Legal Ads Are Off the Table in ChatGPT. Getting Cited Is the Route Left.

By Frank Alfano, LL.B., LL.M.

ChatGPT now shows ads on roughly one in four commercial prompts. If you practise law in Ontario, you are not allowed to buy one.

That door is closed by policy, not by budget. There is no amount of money that opens it.

I spent 33 years in Ontario courtrooms before I started writing production code for a living, and the thing that strikes me about this story is how familiar it feels. Nothing gets into the record because you meant it to. It gets in because somebody entered it properly. The rules of admissibility are completely indifferent to how badly you want the exhibit in.

AI answers work the same way now. Here is what the numbers actually say, and what is left for you.

What the study found

SE Ranking analysed 50,006 commercially oriented prompts across 20 niches in the United States, collected on July 23, 2026. ChatGPT served a sponsored placement on 25.94% of them. Search Engine Land covered the findings on August 10.

One caveat, stated plainly: SE Ranking sells SEO software. A company in the business of selling SEO tools publishing research that makes paid placement look weak is not a disinterested witness. Weigh it the way you would weigh any expert report where the expert has a commercial interest in the conclusion. The figures hold up across the independent coverage, which is why I am willing to build on them, but I am not going to pretend the source has no stake.

Every ad in the sample sat below the generated answer and carried a single sponsored offer, with no competing advertiser alongside it. That is the entire placement. One slot, underneath.

The ad buys a slot beside the answer, not a place inside it

Here is the figure that should reframe the whole conversation.

In 96.37% of those placements, the advertiser was not cited anywhere in the answer itself.

The advertiser's domain appeared among the cited sources in 3.63% of placements. The exact advertised URL appeared in 0.09%, which works out to twelve cases in the entire sample.

Sit with what that means for how a person actually reads the screen. The answer is the part the reader trusts, because a machine wrote it and machines sound certain. The ad is the part underneath, labelled as paid, which every one of us has been trained by twenty years of the internet to skate past. You are renting a billboard beside a conversation you are not in.

Being cited is different in kind, not in degree. A citation is the machine saying your name while it answers. It arrives with the credibility of the answer attached to it, because the reader cannot easily tell where the machine's authority ends and yours begins. That is the whole prize, and it is not for sale.

Lawyers cannot advertise on ChatGPT at all

Now the part specific to your practice. OpenAI's advertising policy bars the category outright. Search Engine Journal reported on August 12 that the rules state ads for legal advice, court-related work and similar services are not allowed, with one narrow exception carved out for general legal education.

Read that exception closely, because it is carrying a great deal of weight. "General legal education" is an article explaining how a Small Claims Court hearing runs, or what a limitation period is. It is not "book a free consultation with our personal injury team." The first is teaching. The second is soliciting a retainer, and that is the thing the policy excludes.

Search Engine Journal also flagged a discrepancy worth knowing about before anyone budgets against it. OpenAI's public ad policies page reads as though approval might be theoretically available for legal services, while the detailed rules say the category cannot be advertised. If you are planning AI ad spend on the strength of the friendlier page, you are planning against the wrong document.

So the HVAC company two doors down can pay to appear underneath an answer about failing furnaces. You cannot pay to appear underneath an answer about wrongful dismissal. Same platform, same city, different rulebook.

Why this lands harder on regulated professions

Every profession is watching AI answers eat the top of its funnel. Most of them still have a paid escape hatch. If a plumber's organic visibility collapses, the plumber can buy their way back into view tomorrow morning with a credit card. It costs money, and the economics may be miserable, but the option exists.

You do not have that option. Because lawyers cannot advertise on ChatGPT at any price, the paid lane is closed at the policy level, which means earned visibility is not one channel among several for you. It is the channel. Everything else you might have used to paper over a weak web presence has been taken off the table for you specifically.

That cuts both ways, and the second way is genuinely good news. In a market where nobody in your category can buy the top of the answer, the competitor with the deeper marketing budget cannot simply outspend you into invisibility. The advantage goes to whoever built the more citable practice. That is a fairer fight than most professions are getting right now, and it rewards exactly the kind of firm that has been doing the unglamorous work for years.

It is also why what regulated professionals need from a web developer is a different specification than what a restaurant needs. The compliance constraints are tighter, the trust bar is higher, and now the paid fallback is gone.

The unglamorous list Google actually publishes

There is a healthy industry right now selling schema packages and "AI optimisation files" to professionals who are nervous about this exact problem. Before anyone sells you one, read what Google publishes for free.

Google's documentation on AI features, last updated December 10, 2025, is blunter than most of the advice being sold about it. It says there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary. In its own words, you do not need to create new machine readable files, AI text files, or markup, and there is no special schema.org structured data that you need to add.

Nobody can sell you the shortcut, because the shortcut does not exist. What Google asks for instead is duller:

• Let the crawlers in. Check robots.txt, and check that your CDN or host is not quietly blocking them.
• Keep the content that matters in readable text, not baked into an image and not locked behind a script.
• Make your structured data match what a human actually sees on the page.
• Build internal links so your pages are discoverable from one another.
• Keep your Business Profile information current.

Google also says the quiet part in the same document: indexing and serving are not guaranteed. All of this makes you eligible. None of it makes you entitled. That distinction is the entire difference between doing the work and buying a promise, and it is the same distinction that separates what it actually takes to rank in 2026 from what gets pitched to you in a cold email.

What I would add for a law practice

The Google list is the floor. It gets you eligible. Three things move a professional practice from eligible to actually cited.

Put a named human on the page. Not "our team" and not "admin." A name, real credentials, and a bio that establishes why this person is qualified to answer this question. Answer engines are built to prefer identifiable expertise, and so are the people reading them. A page about limitation periods written by a named lawyer with a real call to the bar is a different object than the same words with no author.

Write pages shaped like the question, not like the practice area. Your site probably has a page called "Employment Law." Nobody types that at eleven at night. They type "can my employer fire me for refusing to work overtime." One of those is a filing cabinet label and the other is a question a machine can answer by quoting you. This is the whole discipline behind answer engine optimisation, and it is not complicated. It is just unfamiliar.

Be quotable on purpose. An answer engine lifts a sentence or two. Give it clean ones. Lead with the answer, then explain it. A paragraph that opens with "It depends on several factors" is unquotable by construction. A paragraph that opens with "In Ontario, the general limitation period is two years from the day the claim was discovered" is a sentence a machine can lift intact and attribute to you. The mechanics of that are covered properly in how to get cited by ChatGPT.

None of this is separable from the rest of your visibility. Getting cited and getting found are the same body of work seen from two angles, which is what the intersection of SEO, AEO and GEO is about.

The sixty second test

Do this before you close the tab. Open ChatGPT and ask it who handles your practice area in your city. Read the answer twice.

First, look for your name. Then look at what it cited.

If the sources it named are directory profiles you have never logged into, a legal aggregator you do not control, or a competitor's blog post, then somebody else is currently writing your record. That is not a hypothetical risk. That is the answer a potential client got this morning.

The uncomfortable part is that being right is not the remedy. You can be the most competent practitioner in the region and be entirely absent from that answer, for reasons that have nothing to do with your competence and everything to do with whether a crawler could read your site. Being excellent and being invisible are perfectly compatible states, which is also the argument behind why ranking first is not the same as winning the client.

Nothing gets into the record because you meant it to. Somebody has to enter it properly.

If you want your site built so the machines can actually read it, that is the work we do. Start with a free website audit, or read about how we approach SEO and visibility.

Common questions

Can lawyers advertise on ChatGPT?

No. OpenAI's advertising policy states that ads for legal advice, court-related work and similar services are not allowed, with a narrow exception for general legal education. The restriction is OpenAI's own platform policy, not a law society rule. It applies regardless of budget or jurisdiction.

Does a ChatGPT ad get my firm cited in the answer?

Almost never. SE Ranking's study of 50,006 commercial prompts found that in 96.37% of sponsored placements, the advertiser was not cited anywhere in the answer copy. The advertiser's domain appeared among cited sources in 3.63% of cases. Paid placement and earned citation are separate systems.

What counts as general legal education?

Content that teaches rather than solicits. An explainer on how a hearing works or what a limitation period means is education. An offer of a consultation, a case evaluation, or representation is legal services, and that is the part the policy excludes. The line sits at whether you are informing the reader or seeking a retainer.

Is there special schema that gets me into AI Overviews?

No. Google's AI features documentation, updated December 10, 2025, states there is no special schema.org structured data required and no need for AI-specific files or markup. What it asks for is standard practice: crawlable pages, content in readable text, structured data that matches the visible page, internal linking, and a current Business Profile.

How do I check whether AI tools mention my firm?

Ask ChatGPT who handles your practice area in your city, then examine the citations under the answer rather than just the text. The sources it names are the pages currently shaping your reputation in AI answers. If they are directories or competitors rather than your own site, that is the gap to close first.

---

Sources: SE Ranking, ChatGPT ads study (August 10, 2026); Search Engine Land (August 10, 2026); Search Engine Journal on OpenAI's ad policy (August 12, 2026); Google Search Central, AI features documentation (updated December 10, 2025).