ChatGPT is now an ad channel for US law firms, and it is a small one. OpenAI permits ads for legal services in the US only where the advertiser is licensed to practice law, its ads reach only some of ChatGPT’s users, and no firm has more than a few weeks of data. That makes it a good place for a capped, measured test and a poor place for a large budget. This guide sits under our AI search for law firms pillar, because ads in ChatGPT and being named in ChatGPT’s answers are two different things, and firms confuse them.
What OpenAI’s policy allows
OpenAI changed its ad policy so that ads for legal advice, representation and legal services are allowed in the US “only when the advertiser is licensed to practice law in the jurisdiction where the ad is shown” (OpenAI). Trade coverage dates the change to August 31, 2026 (LawLytics, secondary). Our dated note on the policy change records it as news; this guide is the working method.
Three parts of the policy matter in practice:
- The license test is place by place. A firm whose lawyers are licensed in Texas only should not let its ads show in Oklahoma. Location settings have to match licenses, state by state.
- US only. Even a firm with a foreign office can’t run legal-services ads in ChatGPT outside the US under this policy.
- Practice areas are not singled out for exclusion. Reporting on the updated wording says it names immigration and personal injury among the allowed services for licensed lawyers (LawLytics, secondary). Other platform limits, below, still narrow where those ads show.
OpenAI’s pages block automated reads, so we checked the wording through search extracts and trade coverage on October 24, 2026. Read the live policy yourself before you launch.
Who sees ChatGPT ads, and where
OpenAI says ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts don’t see ads, and neither do accounts OpenAI identifies as belonging to people under 18 (OpenAI Help). Temporary chats show no ads (same source).
Ads appear below a relevant conversation, not inside the answer, and OpenAI says they “remain distinct” from ChatGPT’s responses (PPC Land, secondary, quoting OpenAI). Each ad carries an advertiser name, a small icon, a headline, a description, a landing page and an image (same source).
The sensitive-topic limit and what it means for injury
OpenAI says ads are not eligible to appear near sensitive or regulated topics such as personal health, mental health or politics (OpenAI Help). Advertisers in some regulated fields may qualify under strict criteria, but even then their ads are not placed next to personal health conversations (same source).
For a personal injury firm, that is a real constraint. A person describing a neck injury after a crash is having a health conversation. Expect fewer eligible placements than the size of ChatGPT suggests, and expect them to lean toward questions about process (“how long does a car accident claim take”) rather than symptoms. Criminal defense, family law and immigration conversations may also brush against sensitive topics. The pilot will show what actually serves; don’t plan on volume you can’t see yet.
How buying works, as far as OpenAI documents it
OpenAI opened a self-serve Ads Manager to US businesses in a beta announced May 5, 2026 (OpenAI; PPC Land, secondary). As reported:
| Topic | What is reported | What it means for a firm |
|---|---|---|
| Account access | Sign up, then wait for OpenAI to verify the business; ads don’t deliver until setup and billing are complete | Open the account in the firm’s name, on the firm’s card, never an agency’s |
| Bidding | A Reach objective billed per thousand impressions, or a Clicks objective billed per click with a maximum bid | Start on clicks, so spend follows visits you can track |
| Targeting | No exact-match keywords. Advertisers add “context hints” describing the conversations an ad suits; OpenAI weighs those with the landing page and ad text | Write hints about legal questions, never about people’s traits or situations |
| Measurement | A pixel and a Conversions API for actions after the click; aggregated reporting only | Track calls and forms on your own landing page and phone number as well |
| Data | Advertisers don’t see individual conversations | Nothing about a person’s chat ever reaches your firm, and nothing about your clients should reach OpenAI |
Two cautions. First, these details come from OpenAI’s announcement as reported in the trade press, and the product changes quickly; confirm them in Ads Manager on the day you build. Second, context hints guide matching but don’t guarantee where an ad runs, so watch the placement and search-like reports weekly.
Our rules on targeting
We never write context hints aimed at a person’s health, legal trouble, immigration status or finances, and we never upload client lists. Hints describe the question, such as “questions about filing a car accident claim,” not the person asking it. The same policy applies in every channel we run: no Customer Match or lookalike audiences built from client lists, in any practice.
What a short ad must still say
A ChatGPT ad is a communication about a lawyer’s services, so the ABA’s model standard applies: nothing false or misleading, including a true statement that leaves out a fact needed to keep it from misleading (ABA Model Rule 7.1). The model rules also ask every communication to include the name and contact information of at least one lawyer or firm responsible for it (ABA Model Rule 7.2). States add more:
- California requires a lawyer or firm name and an office location in every ad (§6157.2). In a headline and a short description, that takes planning.
- Missouri requires a set sentence in every ad, and Alabama requires the location of a bona fide office, according to our state rules library.
- Results and “expert” claims are limited almost everywhere. Leave them out of a ChatGPT ad entirely.
Our default for every ChatGPT ad: the responsible lawyer’s name, the office city, no results, no comparisons, no “specialist” wording unless your state allows the exact phrase, and nothing that implies OpenAI or ChatGPT recommends the firm. Your attorney approves each version in writing before it runs.
Does my ChatGPT ad need a bar filing?
Filing rules generally follow the kind of communication, not the platform, and no state we know of has written a ChatGPT-specific rule. So the question is whether your state files paid internet ads. The checklist below comes from the filing facets in our state rules data, checked on October 24, 2026; each state page carries the sources and a verification badge.
From Probative's state rules data, filing facets
Does my ChatGPT ad need a bar filing? Ten states have some filing or review step; the rest have none
| State | What the state requires | Status | Likely step for a ChatGPT ad |
|---|---|---|---|
| Florida | File at least 20 days before first use; internet banner ads are listed; firm websites exempt | P | File before launch unless the ad holds only safe-harbor content |
| Texas | File non-exempt ads with the Advertising Review Committee within 10 days after first use | P | File within 10 days of launch |
| Nevada | File within 15 days after first use | P | File after launch; check the internet-ad guidance |
| Louisiana | File at or before first use; safe-harbor-only ads exempt | P | File at launch |
| Mississippi | Submit before first use; some basic-information ads exempt | P | File before launch |
| Connecticut | File at or before first use through e-services | P | File at launch |
| Kentucky | Voluntary advisory review | P | Optional review before launch |
| Indiana | Only targeted solicitations filed | S | Usually none for a general ad |
| South Dakota | Only targeted solicitations filed | S | Usually none for a general ad |
| Wisconsin | Only targeted solicitations filed | S | Usually none for a general ad |
| Alabama | No filing since January 1, 2026; copies kept 6 years; an office location in every ad | P | Keep a dated copy and show the office city |
| New York | No filing, label or retention rule since June 1, 2026; a name and contact in every ad | P | Keep a dated copy of each version |
| All other states and DC | No filing found | P or S | Keep a dated copy and the attorney’s approval |
Three practical points. A changed ad is usually a new ad for filing purposes, so ChatGPT’s automatic variations, if any appear, need a check against your state’s rule. Keep a dated copy of every version and every image, with the approving attorney’s name, whatever your state. And where your state allows an advisory opinion before first use, a new channel is a good reason to ask for one.
Tracking a ChatGPT visit to a signed case
ChatGPT ads are judged the way every channel should be: on signed cases from your own records, not clicks. The setup:
- A dedicated landing page for ChatGPT traffic, written to answer a question first, because visitors arrive from a conversation, not a search box.
- A tracked phone number on that page only, so calls carry their source into your CRM.
- UTM tags on the landing page address, which Ads Manager supports (PPC Land, secondary).
- OpenAI’s pixel or Conversions API for form submissions, set to send a neutral event name such as “lead” and never anything about the matter.
- A CRM source field that holds “ChatGPT ads” from first contact to signed case, so the Monday report can count each stage.
The track signed cases guide explains the stage names and value bands we use on Google; for ChatGPT we track the same stages but send nothing back except the neutral lead event, because OpenAI has not published an offline signed-case import.
The holdout readout
A small channel can take credit for cases your firm would have signed anyway. A holdout separates the two. Before launch, pair your service areas into similar groups and run ChatGPT ads in one of each pair, holding the other out. At the end, compare signed cases per thousand residents between the two sets, with everything else held as steady as you can. The method is set out in The Probative Standard’s holdout test.
Worked example: a fictional estate-planning firm
IllustrativeA 90-day pilot readout: spend, stages and the holdout, on one page
| Line | Test areas (ads on) | Holdout areas (ads off) |
|---|---|---|
| ChatGPT ad spend | $6,000 | $0 |
| ChatGPT clicks | 1,450 | 0 |
| Leads with ChatGPT as source | 38 | 0 |
| Qualified leads | 21 | 0 |
| Consults held | 14 | 0 |
| Signed matters, all sources | 52 | 44 |
| Signed matters per 10,000 residents | 4.1 | 3.6 |
| Readout | Direction is positive; volume is small. Extend 60 days at the same cap before deciding |
In this example the honest answer is “probably helping, not yet proven.” That is a fine result for a pilot. What isn’t fine is a report that says “38 leads from ChatGPT” and stops there.
How we would run a pilot
| Step | What happens |
|---|---|
| Eligibility | List every state where your lawyers are licensed, limit locations to those states, and check your state’s filing rule |
| Written plan | A budget cap, start and end dates, what counts as success, and what ends the pilot early |
| Ads and pages | Ad text and a landing page drafted to your state’s rules; your attorney approves each version |
| Tracking | A tracked page and number, UTM tags and a CRM source field, tested before launch |
| Weekly review | Spend, placements, leads and stages, with any off-topic placements cut |
| Readout | A written readout at day 90 with the holdout, and a recommendation to stop, hold or grow |
What this means for your firm
- Treat ChatGPT ads as a test, not a channel yet. Cap it, date it, and decide in writing what result would end it.
- Match locations to licenses. OpenAI’s policy and your bar both require it.
- Check your filing rule before the first ad. If your state is in Exhibit A, plan the filing into the launch date.
- Put the required words in the ad itself. Name and office city, at minimum, with no results or “expert” claims.
- Don’t buy ads to get named in answers. Ads don’t change answers. Being named in ChatGPT’s responses comes from what the engines read about your firm; see AI search for law firms.
- Measure in signed cases with a holdout. Anything less is a guess.
Probative is AI-first SEO and ads for law firms. Our ChatGPT ads service runs the pilot above as a fixed 90-day plan, with a flat fee shown on the pricing page. The organic side, being named in AI answers, is our AI search service. For state-by-state ad rules, see the lawyer advertising rules guide.
Marketing information, not legal advice. A ChatGPT ad is lawyer advertising under your state’s rules; check with your bar and your own counsel before you run one.