Blog · By industry

AEO for legal services

Published 4 May 2026 · Updated 12 October 2026 · 9 min read

AEO for legal services
Photograph: Dmitrij Paskevic / unsplash

The short answer: AEO for legal services is the work of making a firm retrievable, correctly identified, and safely quotable when an assistant answers a legal question. Three things make it different from other sectors: the answer often omits the jurisdiction, directories and bar listings turn up as the cited source more often than firm sites in the runs we have looked at, and professional conduct rules govern the claims an engine may lift from your page. Fix access first, then identity, then the passages themselves.

Legal buying starts with an informational question. Can my landlord keep the deposit. How long do I have to file. What happens if the other side ignores a notice. Those questions used to send a person to a firm's explainer page. Now a large share of them end in a written answer, and the firm is either named in it or absent from it.

Zero-click pressure, and what the measured numbers do and do not cover

Pew Research Center tracked the browsing of 900 US adults across 68,879 Google searches in March

  1. On result pages carrying an AI summary, people clicked a traditional result on 8 percent of

    2. visits. Without a summary, 15 percent. Clicks on the links inside the summary itself occurred on 1 percent of visits to pages that carried one. A summary appeared on 12,593 of the 68,879 searches, roughly one in five.

Those are all-category numbers, not legal ones, and nobody should quote them as a legal benchmark. They still matter to a firm because the legal funnel is unusually top-heavy with exactly the kind of question a summary handles well: short, factual, procedural, answerable in four sentences. An explainer article that used to earn a click may now earn a mention instead, or nothing. We have not measured how often that substitution happens in legal queries specifically.

This is the point where a managing partner usually asks whether the explainer library was wasted money. It was not. It changed job. A page's value is now measured by whether an engine can lift a correct, attributable passage out of it, which is a different craft from writing for a reader who has already arrived. The split between work you control and work you do not is covered in AEO by industry: where the work differs.

Jurisdiction collapse, the failure mode specific to law

An assistant writes one answer. The user asked "how long do I have to sue for breach of contract", with no state named, and the model composes a single confident paragraph from whatever passages it retrieved. If your Karnataka limitation page and a Texas statute-of-limitations page both land in the candidate set, nothing in either passage tells the model they describe different law, and the answer can merge them.

Two risks follow. Your page can be retrieved for prompts it should never answer, and your firm name can end up attached to a number that is wrong for the reader asking. Neither would be the engine misbehaving. It is doing passage retrieval on text that never said which law it describes. We do not have a measured rate for how often this happens, which is why the check below is a check and not a statistic.

The fix is mechanical. Put the jurisdiction in the H1, in the first sentence, and in the sentence that carries the number, not only in a breadcrumb or a footer. "Under the Limitation Act, 1963, in India, the period is three years from the date the cause of action arises" survives extraction. "The period is three years" does not.

Here is a check worth running this week. Pick your five highest-traffic explainers. For each, run the question twice on two engines, once bare and once with a jurisdiction the page does not cover. If your page is cited in the second run, the page is under-specified, and you are one summary away from being quoted wrongly.

What the engines actually say about eligibility

Google's own documentation on AI features states that there are no additional requirements to appear in AI Overviews or AI Mode, and no special schema.org markup that produces eligibility. Eligibility comes down to two conditions: the page is in the index, and it is permitted to show a snippet. The same page names nosnippet, data-nosnippet, max-snippet and noindex as the controls that limit appearance in AI features.

Read that second sentence again if your firm's site sets a short max-snippet to stop scrapers. By Google's own account that same directive is one of the levers that limits what its AI surfaces may show. We have found inherited snippet limits on firm sites during audits, usually set years earlier for a reason nobody present could still name. We have not counted how common that is.

OpenAI's crawler documentation draws a similar line for its own agents. OAI-SearchBot is the agent behind ChatGPT's search features, and the docs state that sites opted out of it will not be shown in ChatGPT search answers, though the same page adds that they can still appear as navigational links. GPTBot is the training crawler and is a separate decision: the docs say the settings are independent, so a site can allow one and disallow the other. ChatGPT-User is the agent for actions taken on a live user request, and OpenAI states it is not used for automatic crawling and does not determine whether content appears in search. A firm can therefore refuse training and still be findable in ChatGPT search. None of this generalises to engines that publish no such documentation, and we have no way to see inside the ones that do not.

Check your own edge, not your robots.txt. A WAF or bot-management rule can return a challenge page to a search crawler while your robots file says allow, and the robots file will not tell you. Request your own URL with each agent string and compare the status codes and the byte counts. A 200 that returns 4 KB of JavaScript shell is a failure dressed as a success.

Identity: your name is a surname, and surnames collide

Firm names are built from family names, and family names repeat. An engine trying to decide whether the firm in your site, the firm in a bar council listing, the firm in a court order and the firm in a news article are one entity has very little to work with when all four strings differ slightly.

Schema.org has a LegalService type for a business providing legally oriented services, advice and representation. There is an Attorney type beneath it, but schema.org marks Attorney deprecated and points to LegalService as the more inclusive and less ambiguous choice, so use LegalService. Marking up the firm with a stable @id, the registered name, areaServed, and sameAs links to listings you genuinely control does not buy eligibility, as Google's documentation says outright. What it offers is disambiguation. The engines we test do not expose how they resolve entities internally, so treat that as the stated purpose of the markup rather than a measured result. The same reasoning applies in regulated finance, which is why AEO for fintech ends up with a near-identical identity checklist and a very different content one.

The conduct rules are part of the brief

The American Bar Association's Model Rule 7.1 bars a lawyer from making a false or misleading communication about the lawyer or the lawyer's services. The model rules are a template, not law: each regulator adopts and amends its own version, and we have not surveyed those adoptions, so read the rule that governs you rather than the model. Advertising and solicitation rules for advocates in India sit with the Bar Council of India and differ from the US position.

You cannot control the sentence an engine writes. You can control the raw material it has to work from. Two practical consequences. First, a case-results page written as a sequence of outcome figures gives an assistant everything it needs to compose a comparative superlative with your name on it. Adding the facts and the conditions to each entry makes the lift harder and the summary more accurate. Second, a disclaimer parked as the opening paragraph is a candidate for retrieval like any other passage, and a page whose first paragraph is a disclaimer has put a non-answer where the answer should be. Move it below the answer. We cannot see which passage an engine selected, so this is a reasoned precaution rather than something we have measured.

Honest limits

Several things this post does not establish. The Pew figures are all-category US web measurements from March 2025, not legal ones, and they describe Google result pages rather than assistants in general, so they are context and not a legal benchmark. The engine behaviours above are what Google's and OpenAI's own documentation says about their own surfaces on the dates cited; no other engine is covered, and documented behaviour can change without notice. Directory and bar-listing dominance shows up in every audit run we have looked at, but those runs are a small, unsystematic sample of firms rather than a census, and we have not put a figure on it. We also cannot see inside retrieval: where this post says a passage "gets retrieved" or an entity "gets resolved", that is a reasoned account of a mechanism, not an observation. And attribution stays weak, because these surfaces send little or no referrer, so the honest measurement is a tracked prompt set rather than a traffic line in analytics.

Who runs that prompt set is a live question in most firms, usually landing between marketing and the practice groups. Who owns AEO inside a company is worth reading before the first meeting. If you want to see what a finished assessment looks like before commissioning one, the sample audit report walks through the evidence format section by section.

FAQ

Is this just SEO with a new label? Indexing and clean markup serve both, so there is real overlap. The goal differs. A firm wants a liftable, jurisdiction-stamped passage and an entity an engine can resolve, not a position in a list. The terminology argument is unpacked in AEO vs GEO vs SEO: what the three terms actually mean.

Will structured data get my firm into AI Overviews? No. Google states there is no special schema that creates eligibility. Markup helps an engine work out who you are, which affects whether the right firm gets named once a passage is already chosen.

Should a firm block AI crawlers to protect its content? Blocking the training crawler and allowing the search crawler is a coherent position, and the documentation supports separating them. Blocking everything removes the firm from answers people are already reading.

What should a firm fix first? Access, every time. Crawler reachability, rendered HTML, and snippet directives are days of work and they gate everything else. Identity next. Third-party corroboration is slowest and least predictable, so start it early and judge it late.

How do I know whether any of it worked? Run a fixed prompt set, several runs per prompt, across more than one engine, and record three things each time: named or not, claim correct or not, sources used. Change one variable, wait, re-run. A single run is an anecdote.

Sources

Related reading

AEO by industry: where the work differsAEO for fintechAEO vs GEO vs SEO: what the three terms actually meanWho owns AEO inside a companyAEO for businesses with a service areaAEO for marketplaces and aggregators, when the inventory is not yours