Optimising for citations in AI answers, not just blue links

Search stopped being a list of links. A meaningful share of the people who need a bankruptcy attorney now get their first answer from an AI summary, and they act on it without ever scrolling to a result. Ranking and being found have quietly stopped meaning the same thing.

That changes what content is for. A page no longer has to win a position — it has to be the thing a language model reaches for when it assembles an answer, and it has to be quotable enough to survive being summarised.

• KeywordsSemrush and Ahrefs to map the questions people ask before they are ready to use the word “bankruptcy” — debt, garnishment, foreclosure, collection calls. The intent arrives long before the vocabulary does.
• SchemaJSON-LD structured data across the site so a machine can parse without guessing: who the firm is, which three states it is licensed in, what each page answers, and which answers are authoritative.
• CitationContent restructured so each page answers one question cleanly and self-containedly. AI systems quote passages, not pages — a paragraph that only makes sense with the four above it will not be picked up.
• LocalisationIllinois, Georgia and Texas have different exemptions and procedures. Content built per state rather than nationally, so the answer is specific enough to be worth citing.
• MobileThe same technical work that helps a crawler helps a phone: faster pages, cleaner markup, content reachable without interaction.

Impressions in AI surfaces went from roughly 33,000 in May to 129,081 in June. They did not stay there. July settled nearer 95,000 and the line has drifted gently down since. Across the period the pages accumulated about 264,000 AI-surface impressions in total.

I am showing the shape rather than the peak deliberately. A single month’s high is easy to quote and easy to puncture; what actually matters is that the floor moved — the quiet months now sit at roughly three times where this started. Click-through rate rose alongside it, and the traffic arriving was better qualified, which is the part the intake team feels.

• ~33KMonthly AI-surface impressions at the start
• 129,081Peak month — June 2026
• ~3×Where the sustained level now sits versus baseline
• CTRHigher click-through and better-qualified traffic

AI-surface reporting is young and inconsistent between tools. An impression inside a generated answer is not comparable to a classic search impression, and attribution from an AI citation through to a phone call is still weak — you can see that you were quoted, rarely that the quote produced the client.