A regional accounting firm came to us with a service page that had produced nine visits and zero leads in ninety days. Eight weeks after an answer-first rewrite, the same URL was cited by Perplexity on four of the twelve buyer questions we track, organic sessions were up 61%, and the page had produced three qualified inquiries. Nothing else on the site changed during the window.
This case note documents what we changed, in what order, what it cost, and the one intervention that did nothing at all.
Client anonymized as always. The numbers are theirs, rounded only where noted.
The page before: 2,400 words, zero answers
The "before" picture: a 2,400-word outsourced-bookkeeping page ranking at position 28 for its head term, nine organic visits in ninety days, an 84% bounce rate, and zero citations across the twelve tracked buyer questions in three AI engines. It was written like a brochure — a hero slogan, a "why choose us" essay, four paragraphs on company values, and the actual service description starting around word 1,700.
Nothing about it was unusual. Most service pages we audit look exactly like this, because they were written to be approved by the owner, not to answer a buyer.
Two more numbers complete the picture: the page had earned zero external links in four years, and its meta description was the first 155 characters of the hero slogan. Nobody had touched any of it since launch.
The firm's ask was a redesign. We talked them out of it — the template was fine. The words were the problem.
The diagnosis took 40 minutes
We ran the standard audit: ask the twelve buyer questions in ChatGPT, Perplexity, and Google AI Mode, record who gets cited, then score the page against what wins. The findings fit on one screen. Zero question-shaped headings. First passage that directly answers anything at word 1,700. No visible date. No named author. No FAQ. Generic h2s like "Our Approach" that no engine could match to any question a human asks.
The competitors being cited were not better firms. They were better-structured pages — two of them visibly thinner on actual expertise.
We also scored the twelve queries by intent value with the partners. Four were worth winning this quarter, so the rewrite optimized for those four first — and three of them are the ones now citing.
That gap is the entire opportunity. Expertise the engines cannot extract does not exist, as far as a citation is concerned.
The rewrite: six questions, six answers
We rebuilt the page around the six questions buyers actually ask, verbatim as h2s: what outsourced bookkeeping includes, what it costs monthly, how handoff works, what software it runs on, when to switch from DIY, and how it differs from a fractional CFO. Under each, the first paragraph is a self-contained 40-to-90-word answer written to survive being lifted out of context. Supporting detail follows the answer, never the reverse.
Writing order matters: we wrote the six answers first, as standalone passages, then built the page around them. Answers written last always lean on context that will not travel.
Each h2 keeps the buyer's vocabulary. Nobody asks an engine about engagement models. They ask what it costs each month, so that is the heading, nearly word for word.
Six sections, six answers, no section over 220 words. The page reads faster than the brochure did at half the length.
The price answer names real numbers — a monthly range and what moves it. That paragraph alone is now the page's most-cited passage.
What we cut: 1,100 words of reassurance
The rewrite removed 1,100 words: the values essay, three paragraphs of adjectives, two redundant CTAs, and every sentence whose only job was to say "we are trustworthy" without evidence. Final length: 1,300 words. Cutting was not cosmetic. Citation is a density game — the fraction of a page an engine can quote — and reassurance prose is unquotable filler that buries the passages that earn.
The owner pushed back on cutting the values section. We moved it to the about page, where it belongs, and everyone kept their pride.
A side effect worth naming: the page now opens with an answer above the fold on mobile. The old version showed a slogan and a stock photo until the second screen.
Word count is not the enemy. Words that answer nothing are.
What we added: dates, a byline, and an FAQ
Three additions shipped with the rewrite. A visible last-updated stamp, wired to the CMS so it stays honest. A byline naming the managing partner with a one-line credential — 19 years, 240 clients — plus matching Person schema. And a five-question FAQ block with FAQPage schema, built from the questions the firm's own inbox asks most. Total project time including review calls: 11 hours.
The FAQ questions came straight from the office manager's sent folder. The five most-repeated email answers became the five FAQ answers, lightly edited. Twenty minutes of inbox mining beat every keyword tool we own.
None of these three is clever. They are the same three moves we make on nearly every page, because they keep working.
At $200 an hour, call it a $2,200 intervention. Keep that number in mind for the results section.
The eight-week numbers
Week three: first Perplexity citation, on the cost question. Week five: two more queries citing, and the head term moved from position 28 to 14. Week six: first AI Overviews appearance. Week eight, end of measurement: cited on four of twelve tracked queries, position 9, organic sessions up 61% from an admittedly small base, three qualified inquiries, one signed at $1,400 a month.
Honesty about scale: 61% on a near-dead page is dozens of sessions, not thousands. The point is direction and repeatability, not fireworks.
But the signed client repays the $2,200 intervention roughly every seven weeks. That is the arithmetic that matters to a firm this size.
Bounce fell from 84% to 63% as well. People who land on an answer stay for the next one. We treat that as confirmation, not as the goal.
The AI citations arrived before the rankings moved. We see that ordering often enough now to treat it as the leading indicator.
What did nothing, and the checklist
One intervention flatlined: we tested two CTA button rewrites in weeks four through eight and neither moved inquiries at all — the answers were doing the selling before the button entered the picture. Worth knowing, because button copy is where teams love to spend attention. The rest of the playbook, in the order we would run it again: audit the twelve questions, rewrite around six buyer questions with answer-first leads, cut everything that answers nothing, then add date, byline, and FAQ with schema.
The sequence transfers. We have now run it on six pages across four clients, and week-three first citations have shown up on five of the six.
Eleven hours, no redesign, no new tools. The template never changed.
Every anonymized number here comes from the client's own analytics and our monthly citation audit, the same audit that opens every Avakata engagement.