How to write a practice-area page an AI engine will quote

What makes a practice-area page quotable by an AI engine?

A practice-area page is quotable when an engine can cut one paragraph out of it and the paragraph still answers the question correctly, names who said it, and carries its own qualifiers. Most law firm practice pages fail that test because they were written to persuade a visitor who reads top to bottom, not to supply a sentence to a system that reads one block at a time and discards the rest.

Three properties do most of the work. The first is self-containment: the answer does not depend on a heading, an earlier paragraph or a pronoun to make sense. The second is specificity: the passage states the jurisdiction, the deadline, the dollar threshold or the procedural step as a fact rather than gesturing at it. The third is attribution: the firm, the statute or the agency is named in the sentence itself, so an engine quoting it can say whose answer it is.

None of this replaces the record around the page. Firm AEO's earlier posts on directories, reviews and entity consistency cover the corroboration an engine checks before it names a firm. Extractability decides whether the page can be quoted once the engine has decided the firm is worth quoting.

Why do AI engines quote passages instead of whole pages?

AI engines retrieve and rank sections of documents because a single page usually answers many questions, and the answer to any one of them lives in a few sentences. Google made passage-level ranking public in 2020, years before AI Overviews, and its current guidance for AI features adds no new requirement beyond being indexable and eligible for a snippet.

Google's October 15, 2020 announcement, ‘How AI is powering a more helpful Google’, said that understanding individual passages, rather than only whole pages, would improve 7 percent of search queries across all languages as it rolled out. That is the earliest first-party statement that a buried paragraph can be retrieved on its own merits. Generative answers extend the same idea: the engine retrieves candidate passages, then writes from the ones that state the answer most cleanly.

Google Search Central's ‘AI features and your website’ page states that to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, and that there are no additional technical requirements. The same page lists nosnippet, data-nosnippet, max-snippet and noindex as the controls that limit what Google shows. A practice page with a restrictive max-snippet setting, or a data-nosnippet wrapper left on the main content by a theme or plugin, has opted its best passages out of being quoted. Checking the page source for those attributes takes two minutes and belongs at the start of any rewrite.

Ranking position is a weaker predictor of citation than it used to be. Ahrefs reported in March 2026 that only about 38% of pages cited in AI Overviews ranked in Google's top 10. The question of why a lower-ranked competitor gets quoted above a page in position one is its own Firm AEO post; the short version is that the cited page had the better passage.

Where on a page do ChatGPT's cited passages come from?

Mostly from the top. Kevin Indig's Growth Memo analysis, published in February 2026 and reported by Search Engine Land, examined 1.2 million ChatGPT responses and validated 18,012 citations against the source pages. 44.2% of the cited text came from the first 30% of the page, 31.1% from the middle 30% to 70%, and 24.7% from the final third, with a sharp drop near the footer.

Indig called the shape a ski ramp and attributed it to the bottom-line-up-front structure of the journalism and academic writing that language models learn from. The topic split matters for law firms: Indig reported that finance had the steepest ramp, while healthcare was flatter. Legal was not broken out, so the right reading for a practice page is directional rather than precise.

Two cautions before using the figure. The data set covers ChatGPT only, not Google AI Overviews, Perplexity, Claude or Gemini. And the analysis is an independent analyst's study published through a newsletter and the trade press, not a peer-reviewed paper with a released data set. Firm AEO treats it as the best available evidence on position, not as a law.

The practical consequence is still clear. A typical law firm practice page opens with a hero banner, a slogan, two paragraphs about the firm's values and a call-to-action block, and reaches the first factual answer somewhere past the halfway mark. On Indig's distribution, that arrangement spends the most-cited third of the page on the sentences least likely to be quoted.

What does a quotable sentence look like?

A quotable sentence defines or states something directly, names its entities, and sits under a heading phrased as the question it answers. Indig's February 2026 study measured each of those features in the 18,012 cited passages and found cited text was nearly twice as likely to use definitional phrasing, far denser in proper nouns than ordinary prose, and easier to read than text that was retrieved but not cited.

The individual findings, each reported by Search Engine Land from Indig's study, translate into drafting rules for a practice page:

• •Definitional phrasing. Cited passages were nearly twice as likely to use constructions such as ‘X is’ or ‘X refers to’. On a practice page: ‘A wrongful death claim in Texas is a lawsuit brought by…’ rather than ‘When tragedy strikes a family…’.
• •Questions as headings. Cited content was about twice as likely to contain a question mark, and 78.4% of citations tied to a question came from headings. The engine appears to treat the heading as the prompt and the paragraph under it as the answer, so the heading should be the client's question in the client's words.
• •Entity density. Cited text averaged 20.6% entity density, meaning proper nouns such as names, places, statutes and organisations, against the 5% to 8% Indig gives for typical English. Practice pages that name the county, the court, the statute and the firm in the answer sentence sit closer to the cited profile.
• •Readability. Cited passages averaged a Flesch-Kincaid grade level of about 16; passages that were not cited averaged 19.1. The lesson for lawyers is not to simplify the law but to cut stacked subordinate clauses. One idea per sentence, with the qualifier attached to the sentence it qualifies.
• •Tone. Indig reported cited passages clustering around 0.47 on a 0-to-1 subjectivity scale, between dry recitation and opinion. A sentence that states the rule and then says plainly what it usually means for the reader fits that middle.

Before and after: one Texas car accident section, rewritten

The example below is an illustrative composite of the opening Firm AEO sees most often on personal injury pages, rewritten to the drafting rules above. It is a demonstration of structure, not legal content for any firm to publish as written: every legal statement on a real page must be checked by the firm's own attorneys, and the firm's ethics counsel decides what the final copy may say.

Before. Heading: ‘Houston Car Accident Lawyers’. First paragraph: ‘When you've been hurt in a crash, you need an aggressive, experienced advocate in your corner. Our award-winning team has recovered millions for injured Texans and we fight for every client like family. Don't wait. Call now for your free consultation.’ The deadline to sue appears in the eleventh paragraph, as ‘strict time limits may apply to your case’.

After. Heading: ‘How long do I have to file a car accident lawsuit in Texas?’ First paragraph: ‘In Texas, most car accident injury lawsuits must be filed within two years of the crash, under section 16.003 of the Texas Civil Practice and Remedies Code. Claims against a city, county or state agency are different: the Texas Tort Claims Act requires written notice of the claim within six months, and some cities set a shorter period by charter. [Firm name], a Houston personal injury firm, reviews which deadline applies to a claim during an initial consultation.’

The second version answers the question in its first sentence, names the statute, states the number, carries its own exception, and names the firm in the third person. If an engine lifts only the first sentence, the extract is still true. If it lifts the whole paragraph, the reader learns the exception and who said it. The table below scores the two versions against the extraction tests.

Do statistics, sources and keywords change whether a page is quoted?

Statistics and named sources help; extra keywords do not. The Princeton generative engine optimisation study by Aggarwal and colleagues, presented at KDD 2024, tested rewriting methods across GEO-bench, a benchmark of 10,000 queries. Adding quotations and adding statistics produced relative visibility gains of 30% to 40% on the paper's position-adjusted word count measure, while keyword stuffing offered little to no improvement.

One line in the paper speaks directly to law firms. The authors reported that domains such as ‘Law & Government’ benefit significantly from adding relevant statistics to website content. For a practice page, the relevant statistic is usually a public one: the statutory deadline, the dollar threshold for small claims court in the state, the filing fee published by the court, or a figure from the state bar or court administration. Each should carry its source in the sentence, so the extract keeps its attribution.

The keyword finding is the one most practice pages need to hear. Pages written to an older SEO brief repeat ‘Houston car accident lawyer’ a dozen times and say almost nothing a client could act on. The Princeton results suggest that repetition does nothing for generative visibility, and the space it takes is space a quotable answer could occupy.

The paper's benchmark predates AI Mode and was not legal-specific, and its gains are relative improvements on the paper's own metrics, not a promise of how often any firm will be cited. Firm AEO uses it for direction: specific, sourced statements outperform vague or repetitive ones.

How do bar advertising rules change the rewrite?

They make the rewrite stricter, because an extracted sentence travels without the disclaimer at the bottom of the page. ABA Model Rule 7.1 prohibits a false or misleading communication about a lawyer's services, and a sentence that is accurate only alongside a footnote can become misleading once an engine quotes it alone. Every claim in a quotable passage has to be true standing by itself.

In practice that means three edits in the same pass as the structural rewrite. Outcome language such as ‘recovered millions’ either comes out or moves into a sentence that carries its own qualifier, so the smallest extractable unit is not misleading. Unverifiable superlatives such as ‘award-winning’ or ‘best’ come out unless the firm can name the award and the year in the same sentence and its ethics counsel approves the wording. And the firm is described in the third person by what it does, not by what it promises.

If AI tools help draft the rewrite, ABA Formal Opinion 512 (July 29, 2024) places the responsibility for the output on the lawyers who publish it, which in practice means attorney review of every legal statement before it goes live. State advertising rules differ on whether and how website copy must be filed, labelled or retained; Firm AEO's bar advertising rules pillar maps the main regimes, and the firm's ethics counsel decides which apply to a revised page.

How should a law firm audit its existing practice pages?

Start with the questions, not the pages. List the questions clients actually ask about each practice area, map each question to the page that should answer it, and score the first paragraph under each question against the extraction tests. Most firms find the answers exist somewhere on the site but sit in the wrong place, in the wrong form, or under the wrong heading.

The order Firm AEO uses, with timings labelled typical rather than promised:

• •1. Check eligibility first. Confirm each practice page is indexed and has no nosnippet, data-nosnippet or restrictive max-snippet setting over the main content. Typical time: an hour for a mid-sized site.
• •2. Collect the real questions. Pull them from intake notes, call recordings the firm already reviews, Google Search Console queries and the questions the engines suggest. Ten to twenty per practice area is typical.
• •3. Map one question to one passage. Each question gets a heading phrased as the question and one answer paragraph. Questions that need their own page, such as a specific charge or a specific county, get one.
• •4. Rewrite the first sentence under every heading. Direct answer, jurisdiction inside the sentence, the number stated, the source named.
• •5. Move answers up. Put the most-asked factual answers in the first third of the page and move firm biography and slogans below them.
• •6. Run the Rule 7.1 pass. Remove or qualify outcome language and unverifiable superlatives at sentence level, then send the page to the firm's attorneys for legal review.
• •7. Re-measure after re-crawl. Track whether the rewritten passages start appearing in AI answers for the mapped questions, using a fixed prompt set over several weeks rather than a single check.