. It is model-rule guidance, not a substitute for checking your jurisdiction's rules, court requirements or the facts of an engagement. ABA guidance on AI tools

For the exercise below, use only the invented packet. Do not add a client name, real correspondence, credentials or confidential facts to make it more realistic. A synthetic test reduces the need to expose real information; it does not answer every security or suitability question.

Run this fictional document-review exercise

All names, documents and events in the following packet are invented. This is not a client matter. The separate source labels are intentional: they let you see whether the tool preserves the difference between a document's statement and an independently established fact.

The packet

• S1 - Accepted quote, October 5, 2026: Cedar Studio accepts Northline Supplies' written quote for 12 desks, total $6,000, with delivery by October 20, 2026.
• S2 - Payment note, October 6, 2026: The packet records a $2,000 deposit paid by Cedar Studio. No bank record is included.
• S3 - Northline email, October 22, 2026: Northline says eight desks arrived on October 21 and four remain outstanding. No delivery receipt is included.
• S4 - Cedar note, October 23, 2026: Cedar says two of the delivered desks were damaged. No photographs or inspection report are included.

No additional contract clauses, governing-law term, refund promise or later correspondence is supplied. Treat each item as the supplied document's account, not as proof of facts beyond that account.

Use the same prompt for each tool

Synthetic test only. Using only sources S1-S4 above, prepare: (1) a chronology with a source label for every material statement; (2) a list of missing facts and any conflicting accounts; and (3) a neutral internal first-draft summary under 200 words. Distinguish what a source says from independently verified facts. Do not invent contract terms, governing law, remedies, citations, deadlines or later events. Do not force a conflict where the sources merely leave something unknown. If the packet does not answer a question, say so. Do not send anything or change any account.

Keep the input and prompt unchanged when comparing tools. Record the tool, version or displayed model label, settings you chose and test date. If a tool cannot accept the task or provide usable source labels, record that limitation rather than silently changing the test to make it pass.

Check the answer against the packet

Start with the record, not the fluency of the writing. For each material statement, ask which source supports it and whether the output says more than that source does.

• Does it distinguish the promised October 20 delivery date in S1 from the reported October 21 arrival in S3?
• Does it attribute the partial delivery to Northline's email rather than claim a delivery receipt was checked?
• Does it attribute the reported damage to Cedar's note, without inventing an inspection?
• Does it preserve the missing four desks without assuming what happened after October 23?
• Does it leave refund rights, breach, governing law and legal deadlines unresolved?
• Does it ask for the complete terms, payment and delivery evidence, damage documentation and later communications?

The packet does not require a legal conclusion. An invented refund right or confident breach finding is not a helpful extra. Mark unsupported material statements and missing-fact guesses even when they sound plausible.

This exercise tests source handling and an internal draft. It does not validate legal research. For a separate research test, check that each authority exists, supports the exact proposition and quotation, applies in the relevant jurisdiction and is current, including subsequent treatment. An existing citation is not necessarily good law or relevant law. This worksheet is not a citator.

Copy this comparison worksheet

Use one copy per tool and task. Fill the same fields for your current process so the comparison includes the work you already do, not just two AI outputs. Do not put client information into the worksheet.

Test record

• Tool and displayed version/model: ____________________
• Test date and reviewer: ____________________
• Task, packet version and settings: ____________________
• Same input/prompt used across comparisons? Yes / No. Difference: ____________________

Effort, in minutes

• Setup and input preparation: ______
• Output generation: ______
• Corrections and redrafting: ______
• Independent factual/source verification: ______
• Complete-task total, counting overlapping time only once: ______
• Current-process total on the same task: ______

Quality and fit

• Material statements checked: ______
• Unsupported material statements found: ______
• Unknown facts incorrectly filled in: ______
• Important omissions or misattributions: ____________________
• Source labels usable? Yes / Partly / No. Example: ____________________
• Output and export usable in your workflow? Yes / Partly / No. Why: ____________________
• Data-handling answers kept in writing: ____________________
• Unanswered vendor or contract questions: ____________________
• Confirm no confidential information was used: ____________________

Decision

• Continue this narrow test / Change the test / Stop
• Reason, corrections needed and next reviewer: ____________________

A fast first draft is not a pass by itself. Consider stopping and investigating if the output invents an authority or material fact, cannot be traced to a source, or exposes information unexpectedly. One successful run also does not establish reliability. If the narrow test is promising, repeat it on other appropriate fictional examples, including incomplete or conflicting records, before considering a broader pilot.

Compare the complete workflow and cost

Before adopting a tool, check how it fits with your current system. Can you keep the original record, review output before use, export what you need and control who has access? Ask what onboarding and support involve, and what happens when you leave.

Compare the total price for the seats and features you need, any usage limits or additional charges, the contract period, renewal and cancellation terms. A low advertised starting price does not answer those questions. Review current terms directly rather than relying on an old comparison article.

For broader practice-software evaluation, see how to choose law practice management software. If you are evaluating Lexitio, review its current security information and pricing, then ask about anything those pages do not settle for your practice. Marketing pages are a starting point for diligence, not proof that a specific workflow will work for you.

Want to evaluate Lexitio on one workflow?

Request consideration for Lexitio's design-partner program. Describe the one job you want to evaluate and ask whether it is currently supported. Review the current program terms and admission process before applying.

Start the conversation with fictional material, not a client file. The exercise above is a general evaluation method; it is not a claim that Lexitio has completed or passed this test. Applying does not establish that a workflow is supported, grant access or authorize a charge.

You can use the worksheet whether or not you apply. Keep a tool when the checked work and complete workflow justify it for your practice, not because its first answer sounds polished.