AI intake for law firms is a safer first project than an AI legal assistant, and OpenAI dots for Texas law firms fit it. Intake administration needs no legal judgment, the pilot can run on fictional records so no client confidence is at stake, and every output stays a draft that a person must act on. A dot can keep a review queue moving between conversations: list the inquiries nobody owns, draft a summary for the intake manager, and ask a person when a decision is needed. This article lays out a proposed pilot. We have not run it at a client firm, and OpenAI has not reviewed it.
The short version
- Start with fictional records and drafts that stay inside the firm. Settle the data questions before a real inquiry goes anywhere near a dot.
- Keep intake administration apart from conflict checks, legal advice and the decision to take a matter.
- Dots and Astra for Law are separate products with separate access. Having one tells you nothing about the other.
- Score the pilot on the completed handoff, meaning the right person had the right record, not on how many AI tasks ran.

Where an inquiry loses its owner
Picture a fictional Dallas family-law practice. The marketing report counts a completed web form as a lead. The intake team sees a missing callback. The lawyer opens an incomplete record with no reliable account of what the person asked. Three groups are looking at one event through three systems, and a faster writing tool does not tell any of them who owns the next step.
Start by drawing the handoff the firm already has: inquiry received, assigned to staff, reviewed under the intake process, consultation arranged if appropriate, disposition recorded. Then list the fields staff need. A test record might carry a fictional identifier, a received time, a preferred contact method, a practice category and an assigned owner. Nothing in it should guess at the merits of a case from a two-line message.
Real queues have wrinkles, so plan for one. As a hypothetical example, a web form arrives at 9 p.m. on a Saturday from someone who writes mostly in Spanish. The record has an owner field, but the only bilingual staff member is off until Monday. The question for the pilot is not whether the dot can answer the person. It is whether the Monday report shows that record near the top with the weekend arrival time and the language preference intact, so the right person picks it up first.
The same boundary holds in a Houston injury practice or an Austin criminal-defense office. A practice category tells staff where to route a message. It does not establish a deadline, a legal status or whether the firm can take the person on. Those calls belong to the firm's approved process and its lawyers. Our Texas law firm marketing services cover the website and acquisition side of the problem, and we would settle this boundary before any accounts are connected.
What a dot does and what Astra for Law does
OpenAI describes a dot as an always-on agent in ChatGPT that keeps work moving and checks in when a decision is needed. In OpenAI's words, a dot "works between conversations," tracks progress and follows through as things change. It uses supported plugins that are installed and enabled for your account.
Availability is narrower than the name suggests. The same page says dots are rolling out gradually to eligible accounts, and that on Enterprise they are off by default and must be enabled by a workspace administrator. A firm that wants to test this should ask whoever administers its ChatGPT workspace before it plans anything.
For this pilot, the dot's whole assignment is small. It maintains an internal exception list from the approved test records and drafts a summary for the intake manager. It does not email, call, publish or edit a matter record.
Astra for Law does different work. OpenAI describes it as GPT-6 Astra built for professional legal work, paired with a Legal Search Index that covers U.S. case law, statutes, regulations, court rules and administrative decisions. Access starts with selected U.S. law firms through Trusted Access and Codex, for eligible lawyers and people working under their supervision, with API access listed as coming soon. The page tells users to review the answers and cited sources before relying on them.
An authorized lawyer might use that environment to look into a legal question after the firm's intake and data checks are done. The lawyer decides which facts go in, verifies what comes out and chooses what returns to the administrative queue. A dot is not automatically using Astra for Law, and neither product name shows that a case-management connection exists or is permitted.
| Queue item | Proposed AI assistance | Human decision |
|---|---|---|
| Unassigned inquiry | Flag the missing owner in an internal report | Staff assigns responsibility |
| Incomplete test record | List missing fields without guessing answers | Staff follows the approved intake process |
| Legal question | Prepare a research brief in an approved environment | Lawyer verifies authorities and evaluates the issue |
| Potential marketing topic | Draft an anonymous topic suggestion | Lawyer and editor approve a public explanation |
| Ready-to-send message | Prepare wording from an approved template | Authorized person reviews recipient and content |
Test with fictional records before connecting an inbox
Build a test set from the problems your staff already recognize: a duplicate inquiry, an item with no owner, a record missing a phone number, a message outside the firm's practice, and an inquiry the sender marked urgent. Every person and fact stays fictional. The exercise tests routing and reporting. It does not test legal judgment.
Write the expected result before the dot runs. A duplicate should be flagged for staff, not deleted. A missing phone number should stay missing. An urgent label should be escalated under the test procedure and should not turn into an invented court deadline. A record outside the practice should not trigger an automatic rejection.
Then have the intake manager compare the report with the expected answers, and log missed exceptions and false alarms separately. If the dot keeps missing the unassigned record, narrow the task or fix the source format before adding another tool. A miss on a made-up record costs nothing. The same miss on a real one means a person's message sits unanswered.
Our proposed pilot instruction: use only this fictional intake test sheet. Identify missing owners, duplicate-looking records and incomplete fields. Preserve the original text and mark uncertainty. Prepare an internal review list. Do not contact anyone, decide conflicts, assess legal merits, set deadlines or change the source records.
If a firm later considers real records, its responsible lawyer should first look at confidentiality duties and the actual data flow. Removing a name does not sanitize a record when the remaining facts still point to a person or a matter. And an inquiry from someone who has not yet become a client is not automatically safe for a marketing workspace.
Reading Opinion 705 before real data enters
Texas Professional Ethics Committee Opinion 705, issued in February 2025, covers generative AI under the Texas Disciplinary Rules. It says a lawyer should be "reasonably satisfied that the program will not reveal confidential information to others," and it points to four precautions: understand how the technology works, review the terms of service, learn the tool's data-security protections, and train lawyers and staff. It also says lawyers should verify the accuracy of AI responses, and that hourly fees cannot include time "saved" by using the tool. It endorses no vendor.
For this pilot, we would keep a one-page review sheet: the vendor, the account type, the data approved for use, the connected tools, who can read outputs and who can change permissions. The responsible lawyer and the technical owner settle open questions before real data goes in, and the sheet stays next to the workflow instructions so a staff change does not erase the reasoning.
Keeping the release decision with a person

OpenAI's page on dot controls says an automatic review checks an action against your instructions, permissions, custom rules and built-in safety requirements before it affects your accounts or shares information. It also says custom rules are instructions the dot tries to follow and that it can make mistakes, and that asking a dot to draft replies does not give it permission to send them.
We would build on that distinction. For a first pilot, the dot drafts and a person acts. Give staff a plain status sequence, and record which version was approved. If a draft callback message changes, the old approval does not carry over. If the recipient changes, check the message again before it leaves the firm.

Decide where the report goes and when the pilot ends, and use the office's actual time zone. A report that lands after staff have left can look like a managed queue while nobody is there to act on it. Name a fallback owner and keep a manual queue. OpenAI's page also warns that stopping a task does not undo actions a dot has already taken, which is another reason to keep sending, publishing and record changes out of the first pilot. If an app connection expires or the source sheet cannot be read, the report should say so. Silence must never mean that nothing needs attention, so test that failure with the fictional set.
Turning approved questions into useful marketing
A reviewed intake pattern can suggest a public article, but private messages should not be pasted into a content prompt. Ask the responsible lawyer to approve a general topic first. A fictional Dallas practice might want a clearer explanation of how to request a consultation. A Houston firm might need a page that says what kinds of inquiries it handles.
Our Texas legal content marketing page covers that editorial work. For one market in detail, see Dallas law firm SEO and marketing. Do not invent a Dallas office, a success story or a credential to make a page sound local.
Before anything is published, a lawyer should assess the actual advertisement. Our Texas attorney website advertising guide is related reading, and AI-drafted wording is not an exception to the firm's review process.
Measuring whether inquiries reach the right person
For the pilot, track records reviewed, missing-owner exceptions found, exceptions missed, corrections required and staff review time. Record the failures even when the final report looks polished. Those figures come from the firm's own test.
Firms that also manage California content can read our sister brand's California dots and Astra editorial workflow. Both sites are ByteZero brands, and each state's workflow needs its own review. For a lower-stakes first step, the free AI toolkit for Texas law firms walks through a test with fictional records.
Frequently Asked Questions
Can lawyers use AI?
Yes, with conditions. Texas Ethics Opinion 705 (February 2025) says lawyers should understand the technology, protect confidential client information, verify the accuracy of AI responses and not bill hourly for time saved. This article applies that to intake administration. A firm's lawyers still decide what any tool may touch.
What is a useful first dots workflow for a Texas firm?
An internal exception report built from fictional intake records is a practical proposed pilot. Test missing owners, incomplete fields and duplicates before considering real inquiries. The report goes to a named staff reviewer.
What should people not tell an attorney in a first message?
That is a question for the firm's own lawyers. As a design choice for an intake form, we would collect a name, a contact method and a practice category, and leave the facts of the matter for a lawyer to ask about. The wording of any warning on the form should come from the firm's lawyer.
What is the 80/20 rule for lawyers, and does it help pick a pilot?
The 80/20 rule is a rule of thumb that a small share of causes produces most of an effect. It is not a measured figure for any firm. As a planning heuristic, look for the one intake step where inquiries most often lose their owner and test that step first.
Is Astra for Law available to every Texas firm?
No. OpenAI describes initial access for selected U.S. law firms through Trusted Access and Codex, for eligible lawyers and people working under their supervision, with API access coming soon. Contact OpenAI before building a workflow that depends on it.
Does this guide prove the tools work together?
No. It describes a supervised handoff and a test plan, not a verified product integration or a client deployment. Confirm the tools and permissions your account has, then validate the actual setup on fictional records before using real information.
About this guide: Prepared by the ByteZero SEO Team for Law Firm Marketing Texas. It is educational marketing and operations guidance, not legal advice, and no attorney reviewed it. Product pages were checked on November 5, 2026. This is a proposed pilot, not a tested integration and not an OpenAI endorsement, and AI assisted with the drafting and illustrations. A firm's lawyers should evaluate any legal use.



