Everything in this guide is sourced from OpenAI's published ad policies, Help Center documentation, and official announcements, supplemented by reporting from Reuters, CNBC, TechCrunch, and PPC Land, and consumer research from iLawyerMarketing and the Clio Legal Trends Report. This platform is changing monthly. Facts are current as of September 2, 2026, and where something is a beta feature, a reported plan, or my own strategic projection rather than a confirmed fact, I say so explicitly.
One warning before anything else: most of what has been written about ChatGPT ads for lawyers is already wrong. OpenAI prohibited legal services ads at launch in February 2026, and the majority of legal marketing articles still say so. The policy changed in August 2026. Read the rules chapter carefully, because the opportunity it describes is one most of your competitors do not yet know exists.
Table of Contents
- What ChatGPT Ads Is and How Fast It Became Real
- Why This Channel Matters for Law Firms
- The Rules: What OpenAI Allows for Legal Services
- What ChatGPT Ads Look Like: Formats and Placement
- How to Advertise on ChatGPT: Campaigns, Budgets, and the Auction
- Targeting: How You Actually Reach the Right Person
- The Personal Injury Campaign Build, Step by Step
- Strategy by Practice Area
- Compliance: Bar Rules Still Apply
- Paid and Organic: Ads Are Half the AI Strategy
- When to Start, the Early-Mover Advantage, and a 90-Day Pilot Plan
- Where the Platform Is Heading
- Frequently Asked Questions
What ChatGPT Ads Is and How Fast It Became Real
On January 16, 2026, OpenAI announced it would begin testing advertising inside ChatGPT. On February 9, the first ads went live for US users on the Free and Go tiers. By April, OpenAI reported the pilot had crossed $100 million in annualized revenue within roughly six weeks. A self-serve Ads Manager opened in May. Through August the platform expanded to the UK, Japan, South Korea, Brazil, Mexico, 31 European markets, and India, more than 30 new markets in a single month. On August 31, 2026, OpenAI announced ChatGPT Ads had reached $1 billion in annualized revenue run rate, less than 200 days after launch (Reuters, August 31, 2026).
Ads in ChatGPT work differently from every ad product before them, and the essentials are all confirmed by OpenAI's own documentation:
- Where ads appear: at the bottom of ChatGPT's answers, in a clearly labeled sponsored unit, on the Free and Go subscription tiers. Plus, Pro, Business, and Enterprise users never see ads.
- Answer independence: OpenAI's stated core principle is that ads never influence the organic answer. The model's response is generated first and independently; the ad is matched to the conversation afterward. Paying does not make ChatGPT recommend you in the answer itself.
- Who sees them: adults only. Users identified as under 18 are excluded, and entire categories of sensitive conversations never show ads at all.
- How they are bought: through OpenAI Ads Manager at ads.openai.com (currently labeled beta), through OpenAI's Ads Solutions team for managed accounts, and through agency and technology partners.
Scale of the audience: ChatGPT serves roughly 800 million weekly users globally, and the majority sit on the ad-supported free tier. For context on how quickly the advertiser side is filling in, an Adthena analysis of the US pilot found ads appearing on about 24 percent of eligible queries, with retail brands taking 39 percent of impressions (PPC Land, August 2026). Read that carefully: three-quarters of eligible conversations still show no ad at all, and the legal category barely registers yet. That is what an unclaimed channel looks like in its first year.
Every prior advertising land grab (Google Ads in the early 2000s, Facebook in the early 2010s, TikTok in 2020) rewarded the advertisers who learned the platform while inventory was cheap and competitors were skeptical. ChatGPT Ads is at that stage right now, and for legal services specifically, the door only opened in August 2026. The firms that build competence in the next two quarters will buy attention at prices that will not exist in two years.
Why This Channel Matters for Law Firms
The case for caring is not that ChatGPT ads exist. It is that your future clients are already inside ChatGPT asking the questions that used to go to Google, and they are asking them in a more revealing way.
Sources: iLawyerMarketing 2026 Consumer Study (1,110 US consumers); Clio Legal Trends Report.
Two properties make the channel unusual for legal marketing. First, the intent is richer than a search query. A Google search gives you "car accident lawyer near me," five words. A ChatGPT conversation gives the platform something closer to a full intake narrative: "I was rear-ended on I-25 last week, the other driver's insurance offered me $8,000, my neck still hurts, should I accept it or talk to a lawyer first?" The person describes their situation, their objection, and their stage of decision in natural language, and the ad system matches against all of it. No advertising platform in history has had access to intent expressed this completely.
Second, the conversation happens earlier in the journey. People take legal questions to ChatGPT while they are still deciding whether they have a case at all, days before they would ever type a lawyer search into Google. Reaching someone at that stage, before they have seen a competitor's name, is the same logic that makes top of funnel strategy work, except here the "content" moment and the "ad" moment happen in the same window. The firm that appears while the question is forming shapes the shortlist everyone else has to fight into later.
This sits inside a bigger shift we mapped in the AEO and SEO guide for law firms: AI assistants are becoming a primary discovery layer for legal services, and visibility inside them now has two components, the organic answer and the paid unit beneath it. This guide covers the paid half; chapter 10 explains how the two halves reinforce each other.
The Rules: What OpenAI Allows for Legal Services
This is the chapter to read twice, because the policy has flipped twice in seven months and most published guidance is out of date.
The policy timeline
The current rule, quoting the operative conditions from OpenAI's ad policies (v1.5, August 2026): ads for legal advice, representation, or legal services are permitted in the US only, and only when the advertiser is licensed to practice law in the jurisdiction where the ad is shown. The policy explicitly includes services related to immigration, personal injury, legal claims, and document preparation. Legal services is treated as a regulated category, which means manual, case-by-case advertiser approval rather than instant automated onboarding, and you should expect to provide proof of licensure during account verification.
Three practical consequences follow directly from the licensing language:
- Geographic targeting is not optional for law firms; it is a compliance requirement. If your attorneys are licensed in Colorado, your ads can show in Colorado. The jurisdiction rule is what makes the state-level targeting in chapter 6 the backbone of every legal campaign.
- Multi-state firms have a structural advantage. A firm licensed in six states can build six state-targeted campaigns; a single-state firm cannot follow. This compounds the expansion logic in our guide to scaling a law firm into multiple states.
- Expect friction, and treat it as a moat. Manual review means slower onboarding, creative rejections, and conservative interpretations, especially early. Every hoop that slows you down is also slowing the 50,000 other firms that will eventually want in.
Placement exclusions still protect sensitive moments
Separate from who may advertise, OpenAI maintains placement rules about where ads may appear. Conversations involving mental health, acute emotional distress, and other sensitive contexts do not show ads regardless of advertiser category. For a personal injury firm this matters: a user describing a crash in visibly traumatic terms may never see your ad, while the same user asking practical questions about insurance settlements will. Do not expect impressions on every relevant conversation, and recognize that the filter also protects your brand from appearing in moments that would reflect badly on it.
Verify before you spend. This policy has changed twice in seven months and will change again. Before committing budget, read the live policy at openai.com/policies/ad-policies yourself and confirm the current rules for legal services in your states. Any agency pitch that cannot quote the current policy version is a red flag, and plenty of agencies are still selling guidance written in June 2026 that says the opposite of what the policy says today.
What ChatGPT Ads Look Like: Formats and Placement
The core unit is deliberately simple. After ChatGPT finishes its answer, a single sponsored card can appear beneath it, visually separated from the response and labeled as sponsored. The card contains:
- Advertiser name and favicon, pulled from your verified brand identity
- Title, a headline capped at 50 characters
- Copy, a description capped at 100 characters
- Image, a single creative asset
- Landing page URL, where the click goes
Before accepting, it is worth understanding what the offer covers. Early settlement offers are often made before the full extent of injuries is known, and accepting one usually waives your right to seek more later. A few things to consider:
1. Ongoing symptoms like neck pain can indicate injuries that have not fully presented yet...
2. The offer should account for medical bills, lost wages, and future treatment...
Recreated illustration of the placement, styled for this guide. The organic answer always renders first and is never altered by the ad; the sponsored card appears below it, clearly labeled. Format per OpenAI Ads Manager documentation, September 2026.
The anatomy of the ad card
Beyond the standard card, the format set is expanding. Confirmed and in market or beta as of September 2026: an e-commerce format with price and review data (retail-focused, launched May 2026), and call-to-action buttons inside the ad unit such as Book now, Sign up, and Learn more, in early testing with the button auto-selected by the system. Reported but not yet confirmed by OpenAI documentation: conversation-capable ad experiences, pay-per-action pricing, video formats later in 2026, and ad placements inside third-party custom GPTs with creator revenue sharing. Plan your strategy on the standard card; treat everything else as upside.
Three properties of the unit shape how you should write for it:
- One slot, not ten. A Google results page holds several ads plus organic listings. A ChatGPT answer shows at most one sponsored unit in the conversation view. Winning the impression means being the single most relevant advertiser for that conversation, which makes the targeting quality in chapter 6 matter more than bid size alone.
- It appears after the answer. The user has just read a thorough, neutral explanation of their situation. An ad that repeats the answer adds nothing; an ad that offers the logical next step (a case evaluation, a settlement review, a local specialist) completes it.
- Trust context is unforgiving. Users are in a conversation they experience as private and helpful. Hype-heavy creative that would be normal on Facebook reads as invasive here. The ads that work borrow the assistant's own tone: calm, specific, useful.
User controls exist and matter. Users can dismiss ads, opt to see fewer of them, and control whether their chat history personalizes ad matching. OpenAI has been explicit that user trust outranks ad revenue, which is why formats stay conservative. Build creative that survives in that environment rather than fighting it.
How to Advertise on ChatGPT: Campaigns, Budgets, and the Auction
How to run ads on ChatGPT, mechanically: everything is managed in OpenAI Ads Manager at ads.openai.com. The hierarchy will feel familiar if you have run Google or Meta campaigns, with one important difference at the ad group level:
Here is the campaign creation screen itself, recreated with the settings a Colorado personal injury pilot would use, and the six decisions it forces:
Recreated from OpenAI Ads Manager Beta, September 2026, with an example law firm configuration. Fields and defaults may change as the beta evolves.
Objectives and pricing. The platform launched on impression-based pricing (reported at a $60 CPM in the February US launch) and has since added click-based buying, with conversion-optimized bidding in development per industry reporting. In Ads Manager you choose an objective, set either a daily or total budget, and the system runs a relevance-weighted auction: your effective rank is a function of bid and predicted relevance to the conversation, so a highly relevant ad can beat a higher bid, the same principle as Google's quality score. Reported minimums for self-serve have been low (India launched with a daily minimum near $8), which means a firm can pilot this channel for less than the cost of one Google Ads click in a competitive PI market.
Measurement. Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM, and conversions where conversion measurement is configured, viewable as tables, trend charts, and CSV exports. That is the platform layer, and on its own it will mislead you: platform metrics stop at the click, and a law firm makes money at the signed fee agreement, five steps later. Tracking this channel properly means wiring five layers so a dollar of spend is traceable to a signed case. This is the same end-to-end chain we rebuild in marketing attribution engagements, applied to a brand-new source.
The five-layer tracking stack
Layer 2, the UTM convention, is where most tracking dies, so fix it in writing before launch. One pattern, used on every ad, no exceptions:
utm_source=chatgptandutm_medium=paidon every URL, so the channel aggregates cleanlyutm_campaignmatching the campaign name exactly (pi-colorado-evaluation)utm_contentcarrying the ad group and ad variant (car-accidents-settlement-review-a)
With that discipline, your analytics can answer "which ad group produced the consultation" instead of lumping everything into referral traffic. Without it, ChatGPT visits scatter into direct and referral buckets and the channel becomes unmeasurable within a month.
Connecting it to GA4
Google Analytics 4 does not know ChatGPT Ads exists, so out of the box the traffic lands in the wrong buckets: clicks from the ChatGPT app can arrive with no referrer and show as Direct, and untagged web clicks show as chatgpt.com referral, mixed in with organic ChatGPT mentions of your firm. The UTM convention above is what fixes this, and four GA4 configurations make the channel fully readable:
- Verify the tags land. With
utm_source=chatgptandutm_medium=paid, GA4 records session source/medium as chatgpt / paid. Run one test click from a tagged URL and confirm it in Reports, then in Explorations, before spending. - Build a custom channel group. Create a channel called "AI Paid" defined as source = chatgpt and medium = paid, and a second called "AI Organic" for source matching chatgpt.com, perplexity, gemini, and copilot referrals with any other medium. Now paid ChatGPT traffic, organic AI referrals, and Google Ads sit as three separate lines in every report, which is exactly the comparison a managing partner needs to see monthly.
- Mark the money events as key events. Form submission, tracked call, chat started, and consultation booked should be GA4 key events, with your call tracking platform pushing call events into GA4 so phone conversions attribute to the chatgpt / paid session that produced them.
- Read it in Explorations, not just standard reports. A simple exploration with session source/medium and session campaign as rows, plus sessions, engagement rate, key events, and event value as columns, filtered to source chatgpt, gives you ad-group-level performance (via utm_content) that standard reports flatten.
Two GA4 honesty notes. Attribution in GA4 is click-based, so the view-through effect described above (users who see the ad, never click, and later search your name) will credit Organic Search or Direct, never ChatGPT; that is a known blind spot to interpret around, not a configuration error. And GA4's default last-click lens will under-credit ChatGPT when a prospect clicks your ad on Tuesday and returns via Google search on Friday; check the Attribution paths report monthly so early-funnel assists from the channel are visible before you judge it.
Setting expectations: what the first 90 days of data actually look like
Set these expectations with whoever reviews the numbers, before launch. Volume will be low: this is one ad slot on a fraction of eligible conversations in your geography, so expect hundreds of impressions per day in a metro market, not tens of thousands, and single-digit daily clicks at pilot budgets. Signal accumulates slowly: with a handful of clicks per day, an ad group needs weeks to produce a readable conversion rate, which is why the pilot runs 90 days and not 14. Some return will be invisible: branded search lift and assisted conversions will not appear in any ChatGPT column. And there is no legal benchmark to compare against yet, so your baseline is your own Google Ads cost per consultation, not an industry number someone invented. A channel review at day 30 that says "only 41 clicks and 3 consultations" is not a failing channel; at typical pilot spend it may already be beating your search economics. Judge it on cost per consultation trending, then cost per signed case at day 90.
Layer 3 has one legal-specific trap: the phone. A large share of legal conversions are calls, not forms, and a call from a ChatGPT-sourced visitor is invisible unless you run call tracking with dynamic number insertion, which shows a unique tracking number to visitors arriving with your ChatGPT UTMs and logs the call against that source. Configure your conversion events in Ads Manager (and your analytics goals) around the actions that actually precede a case: form submission, tracked call over 60 seconds, live chat started, and consultation booked. Where OpenAI's conversion measurement supports server-side postbacks, send the consultation-booked event, not just the page view, so any conversion-optimized bidding learns from real intent.
Layers 4 and 5 are where the channel gets judged. Your intake system should stamp every new contact with its source automatically from the UTM, and intake staff should confirm it verbally ("how did you hear about us?") because self-reported answers catch what tags miss. Then, monthly, run the reconciliation: spend from Ads Manager, consultations and signed cases from the CRM by source, fees from the signed cases. The output is one row in the same scorecard every channel lives in:
| Metric | Where it lives | What it tells you | Decision it drives |
|---|---|---|---|
| Impressions and CTR | Ads Manager | Whether your context hints are matching the right conversations and your creative earns the click | Rewrite hints and creative; a CTR far below your other ad groups means a matching miss |
| CPC and spend by ad group | Ads Manager | What attention costs per intent theme | Shift budget toward ad groups with cheap, converting clicks |
| Landing page conversion rate | Analytics | Whether the page completes the promise the ad made | Fix the page before blaming the traffic; ChatGPT clicks are high intent |
| Cost per consultation | Analytics + CRM | The channel's real lead economics | Compare against your Google Ads baseline at the same stage |
| Consult-to-signed rate by source | CRM | Whether ChatGPT-sourced prospects are qualified or curious | If they consult but do not sign, the issue is targeting breadth or intake handling |
| Cost per signed case and fees per case | CRM + finance | The only verdict that matters | Scale, hold, or exit the channel; nothing else settles the question |
Two measurement cautions specific to this channel. First, expect view-through effects: many users will see your ad, not click, and later search your firm's name. Watch branded search impressions in Google Search Console and your direct traffic line after launch; a rise alongside ChatGPT spend is part of the channel's return even though no report will hand it to you labeled. Second, respect minimum data before deciding anything: a new ad group needs a few hundred clicks or several weeks, whichever comes first, before its conversion rate means something. Killing an ad group on twelve clicks is not discipline, it is noise worship, the exact reporting mistake we cover in the marketing measurement chapter of the KPI guide: three numbers per channel, judged with enough volume to trust.
Review rhythm: weekly for platform and landing page metrics while the pilot runs, monthly for the CRM reconciliation and cost per signed case, quarterly for the scale-hold-exit decision against every other channel in the mix.
Review process. Campaigns are submitted for review before serving, and legal services advertisers go through the stricter regulated-category path. Ads can be rejected or paused for policy, verification, or billing reasons, with the reason shown in Ads Manager. Build a week of review slack into any launch plan.
Targeting: How You Actually Reach the Right Person
This is the question every managing partner asks first: can I target by question, by keyword, by location? The honest answer is that ChatGPT targeting works differently from every platform you have used, and understanding the difference is the entire skill. There are three targeting mechanisms, plus system-level signals you influence but do not control.
Mechanism 1: Context hints (targeting by conversation, not keyword)
There is no keyword bidding in ChatGPT Ads. Instead, each ad group carries context hints: natural language descriptions of what you offer, who it helps, and when it is useful. The system reads the meaning of the live conversation and matches it against the meaning of your hints. OpenAI's own documentation is explicit that hints are not exact-match controls and do not guarantee your ad appears for particular words; people do not need to use your phrasing at all.
The practical craft, from OpenAI's guidance plus how semantic matching behaves: write hints as situations, not terms. A weak hint is a keyword transplant: "car accident lawyer." A strong hint describes the conversation: "a person who was injured in a car or truck accident, is dealing with the other driver's insurance company, received a settlement offer they are unsure about, and is deciding whether they need a lawyer before accepting." Cover the natural variations of the same need in separate phrases: fault disputes, medical bills after a crash, what a claim is worth, deadlines to file. Each ad group should hold one intent theme; meaningfully different situations (car accidents versus slip and fall) belong in separate ad groups with their own creative and landing pages.
On Google you target the words people type. On ChatGPT you describe the situations people are in, and the platform's model decides whether a live conversation matches your description. That means your targeting quality is a writing exercise: the firm that describes its ideal client's situations most precisely, in the most natural variations, wins the matching. Your intake team's knowledge of how injured people actually talk is now a targeting asset.
Recreated from OpenAI Ads Manager Beta, September 2026. Note that the interface itself confirms the model: hints guide matching, they are not keyword rules.
Mechanism 2: Geographic targeting (country, state, metro, ZIP)
Campaigns target by country everywhere the platform operates, and within the United States by state, metro area (DMA), and ZIP code. For law firms this is the load-bearing wall of the whole structure, for three reasons: the ad policy requires licensure in the jurisdiction where the ad shows, so state targeting is how you stay compliant; case value and venue quality vary by county, so ZIP and DMA targeting lets you concentrate spend where your best cases originate; and geography is how a $3M single-market firm avoids paying for impressions a statewide competitor would waste. A Denver firm can run Denver DMA plus the ZIP codes along its key corridors and nothing else.
Mechanism 3: Custom audiences (your lists, in or out)
Ads Manager supports custom audiences built from uploaded customer lists, used two ways: inclusion, to re-reach people already in your funnel (unconverted consultations, aged leads, newsletter subscribers), and exclusion, to stop paying for impressions to current clients and already-signed cases. The exclusion use is the one most firms skip and the one with immediate payback. Audience matching runs on hashed identifiers, the same mechanic as Meta and Google customer match.
What the system decides for you
Beyond your three levers, OpenAI's matching also draws on signals you do not directly control: the content of the current conversation, the user's location, and, where the user has personalization enabled, their past conversations and past interactions with ads. Users identified as under 18 never see ads, and sensitive conversations are excluded entirely. What does not exist, by design: demographic targeting, interest categories, lookalike audiences, individual user profiles you can buy against, and any placement that alters the organic answer. If an agency pitches you "demographic targeting on ChatGPT," they are describing a platform that does not exist.
| You want to target by | Does it exist? | How you actually do it |
|---|---|---|
| Questions people ask | Yes, indirectly | Context hints describing the situations behind those questions; the model matches meaning, not phrasing |
| Keywords | No exact match | Fold important terms into natural-language hints; they inform matching without guaranteeing it |
| Geography | Yes, precisely (US) | State, DMA, and ZIP targeting at the campaign level; mandatory discipline for legal because of licensing |
| Your own lists | Yes | Custom audiences for inclusion and exclusion via hashed list upload |
| Demographics, interests, lookalikes | No | Does not exist on this platform; relevance comes from the conversation itself |
| Time and device | Partially | Flight dates with start and end times, elastic budget pacing (up to 2x daily, 7x weekly), and an eligible-platforms selector at campaign level; no dayparting in the beta |
The Personal Injury Campaign Build, Step by Step
Here is what a complete build looks like for a hypothetical personal injury firm licensed in Colorado, serving the Denver metro. Adapt the specifics; keep the structure.
Step 1: Account and verification
Create the Ads Manager account on a work email, complete business verification, and prepare licensure documentation for the regulated-category review. Set up billing, add your logo, and invite whoever manages your marketing. Budget a week or two for legal-category approval before you plan any launch date.
Step 2: Campaign structure
| Level | Setup | Reasoning |
|---|---|---|
| Campaign | "PI - Colorado - Evaluation" · Clicks objective · $50/day pilot budget · Geo: Denver DMA plus 15 target ZIP codes | Click buying keeps the pilot measurable; geography matches licensure and best venues |
| Ad group 1: Car accidents | Hints describing crash victims dealing with insurers, settlement offers, fault disputes, medical bills after a collision | Highest volume intent theme, its own creative and landing page |
| Ad group 2: Serious injury | Hints describing hospitalization, surgery, long recovery, lost work, life-changing injuries from any accident type | Severity language separates high-value cases and deserves its own message |
| Ad group 3: Insurance disputes | Hints describing lowball offers, denied claims, adjuster pressure, "should I accept this settlement" | The single most advertiser-relevant conversation in PI: money is on the table and the decision is now |
| Ad group 4: Wrongful death | Hints describing a family that lost someone in an accident and is asking about accountability and claims | Handled separately so tone can be appropriately restrained; expect many conversations here to be excluded as sensitive |
Step 3: Creative that completes the answer
Remember the placement: your card appears under a thorough, neutral answer about the user's situation. The winning message offers the next step the answer cannot provide. Two examples of the pattern (write your own in your firm's voice and within your bar's rules):
- Title: "Get Your Settlement Offer Reviewed Free" · Copy: "A Colorado injury attorney will tell you what your claim is actually worth before you accept. No fee unless you win."
- Title: "Talk to a Denver Injury Lawyer Today" · Copy: "Free case evaluation. We handle the insurance company so you can focus on recovery."
Recreated from OpenAI Ads Manager Beta, September 2026. The hard limits are real: 50 characters of title, 100 of description. Every word has to earn its place.
Run two to three ads per ad group so the system can learn. Match the landing page to the ad group's intent, not to your homepage: the settlement-review ad lands on a settlement-review page with its own form and phone number, built on the principles in law firm conversion optimization and the lessons from our PPC landing page work.
Step 4: Intake readiness before launch
A ChatGPT click is a person mid-decision. If the click leads to a form nobody answers for a day, the spend is wasted. Before turning the campaign on, confirm the standards from our intake benchmarks guide: live response inside five minutes, after-hours coverage (this audience chats at midnight; pair the campaign with AI intake), and lead-source tracking that tags every contact as chatgpt-ads via UTM so cost per signed case is computable from week one.
Step 5: The feedback loop
Weekly for the first month: check which ad groups win impressions and clicks, read what converts on the landing pages, and refine hints the way you refine intake scripts, adding the situations your intake team actually hears on the phone from ChatGPT-sourced callers. Monthly: reconcile signed cases from the CRM against spend. Kill ad groups with clicks but no consultations; scale the ones producing signed cases; expand geography only after the core structure converts.
Strategy by Practice Area
The build above translates across practice areas; what changes is the conversation you describe in your hints and the urgency curve of the click.
- Criminal defense: the highest urgency fit on the platform. Arrest and charge conversations cluster at night and on weekends, exactly when ChatGPT is the only "person" available to ask. Hints should describe post-arrest situations: bail questions, first court dates, DUI stops, license consequences. Creative promises immediacy ("Talk to a defense lawyer today, 24/7"), and the campaign is only worth running if someone actually answers at 2 AM. Conversion economics follow the patterns in our practice-area benchmarks: criminal leads convert at the highest rates of any practice area when response is fast.
- Family law: long, research-heavy decision cycles. People ask ChatGPT about divorce process, custody standards, and support math for weeks before contacting anyone. Lead with a guide or consultation framed around clarity rather than commitment, and expect the sensitive-context filter to suppress some emotionally raw conversations.
- Immigration: explicitly named as permitted in the US policy. High ChatGPT usage among immigrant communities researching processes in their own words makes context-hint targeting unusually strong, but the licensing-jurisdiction rule and state targeting still govern where ads show.
- Estate planning: the calmest fit. "Do I need a trust or a will" is a classic ChatGPT conversation, ad-eligible, low emotional charge, and well matched to an educational lead magnet with a consultation follow-up.
In every practice area the same rule holds: the ad group map is your intake team's call log translated into situations. If you want the exact conversation themes to describe, ask the people who answer your phones what confused, worried people actually say.
Compliance: Bar Rules Still Apply
OpenAI's approval is the second gate, not the only one. Every state bar's advertising rules apply to ChatGPT ads exactly as they apply to billboards and Google Ads, and a new channel is precisely where regulators look first. The non-negotiables, drawn from the ABA Model Rules 7.1 through 7.3 pattern most states follow:
- No false or misleading claims. No outcome guarantees, no "we win 95 percent of cases" without substantiation, no implying the ad's appearance in ChatGPT is an endorsement by the platform.
- Required disclaimers travel with the ad. If your state requires "attorney advertising" labels, responsibility disclosures, or office-location statements, they must survive in a format with a short title and two sentences of copy. Some firms will need the landing page to carry what the card cannot.
- Jurisdiction honesty. Geographic targeting must match licensure, which OpenAI enforces, but your ad copy must also avoid implying you practice where you do not, which your bar enforces.
- Solicitation boundaries. A contextual ad shown to someone discussing an accident is advertising, not direct solicitation, under the prevailing understanding of the rules. But this is a new kind of placement mechanism, and state bars have not yet issued guidance specific to conversational AI ads. That absence of guidance is a reason for documented conservatism, not adventurousness.
- Turn off automatic text customization until your bar analysis says otherwise. Ads Manager includes a setting that lets AI generate personalized and translated variants of your headlines and descriptions, shown without your individual review. For most industries that is a convenience; for lawyers it means ad copy running under your name that no attorney approved, which collides with the ad-review obligations most states impose. Disable it, or get written guidance from your bar before enabling it.
- Keep records. Save every ad version, its running dates, its geographic scope, and its approval trail. If a bar inquiry ever comes, the firm with documentation has a conversation; the firm without has an issue.
A note on the sensitive-context filter working in your favor. The scenarios most likely to trigger bar concern about AI-era solicitation (reaching people in acute distress) are the same scenarios OpenAI's placement rules already exclude from ads. The platform's conservatism is not an obstacle to a compliant strategy; it is part of one.
Paid and Organic: Ads Are Half the AI Strategy
Because OpenAI keeps ads strictly separate from answers, there are two independent ways your firm can appear in the same conversation: as the sponsored card below the answer, and inside the answer itself when ChatGPT organically recommends firms. Money buys the first. Only authority buys the second.
The organic half runs on the visibility mechanics we documented in the AEO and SEO guide: AI assistants ground their local recommendations in a small set of trusted sources, dominated by a handful of legal directories, review profiles, and authoritative content. Analyses of legal AI visibility find that when a consumer asks an assistant to recommend a lawyer, the answer draws on roughly seven ranking directories, which means directory presence, review velocity (the system behind our review strategy guide), and structured, citable content decide whether you are the name in the answer.
The strategic point is the compounding effect between the two. A user who sees your firm recommended in the answer and then sees your ad beneath it gets the strongest possible one-two impression on the platform. A user who sees only your ad, on a conversation where a competitor is the organic recommendation, is watching you fight uphill. Run the ads for immediacy; build the authority so the answer itself starts agreeing with them. Firms that treat the ad channel as a substitute for AI visibility work will pay rising CPMs to compensate for an answer that never mentions them.
When to Start, the Early-Mover Advantage, and a 90-Day Pilot Plan
Be honest about what does not exist yet: there are no reliable legal-vertical benchmarks for ChatGPT Ads. The channel is months old, legal advertisers were only admitted in August 2026, and anyone quoting you a "typical cost per case on ChatGPT" is inventing it. What is known: launch pricing ran around a $60 CPM in February, click-based buying now exists, self-serve minimums are small, and inventory remains underfilled with ads on roughly a quarter of eligible queries. Early-channel dynamics favor pilots: low competition, cheap learning, and unusually high-intent clicks.
When should a firm consider ChatGPT ads?
Not every firm should touch this channel yet, and the honest filter has little to do with the platform. Run the pilot when these five things are true:
- Your core channels are not starving. ChatGPT ads come out of the testing budget, not the budget that feeds Google Ads and local search visibility. If cutting 5 to 10 percent of spend to fund an experiment would break your lead flow, fix that economics first.
- Intake can answer in minutes, around the clock. The channel produces conversational, after-hours prospects. A firm that returns web leads the next morning will convert almost none of them and will wrongly conclude the channel failed.
- Tracking exists end to end. If you cannot currently name your cost per signed case from Google Ads, you will not be able to judge ChatGPT either. The five-layer stack above is a prerequisite, not an upgrade.
- Your landing pages already convert. The pilot borrows its landing pages from your existing paid program. Pages converting under 5 percent will waste high-intent clicks from any source.
- You are licensed where the money is. The jurisdiction rule means the channel is only as big as your licensure map. A firm licensed in one state with strong case economics has plenty to work with; make sure the state's conversation volume justifies the setup effort.
Wait if intake is the current constraint (the KPI work in our benchmarks guide pays back faster than any new channel), if the marketing budget is under roughly $10,000 per month (the pilot would crowd out proven spend), or if nobody on the team can own the weekly review rhythm, because an unwatched experimental channel is just a donation to OpenAI.
The early-mover math: why starting now beats starting in 2028
Every major ad platform has followed the same price curve, and it only ever moves one direction. Google search clicks for personal injury cost under a dollar in 2003 and run $300 to $900 in competitive metros today. Facebook CPMs roughly quadrupled between 2015 and 2021 as advertisers piled in. The pattern is not mysterious: auction prices are set by competitor density, and competitor density only grows. ChatGPT Ads is currently at the emptiest point it will ever be, with ads filling about a quarter of eligible US conversations, retail dominating the early inventory, and the legal category open for less than 60 days as of this writing.
Starting early buys four compounding advantages, and only one of them is cheap clicks:
- Cheap attention, briefly. A quiet auction means the relevance-weighted system rewards good context hints over big bids. When the PI firms that bid Google clicks to $500 arrive, and they will, relevance alone will no longer carry a small budget.
- A learning curve competitors have to start from zero. Context-hint writing is a craft with a feedback loop. A firm with twelve months of tested hints, creative, and conversion data enters the crowded era knowing exactly which conversations produce signed cases; the late arriver pays 2028 prices for its 2026 education.
- Approval and account history. Legal is a manually reviewed category. Early advertisers clear verification, accumulate account standing, and learn the review process while it is a queue of dozens, not thousands. Platform history has quietly advantaged incumbents on every ad network that ever existed.
- Being first in the mental file. For the next year or two, a law firm ad inside ChatGPT is novel enough to be remembered. The firm a prospect first sees attached to helpful AI answers occupies ground that later entrants must pay to contest, the same first-association effect that made early Google advertisers disproportionate winners in their markets.
The counterweight, stated plainly: early also means volatile. The policy has flipped twice in seven months, formats are in beta, and volumes are modest. That is exactly why the right early posture is the small instrumented pilot below rather than a big commitment: cap the downside at a few thousand dollars, and let the upside be a two-year head start on a channel your competitors will eventually be forced to buy at full price.
A disciplined 90-day pilot for a firm spending $15,000 or more per month on marketing overall, sized inside the allocation logic from how much a law firm should spend on marketing:
| Phase | Spend | Goal | Decision at the end |
|---|---|---|---|
| Days 1-30 | $1,500 ($50/day) | Account approved, one campaign live, three or four ad groups, measurement wired end to end | Are we winning impressions and clicks on the right conversations? Refine hints and creative |
| Days 31-60 | $2,250 ($75/day) | Consultations booked from the channel; intake tagging every ChatGPT-sourced contact | Cost per consultation versus your Google Ads baseline; kill or feed ad groups accordingly |
| Days 61-90 | $3,000 ($100/day) | First signed cases attributable to the channel | Cost per signed case versus every other channel in your mix; scale, hold, or exit with data |
Total pilot exposure: under $7,000, or roughly the fee value of a single modest PI case, to build competence on a channel your competitors have not touched. The pilot fails informatively too: if 90 days produces clicks but no consultations, you have learned the channel's intent quality for your market before anyone overspent, and the fix is usually the landing page or intake speed, both of which improve every other channel when repaired.
The rational posture for a growth-minded firm is a small, instrumented pilot now, not because ChatGPT ads are proven for legal (they are not yet) but because the cost of learning is at its lifetime low while the policy window is open and the auction is quiet. Channels reward early competence; they never again sell it this cheap.
Where the Platform Is Heading
Based on OpenAI's announcements and credible industry reporting, the near-term roadmap includes video ad formats, ad placements inside third-party custom GPTs with revenue sharing for creators, conversation-capable ad experiences, and pay-per-action pricing tied to completed outcomes rather than clicks. OpenAI is also under visible pressure: reporting around the $1 billion milestone noted the company sits well short of its reported $2.5 billion 2026 advertising target, which historically predicts one thing on any platform: more inventory, more formats, and more advertiser tooling, faster.
For law firms, two projections worth planning around, clearly labeled as my synthesis rather than announced fact. First, the legal category will get more crowded and more expensive: the same PI firms that bid Google clicks to $500 will arrive here, and the quiet auction of late 2026 will not survive 2027. Second, expect agentic behavior to reshape what an ad click means: as ChatGPT gains the ability to book appointments and complete actions inside the conversation, the winning law firm ad eventually books the consultation directly in the chat. The firms whose scheduling, intake, and follow-up systems are already automated will plug into that world in a week; firms running on sticky notes will not. That readiness is the same operational foundation covered in how to scale a law firm, and it is the real preparation for every AI channel, this one included.
The wider market context for AI's impact on legal marketing spend sits in our law firm marketing trends report and the 2026 growth intelligence report.
Want a Clear Read on Whether This Channel Fits Your Firm?
Get an honest assessment of your market, your intake readiness, and where ChatGPT ads belong in your channel mix, before you spend a dollar on it.
Schedule a Free Consultation →Frequently Asked Questions
Can law firms advertise on ChatGPT?
Yes, in the United States. OpenAI's ad policy (updated August 2026) permits ads for legal advice, representation, and legal services when the advertiser is licensed to practice in the jurisdiction where the ad is shown, explicitly including personal injury, immigration, legal claims, and document preparation. Legal services is a regulated category with manual advertiser approval, and legal ads remain prohibited outside the US. Earlier guidance saying law firm ads are banned reflects the pre-August policy.
How does targeting work on ChatGPT ads? Can I target keywords?
There is no keyword bidding. Targeting runs on three mechanisms: context hints, which are natural language descriptions of the conversations and situations where your ad is relevant, matched by AI against conversation meaning rather than exact words; geographic targeting; and custom audience lists for inclusion or exclusion. Demographic, interest, and lookalike targeting do not exist on the platform.
Can ChatGPT ads target by geographic location?
Yes. Campaigns target by country, and within the United States by state, metro area (DMA), and ZIP code. For law firms geographic targeting is also a compliance requirement, because OpenAI only permits legal services ads in jurisdictions where the advertiser is licensed to practice.
How much do ChatGPT ads cost?
The US launch priced impressions around a $60 CPM, and the platform now supports click-based buying with a relevance-weighted auction, plus small self-serve daily minimums. There are no reliable legal-vertical benchmarks yet; the channel is months old and legal advertisers were only admitted in August 2026. A disciplined pilot can test the channel for under $7,000 over 90 days, measured on cost per signed case.
Do ChatGPT ads change what ChatGPT recommends in its answers?
No. OpenAI's stated principle is answer independence: the organic response is generated first and is never influenced by advertising. Ads appear as a clearly labeled sponsored card below the answer. Appearing inside the answer itself is a separate, unpaid outcome driven by directories, reviews, and authoritative content, which is why paid ads and AI visibility work are complementary strategies.
How are ChatGPT ads different from Google Ads for law firms?
Three differences matter most. Intent arrives as a full situation described in conversation rather than a short query, so targeting is written as situation descriptions instead of keywords. Only one sponsored unit shows per answer, so relevance beats bid more than on Google. And the audience is earlier in its decision, often days before it would search for a lawyer, which makes ChatGPT ads better at shaping the shortlist and Google better at capturing the final click.
Does ChatGPT have ads now?
Yes. OpenAI began testing ads in ChatGPT in the United States on February 9, 2026, on the Free and Go tiers, and has since expanded to more than 40 markets. Ads appear as a clearly labeled sponsored card below the answer. Plus, Pro, Business, and Enterprise subscribers do not see ads. The business reached a $1 billion annualized revenue run rate within 200 days of launch.
What do ChatGPT ads look like?
A single sponsored card below the answer, visually separated and labeled, containing the advertiser's name and favicon, a short title, one to two sentences of copy, an image, and a landing page link. An e-commerce carousel format and call-to-action buttons are in testing. The recreated illustration in chapter 4 of this guide shows the exact placement.