A UK legal practice

Problem

The firm had a website that worked as a brochure and did nothing else. It was built on a website builder, and the platform had left its fingerprints in places that mattered.

• Generated and placeholder URLs, including addresses of the form blank and blank-1, that described nothing about the page behind them.
• Legal subjects covered thinly, with no consistent structure between language versions.
• Page architecture, titles and metadata inherited from the template rather than decided.
• No measurement of whether AI systems were reading the site at all.

None of this produced a complaint. A firm in this position does not lose enquiries it can see; it never receives the ones it might have had.

Diagnosis

Two questions had to be separated before anything was rebuilt, because they fail independently.

• Can the material be found? Which subjects the firm could realistically compete for, and whether the pages that should answer them existed in each language.
• Can the material be read? Whether automated systems, including those behind AI assistants, were fetching the site, and at what volume. A system can only use a firm's material if it has retrieved it first.

The second question is the one nobody was asking. It is answered in server logs, not in an analytics dashboard.

Implementation

• Content. Five legal subjects developed and structured across three languages, producing 15 indexed pages covering the firm's core areas.
• URL structure. Platform generated addresses replaced with subject specific URLs as part of the rebuild.
• Architecture and metadata. Page structure, titles, metadata, the multilingual structure and the internal organisation of the site rebuilt alongside the content.
• Instrumentation. Server log analysis set up so that crawler and AI activity could be reported separately from search performance.

Measurement

Two sources, kept apart from the first day and never added together.

• Google Search Console for impressions, clicks, queries and average position.
• Server logs for automated activity. A visit is one session by one automated system. A request is one page fetch inside that session. Monthly trends here are reported in visits, the breakdown by provider in requests, and the two are never compared with each other.

Mixing the two would produce an impressive number and a meaningless one. Every figure on this page carries its source.

Result: search

Impressions rose from 559 in January to 1,439 in August, an increase of 157%, with 7,306 impressions across the eight months.

Clicks are small in absolute terms, so month to month percentages are not meaningful. Comparing the two halves of the period is: clicks rose from 20 in January to April to 44 in May to August, an increase of 120%, while click-through rate rose from 0.80% to 0.92%. The site was appearing more often and being chosen more often when it appeared.

The five new subjects had recorded no impressions before January 2026. They earned 2,358 impressions and 20 clicks during the period, which is 32% of the site's impressions and 31% of its clicks from a small share of its pages. All five reached an average position on the first page of Google in at least one language.

Of the 204 queries the site appeared for, 97% were non-branded, 89 sat at average position 10 or better and 28 at position 3 or better. The firm was being found through legal questions, not through people already searching for its name.

One further finding shaped the next content plan: the language versions did not behave alike. One non-English version produced a click-through rate almost three times higher than the English pages, while English generated most of the exposure. Exposure and conversion were coming from different places.

Result: who is reading the site

Automated visits rose from 249 in January to 3,197 in August, an increase of 1,184%, close to thirteen times the January level. The daily average moved from 31 in July to 103 in August.

Full provider level detail is available for July and August, so those months are compared directly. Requests from AI systems rose from 967 to 3,839, an increase of 297%. AI systems accounted for 66% of all automated requests to the site in August, against 37% in July.

All six major providers were present, identified through nine separate crawlers. The largest single change came from one search crawler operated by a major AI provider, which rose from 63 requests in July to 1,121 in August. That crawler fetches pages when content is being retrieved for search and answer generation, so it is a direct technical signal that the firm's material is being pulled into that pipeline.

For a firm of this size, coverage across every major provider is unusually broad.

What this does not show

Three limits, stated because a case study that hides them is not evidence of anything.

• Crawler activity is not citation. Retrieval means the material can be used. It does not prove that any particular AI answer named or linked the firm. Citations were not measured in this period and no figure here should be read as one.
• August is a transition month. Average position held between 6.4 and 8.7 from January to June, then moved to 11.3 in July and 19.7 in August, while impressions stayed high. The rebuild of content, URLs and architecture was being implemented during August.
• September is a new baseline. The new site went live on 1 September 2026. Because URLs and internal linking changed substantially, the following months are measured as a fresh series rather than as a continuation of the old one. The first month after a change of this size is always unsettled.