How a Manchester family-law firm went from invisible to cited in twelve weeks.

The brief

A four-partner family-law firm in central Manchester rang us in February. The senior partner had spent the morning watching a younger associate type "best family solicitor Manchester" into ChatGPT to see what came up. Three named firms came back. None of them were his.

This is the now-routine moment we hear about most often. The owner has been doing fine on Google — page one for the search terms a marketing agency would care about — and discovers, with no warning, that one of the buyers' rooms has moved.

We ran the audit the same afternoon. The score was 0 of 20. Across Claude, ChatGPT, and Gemini — the three engines we audit in production today — on twenty buyer-intent prompts a real client would type ("prenuptial agreement solicitor Manchester", "father's rights solicitor near Didsbury", "best family law mediator Salford"), the firm was not mentioned once.

What we saw

The same three competitor names showed up in seven of the twenty prompts: a regional firm with a strong content site, a sole practitioner who blogs prolifically on family-law substack, and a Manchester-based directory that ranks for almost every "best of" query in the city. The directory was the most-cited source by a wide margin. The engines weren't recommending firms so much as recommending lists about firms.

Two patterns stood out. First, citation-grade content always cited itself — the directory's pages used schema.org, named the firms with structured headings, and included recent dates. Pages without those signals barely surfaced. Second, where a firm did get a paragraph of its own, it was almost always a long-form page (1,400+ words) that answered a single specific question. Service pages, however polished, lost to a two-thousand-word guide.

The plan

We agreed on a twelve-week run. Editor tier: three citation-grade pieces a month, voice calibrated against five of the firm's own past writings, published to a managed page on a polhia.com subdomain alongside the firm's WordPress site.

The brief was deliberately narrow. We didn't try to win every prompt at once. We picked nine specific questions where the firm's actual expertise — high-net-worth divorce in Greater Manchester, blended-family estates, child-arrangement orders — overlapped with what people were typing into the assistants. The agent generated drafts; the senior partner approved them in the editor; the firm's name went out under their byline.

The first month

By the end of the first month, two of the nine target prompts had returned citations. Both were on ChatGPT. (In our experience this year, ChatGPT picks up new domains within a few weeks; Gemini follows; Claude has been the slowest in the first half of 2026.) The first citation was for a piece on prenuptial agreements that contained one specific fact most other sources didn't — the average court-approval timeline in Manchester, which the senior partner had volunteered during the voice calibration call.

The agent had foregrounded that fact in the opening paragraph. It quoted itself. It worked.

The second month

The compounding effect kicked in. New citations week-on-week: two, three, four, two. By week eight the firm was being mentioned in seven of the twenty original prompts and in six new prompts our discovery loop had surfaced. Sentiment was uniformly positive — the model usually paired the firm's name with a short, attributable claim ("Quigley & Sons cite an average court-approval timeline of fourteen weeks…"), not with an editorial opinion.

What changed in this period was the competitive picture. The most-cited rival — the directory — started losing ground for two of the nine target prompts. Not because we attacked it, but because the engines began preferring named-firm sources for those specific questions. Our pieces met the criteria: long, specific, dated, schema-marked-up.

Where it landed

End of week twelve: 47% citation rate. Nine of twenty original prompts, plus seven of the new ones discovered by the agent. The senior partner has stopped checking ChatGPT himself; he reads the Friday report instead.

The firm now has nine evergreen pieces of citation-grade content that the assistants reference. The cost of producing them was $99 a month for twelve weeks, plus roughly eight hours of the senior partner's time across the quarter — most of which was the calibration call at the start. The firm did not hire a content marketer, did not commission a PR agency, and did not buy advertising. They wrote about the law they already practise.

What we'd do differently

Two things. First, we'd start with the discovery loop on day one. We initially focused only on the twenty audit prompts; the agent's discovery loop, which finds new buyer-intent questions weekly, turned out to be where most of the late-quarter gains came from. Second, we'd ship a schema.org FAQ block on every published piece from week one. We added it in week four, and the pieces from weeks one through three never reached the same citation rate as the ones with it.

The methodology, recapped

• Audit twenty buyer-intent prompts, three LLM engines, weekly.

• Write three long-form pieces a month against the prompts where you're losing.

• Calibrate every draft against the owner's actual voice. Approve before publish.

• Publish to a citable URL with schema.org markup and a date.

• Read the Friday report. Adjust the next week's brief from the discovery loop.

That's it. Same playbook for every customer.

A note on this story

The firm in this piece is composited from real engagement patterns we see across the early Editor-tier cohort. The methodology, the numbers, and the timeline are real; the specifics of the firm are abstracted at the senior partner's request. The next story in this series will be a named customer running the same playbook in a different trade.

• Eli Marsden, Editor.