The Small Firm Stack
Guide

Where does your client data go? A plain-English checklist for vetting AI tools

A simple checklist, tied to ABA Formal Opinion 512, for finding out what an AI tool does with your clients' information before you use it.

Last checked

When you type a client's facts into an AI tool, or let an AI answer your phones, that information goes somewhere. This guide gives you a short list of questions to answer before you use any AI tool with client information. Every vendor example below links to the vendor's own words.

This guide is general information, not legal advice. Check your own state's rules and ethics opinions.

Why this matters: what the ABA says

In July 2024, the American Bar Association's ethics committee issued Formal Opinion 512 on generative AI tools (ABA Formal Opinion 512, checked Oct 19, 2026). ABA opinions are guidance, and your own state's rules and ethics opinions control. Here is what it says, in its own words.

You have to look before you leap:

"Before lawyers input information relating to the representation of a client into a GAI tool, they must evaluate the risks that the information will be disclosed to or accessed by others outside the firm."

You have to read the fine print, or have someone who did:

"As a baseline, all lawyers should read and understand the Terms of Use, privacy policy, and related contractual terms and policies of any GAI tool they use to learn who has access to the information that the lawyer inputs into the tool or consult with a colleague or external expert who has read and analyzed those terms and policies."

Tools that learn from your inputs need client consent: For "self-learning" tools, the opinion says "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool." And it warns that "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient."

Your firm needs a policy: "Managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI."

The opinion lists vendor checks. Borrowing from earlier cloud computing opinions, it says lawyers should:

"ensure that the [GAI tool] is configured to preserve the confidentiality and security of information, that the obligation is enforceable, and that the lawyer will be notified in the event of a breach or service of process regarding production of client information"

"investigate the [GAI tool's] reliability, security measures, and policies, including limitations on the [the tool's] liability"

"determine whether the [GAI tool] retains information submitted by the lawyer before and after the discontinuation of services or asserts proprietary rights to the information"

(All quotes from ABA Formal Opinion 512, checked Oct 19, 2026.)

The checklist below turns those ideas into plain questions.

The checklist: 10 questions for any AI tool

Print this. For each tool, write the answer and where you found it.

1. Does it train AI on my data?

This is the big one, because it connects to the ABA's "self-learning" warning. Look for a clear "we do not train" line in the terms, not just on a marketing page.

  • Clear example: Lawmatics' terms say: "We do not use Customer Data to train or fine-tune artificial intelligence models" (Lawmatics terms, checked Oct 19, 2026).
  • Trains after removing names: Otter's privacy policy lists "training our proprietary AI technology on de-identified audio recordings and on transcriptions (which may contain Personal Information)" (Otter privacy policy, checked Oct 19, 2026).
  • Broad wording to ask about: 8am's FAQ says "we may use inputs and outputs to develop and improve our products" (8am IQ FAQ, checked Oct 19, 2026). LexisNexis says it never trains on customer data, but also that "Identifiable information is removed from AI interactions so that we can improve performance" (Protégé page, checked Oct 19, 2026).
  • Read every word: Smokeball says it "does not provide customer data to third parties for AI model training" (Smokeball security, checked Oct 19, 2026). That covers third parties. Ask what Smokeball itself does.

Red flag: "improve our services" with no mention of AI either way. Ask in writing.

2. Which outside AI companies see it?

Many AI tools send your text to a large AI provider. For example, 8am says 8am IQ sends documents to an OpenAI model (8am IQ FAQ, checked Oct 19, 2026), and Spellbook names OpenAI and Anthropic (Spellbook security, checked Oct 19, 2026). Find out which provider your tool uses, and whether that provider keeps anything.

  • Good example: Thomson Reuters says all requests to AI providers use "zero-retention API calls," and that OpenAI and Google are "contractually prohibited from using any customer data to train their models" (CoCounsel FAQ, checked Oct 19, 2026).
  • Good example: CallRail publishes a list that names AssemblyAI, OpenAI, Google Gemini, ElevenLabs, and others (CallRail trust center, checked Oct 19, 2026).
  • Vague example: Abby Connect says transcripts go to "a third-party AI service provider" without naming it (Abby privacy policy, checked Oct 19, 2026).

3. Where is it stored and processed?

Country matters if your clients or your bar care about data location. Be careful: storage and processing can be different places.

  • Clio's security page says AI data is "processed in your region," but its help center says AI features "may process your queries on servers located outside your home jurisdiction" (Clio security, Clio help, checked Oct 19, 2026). When a vendor's pages disagree, ask which is right.

4. How long is it kept, and what happens when I leave?

The ABA asks whether the tool keeps your data "before and after the discontinuation of services." Look for real numbers.

  • Clear exit plan: Lawmatics gives you 60 days to retrieve data after you leave, then deletes it within 30 days (Lawmatics terms, checked Oct 19, 2026).
  • Fixed delete date: CallRail deletes call records and recordings after 25 months and says you cannot extend that, so legal holds mean exporting yourself (CallRail retention policy, checked Oct 19, 2026).
  • Deleted when you leave: Posh says all data is deleted after your account ends and it will not send you data from a closed account (Posh terms, checked Oct 19, 2026).

5. Who owns it?

You want "you own your data." Watch for anything less.

  • Smith.ai's terms say: "All Subscriber Information is jointly owned by Smith and you" (Smith.ai terms, checked Oct 19, 2026).

6. Will they tell me about a breach, and how fast?

The ABA says you should be notified of a breach. Look for a number of hours or days.

7. Is there an independent security audit?

A SOC 2 Type II report or ISO 27001 certificate means an outside auditor checked the vendor's controls. Ask to read the report, not just see a badge. Check that the audit is for the vendor, not only its cloud host.

8. What is their liability if something goes wrong?

The ABA mentions "limitations on the [the tool's] liability." Some caps are very low.

  • Ruby caps liability at the lesser of 15 days of fees or $500 (Ruby data addendum, checked Oct 19, 2026). Posh uses the same cap and requires written notice of a claim within 14 days (Posh terms, checked Oct 19, 2026).

9. Can I turn AI off?

Some tools let an admin switch AI off for the whole firm. Clio's help center explains how, but also says that if your license includes AI features, firm level AI access "stays on" and to contact support (Clio help, checked Oct 19, 2026).

10. Will they sign a BAA if I handle medical records?

Firms that handle injury or disability cases may hold clients' medical records. If you need a business associate agreement, confirm the vendor will sign one, and for which product. See each listing's Data Trail card.

What to do with your answers

  1. Sort tools into three groups: clear "no training" plus a named AI provider; unclear; and trains on your data.
  2. For unclear tools, email the vendor the questions above and save the reply.
  3. For tools that learn from inputs, do not put client information in without informed consent that meets the ABA's standard, or use them only for tasks with no client information.
  4. Write a one page firm AI policy that lists approved tools and what may go into each. The ABA expects managers to set one.
  5. Recheck every year, or when a vendor updates its terms.

Our take

This is our opinion. Most vendors in our directory have at least one line worth asking about. That is normal. The goal is not to find a perfect tool. It is to know exactly what you agreed to, so you can explain it to a client if they ask.


Last checked: Oct 19, 2026. Published by the team behind The Small Firm Stack, who also build phone and intake systems for law firms. See our About page for the full disclosure. We never list, rank, or review ourselves. No vendor paid for this page and it has no affiliate links. Not legal advice.

Sources

Vendor pages and public records this page is based on.

  1. americanbar.org · /content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
  2. lawmatics.com · /terms-of-use
  3. thomsonreuters.com · /en/cocounsel
  4. otter.ai · /privacy-policy
  5. otter.ai · /privacy-security
  6. supportcenter.mycase.com · /en/articles/9370197-8am-iq-faqs
  7. lexisnexis.com · /en-us/products/protege.page
  8. smith.ai · /receptionists/terms
  9. ruby.com · /data-processing-addendum/
  10. posh.com · /terms/
  11. trust.callrail.com · /
  12. support.callrail.com · /hc/en-us/articles/5711337552269-CallRail-s-data-retention-policy
  13. clio.com · /security/
  14. help.clio.com · /hc/en-us/articles/41990965598491-AI-Features-in-Clio-Manage
  15. smokeball.com · /security
  16. abbyconnect.com · /privacy-policy/
  17. spellbook.legal · /security