Free · 60 s Free AI-readiness audit See your site the way ChatGPT does.
Personal reply within one business day.
CLIENT · NAME CHANGED

Law · Content engine + Edge Hub · 2026

How a German law firm put 1,100+ pages in front of AI search

For years the firm published its expert articles on a lawyers’ directory, so the directory got the credit. I built a system that brings them home to the firm’s own website. Each article is rewritten so AI assistants can quote it, every claim is checked against the original, and a lawyer approves each page with one click.

Live · built Jul–Sep 2026 · handed over Sept 2026 · numbers checked 2026-10-05

01 · Before

The firm’s articles were building someone else’s website

The firm had done the hard part long ago: written expert answers to the questions its clients ask. But it published them on a lawyers’ directory. Each article made the directory stronger. Someone who asked ChatGPT or Perplexity was sent there, not to the firm. Copying the articles by hand was out of the question. So was replacing a website the partners liked.

Before: who got the credit for the firm's expertise AI assistants and search engines cite a third-party legal directory, which holds many of the firm's articles. The firm's own website has no knowledge section, so the link to the firm is broken. ASKED BY BUYERS AI assistants & search cites THIRD-PARTY Legal directory many of the firm’s articles live here THE FIRM Own website no knowledge section to cite a contact form, nothing to quote
FIG 01 — Before: the expertise built someone else’s domain
Starting pointbefore
articles
problem: Years of expert articles, all on a third-party directory
own website
problem: No knowledge section, nothing for AI assistants to quote
AI visits
risk: Not measured. Nobody knew whether ChatGPT or Claude ever read the site
enquiries
risk: No way to tell which article brought a client in
website
note: Liked by the partners. To be left alone
risk
risk: Legal content: one invented claim is one too many
The brief: bring the articles home, leave the website alone, publish no unchecked claim.

Recognise this? See what AI sees on your site.Free AI-readiness audit · about 60 seconds.

02 · What I built

Two systems, one engine Gold marks a person’s decision. Teal marks AI at work.

Part 1 · The content engine From the directory to approved pages on the firm’s own website

Every 5 min, the system checks the directory and keeps a copy of each original. Claude, an AI model, rewrites each article into a short page that answers the question first, with an FAQ. A second AI checks every claim against the original. A claim it can’t find holds the draft for a person. For now, a lawyer releases each page with one click from a daily email. The firm can also let categories it trusts publish on their own; a draft that fails the check still waits.

  1. Step 1 (input): Directory profile, the firm’s articles.
  2. Step 2 (automated): Fetch & keep a copy, every 5 min.
  3. Step 3 (ai agent): Rewrite, answer first + FAQ. Rejected here: Over budget (€100 cap).
  4. Step 4 (ai agent): Claim check, against the original. Rejected here: Unsupported claims (flagged · held).
  5. Step 5 (human gate): Lawyer approves, pilot mode, one click.
  6. Step 6 (live): Published, firm’s own website.
  7. Step 7 (automated): AI visit count, 16 AI bots by name.
FIG 02 — The content engine, part 1 Seven steps, one person’s approval. Rewriting stops at the €100 monthly cap. A claim the original doesn’t support holds the draft for a person.
PLATE 01 — A knowledge page with its claim check, desktop · the daily approval email, phone · rebuilt mock, name changed

Part 2 · The Edge Hub One Word file became 230+ linked pages. The website itself was never touched.

The firm’s second website had its articles in one Word document. I turned them into pages with ordinary code, word for word. Claude wrote the short descriptions search results show. The pages sit inside the site’s existing header and footer, so they look like part of it. Then AI suggested links between the articles. Two checks threw out every link that was invented or off-topic.

  1. Step 1 (input): One Word file, 230+ articles.
  2. Step 2 (automated): Turn into pages, word for word.
  3. Step 3 (ai agent): Link suggestions, 1,175 proposed.
  4. Step 4 (automated): Do they exist?, phrase + page. Rejected here: −72 invented (phrases & pages).
  5. Step 5 (ai agent): Relevance check, second pass. Rejected here: −179 (off-topic).
  6. Step 6 (human gate): Person approves, one review.
  7. Step 7 (live): Live on the website, 924 links.
FIG 03 — The Edge Hub, part 2 AI suggests links. Code checks that each linked phrase and each target page exists. A second AI pass judges relevance. A person approves the result.

1,175 links suggested by AI

Check 1 · Does the link exist? −72 · 36 with made-up link text · 36 to pages that don’t exist
Check 2 · Is it on topic? −179 · off-topic
Check 3 · Approved 924 published

What the AI proposed — and what survived

1,175 links proposed. 924 survived.

One Word file became 230+ articles. An AI model proposed internal links between them. Three gates checked every suggestion before anything went live — and every reject was logged, not hidden.

  1. 1,175 links suggested by AI
  2. −72 invented: 36 with made-up link text, 36 to pages that don’t exist
  3. −179 off-topic
  4. 924 approved and published · avg 3.97 per article

Each dot is one real suggestion from the pipeline log.

Part 2 · Edge Hub internal links · Corvane Legal (client name changed)

03 · Who did what

Who did what AI drafts and checks. A lawyer decides what goes live.

  1. agent://researcher

    Mapped the directory, the firm’s website and which AI tools could reach it.

    Never allowed to: decide what gets built or published.

  2. agent://architect

    Drafted the database, the page design and a way to add the knowledge section without touching the WordPress site.

  3. human://vegardHUMAN GATE

    Approved the plan and set one rule: no page goes live without passing the claim check, and nothing publishes on its own unless the firm allows it for that category.

  4. agent://writer

    Rewrites each article into a page that answers the question first, with an FAQ, in the firm’s voice.

    Never allowed to: add a fact that is not in the original.

  5. agent://fact-checker

    Marks every factual claim as supported, partly supported or unsupported by the original. An unsupported claim holds the draft for a person.

    Never allowed to: let an unsupported claim through.

  6. human://clientHUMAN GATE

    A lawyer approves pages from the daily email and decides which categories may publish on their own. One click per article; each link works once.

  7. agent://tester

    Runs 649 automated tests against real copies of the directory pages, on every change.

  8. agent://watcher

    Counts visits from 16 AI crawlers by name, warns when they stop coming, and stops all AI spending at the €100 monthly cap.

    Never allowed to: spend past the cap.

  9. human://vegardHUMAN GATE

    Approved every release, then moved the whole system into the firm’s own hosting account in Sept 2026.

04 · Results

Results so far Each number dated, with how it was counted.

Law firm content engine: results with method and date
NumberWhat it measuresHow it was measuredObserved
1,100+✓ verified pages live on the firm's own domain live sitemap count 2026-10-05
230+✓ verified pages on a sister domain live sitemap count 2026-10-05
924✓ verified AI-suggested links between articles that passed both checks and went live pipeline log 2026-10-05
72✓ verified invented links thrown out (36 with made-up link text, 36 pointing to pages that don’t exist) deterministic validation log 2026-10-05
179✓ verified irrelevant links rejected by relevance check pipeline log 2026-10-05
649✓ verified automated tests test suite count 2026-10-05
16✓ verified AI crawlers tracked by name crawler module 2026-10-05
Not measured yet How often ChatGPT, Perplexity and Google’s AI answers name the firm being measured now; first report at the 90-day review —
Not measured yet Enquiries traced to a specific article tracking is live; too early for a count —

Page counts come from each website’s own list of pages (its sitemap). I publish traffic and AI-mention numbers only after a full quarter of data.

05 · Running costs

What it costs to run Capped, predictable, billed to the firm directly.

AI rewriting runs at a fixed pace. If anything goes wrong, a hard €100 monthly cap stops every AI call. The bills go to the firm’s own accounts. I add no markup.

AI rewritingClaude, at a fixed daily pace
~€33/mo
Monthly capa hard stop, set by the firm
€100
Back catalogue, one-offevery old article reworked, spread over several months
~€230
Fetching the articlesa month, inside the provider’s free tier
~330 credits
Stored originalsa copy of every original page, so any page can be redone for free
~$0.13/mo

06 · Ownership

What the firm owns

  • The hosting accountThe whole system moved into the firm’s own Cloudflare account in Sept 2026.
  • The codeAll of it. Every update goes live from the firm’s own account.
  • Every page and every originalPublished pages, drafts, earlier versions and the stored originals.
  • The AI account and the budgetThe firm’s own account with Anthropic, the company behind Claude. The €100 cap is a setting it controls.
  • The tests649 automated tests that check the system after every change.
  • Admin, manual and a way backA protected admin area, a written manual, and a one-step way to undo a release.

07 · What I’d build for you

Turn what your firm has already written into pages AI assistants quote

If your firm has years of expert writing in Word files, PDFs, a directory or an old blog, I’ll build the same system for you. Start with an Edge Hub beside your current site, or go straight to the full engine.

  • On your own website and your own accounts from day one
  • Every claim checked against your original. You approve each page and choose what may publish on its own
  • A hard monthly AI budget that you set
  • A count of which AI tools visit, and which article brought each enquiry

Not for you if you want volume without review. Every page passes the claim check, and a person decides what publishes.

The offer that fits · Quoted by AI

Content Engine · Full Engine

from €9,500$10,500 typical €9.5k–18ktypical $10.5k–20k

6–9 weeks · Continuous capture → rework → fact-check → approval → publish.

Or start smaller · Quoted by AI

Content Engine · Edge Hub

from €4,500$4,950 typical €4.5k–7ktypical $5k–7.5k

2–3 weeks · Your back catalogue as a knowledge hub beside your current site.

Not sure which fits? Book 20 minutes with me
08 · Under the hood For your web developer: the technical detail
Runtime
Cloudflare Workers in TypeScript, no framework. D1 with 36 migrations, KV for pages and the raw archive, Cron every 5 min with a scheduler lock and a heartbeat ping to an external monitor.
Pipeline
scan → fetch → parse → rework → verify → publish → digest. The parser fails loudly: if the source markup changes, the source is flagged and an alert goes out. Nothing is guessed.
Models
Claude Sonnet rewrites; Claude Haiku writes meta descriptions and runs structural QA. JSON-LD is assembled in code, never by a model.
Approvals
Approval links are random, single-use, bound to one article, stored only as a hash, and expire after 72 h. Client preview links are revocable.
Edge overlay
The Worker is bound to one path route, so it cannot touch any other URL. If it throws, the request passes through to WordPress. The Edge Hub borrows the live header, footer and cookie banner and caches that shell for 1 h.
Link validation
Every AI-suggested link must use an anchor that exists word for word in the article and point to a page that exists. 72 failed. A relevance pass then rejected 179 more, leaving an average of 3.97 links per article.
AI crawlers
Explicit allowlist, llms.txt, sitemaps, IndexNow, and per-bot hit counts for 16 named AI crawlers.
Attribution
The article a visitor read travels in the URL to the contact form. No cookies, no storage.
Tests
649 Vitest tests in 42 files, runnable offline in mock mode against real captured pages.
Handover
Zero-downtime move into the firm’s own Cloudflare account in Sept 2026. CI deploys on every push to main, admin sits behind Cloudflare Access, a written runbook, and a rollback Worker kept in place.