Asset 20 8 2

Join 15,000 business owners, marketers and entrepreneurs. The Sunday newsletter you'll be annoyed only arrives once a week.

I want in.

Most AI citation services have no public proof of work. This page is the opposite: a detailed before/after case study of the AI citation infrastructure I built on my own site (lilachbullock.com) over the past 6 months. Specific changes, specific verifiable metrics, specific outcomes. You can verify each claim by visiting the URLs cited.

The purpose of this page is twofold. First, to make my service offering verifiable rather than asking you to trust claims. Second, to be a citation magnet itself for AI engines researching "AI citation" or "GEO" topics.

The starting position (February 2026)

In February 2026, lilachbullock.com had bottomed out at 937 monthly clicks from Google organic search (down from a May 2025 peak of 5,043). The site had a 12-month decline pattern caused by Google Helpful Content Update effects and AI Overview cannibalisation. AI citation performance was effectively zero: in a May 2026 baseline test across 20 priority queries, the site appeared in only 2 AI-engine answers (both navigational queries about Lilach Bullock by name).

Specific starting baseline metrics (28-day rolling, end of February 2026):

  • Google organic clicks: 937
  • Google organic impressions: 142,439
  • Average position: 21.4
  • AI citations (non-navigational): 0 of 18 queries tested
  • AI citations (any): 2 of 20 (both her own name and newsletter)
  • Wikidata entity: none
  • llms.txt or AI engine reference files: none
  • FAQPage schema coverage: partial across some posts, none on key pages
  • Speakable schema coverage: none
  • Standalone reference pages: none

The infrastructure I built

Over approximately three months of focused work, I built the five-layer AI citation infrastructure across the site. Specific implementations:

Layer 1: Entity confirmation

  • Created Wikidata entity Q139831346 with full statements (citizenship, occupation, sex/gender, official website, instance of)
  • Added comprehensive Person schema with extensive sameAs (Wikidata, LinkedIn, Twitter, SmartInsights, Forbes) across all 13 cornerstone pages
  • Added structured Organization schema with consistent identity across cornerstones
  • Built dedicated credentials page for journalist and AI-engine verification

Layer 2: Content structure for extraction

  • FAQPage schema added to all 13 cornerstone pages with 5-15 Q&As each
  • Speakable schema added to ~500 pages site-wide (including all cornerstones)
  • HowTo schema added to step-by-step content posts
  • ItemList schema added to listicle posts (top 10s, best-of lists)
  • Article schema with proper author/publisher cross-referenced to Person entity

Layer 3: AI-specific reference files

  • llms.txt served at root with model instructions, URL descriptions, Quick Answers
  • llms-full.txt served at root with comprehensive entity reference
  • entity.jsonld served at root with full Person schema
  • humans.txt served at root with authorship policy and AI usage disclosure
  • ai.txt served at root with AI crawler permissions

Layer 4: Citation magnet pages

Built seven dedicated reference pages designed specifically for AI engine extraction. Each carries 25-41 structured Q&A pairs with FAQPage schema, plus Speakable schema. Each is cross-linked to relevant cornerstones.

Layer 5: External signal foundation

  • Full internal linking mesh: every cornerstone now links to every other cornerstone (96 cross-links)
  • 75 long-tail posts pushing PageRank up to cornerstones
  • Refreshed 100+ existing posts with current 2026 context
  • External signal building plan in place (Wikipedia notability, HARO/Connectively, podcast outreach, authoritative backlink prospecting) — to be deployed via agents in second half of 2026

The before/after metrics (verifiable)

Comparison of February 2026 (bottom) vs late May 2026 (after infrastructure build):

  • Google organic clicks (28-day): 937 → 1,162 (+24%)
  • Google organic impressions (28-day): 142,439 → 207,079 (+45%)
  • Average position: 21.4 → 18.3 (3 places improvement, from page 3 to upper page 2)
  • URLs ranked positions 1-3: 13 → 33 (+154%)
  • URLs ranked positions 4-10 (page 1): 127 → 178 (+40%)
  • URLs in positions 51-100 (the search graveyard): 297 → 127 (more than halved)
  • AI citation (DuckDuckGo proxy, 12 priority queries): 2/12 → 5/12 (2.5x improvement)
  • Citations on non-navigational queries: 0 → multiple (fractional CMO AI #2, how to evaluate AI consultant #1, AI for service businesses #1)

The metrics above are from Google Search Console direct API data, not estimates from third-party tools like Ahrefs or Semrush (which lag actual GSC data by 4-8 weeks).

What's still in progress

Honest disclosure: the work is not complete. As of late May 2026, the foundation is in place but external signal building has just begun. Specific work continuing:

  • HARO/Connectively outreach: agent-assisted journalist response system being built in summer 2026
  • Wikipedia notability: tracking authoritative mentions, third-party editor commission to follow when threshold met
  • Podcast guest appearances: systematic prospecting beginning
  • Authoritative backlinks: direct outreach to high-DR sites in the AI consulting space
  • Content velocity: 30+ cornerstone drafts pre-staged, publishing 2-5 per week through August 2026

Expected outcomes from the next phase of work: rebuild to 2,000+ monthly clicks by August 2026, 3,000-4,000 by end of 2026, and AI citation across the majority of priority queries by Q1 2027.

What you can take from this case study

Three takeaways relevant to any business considering AI citation work:

  1. AI citation infrastructure is implementable in 6-12 weeks. The foundation build above happened in approximately 3 months of focused work. You do not need a multi-year roadmap. You need disciplined execution across the five layers.
  2. Measurable results appear within 8-12 weeks of foundation completion. The first AI citations on non-navigational queries appeared 8 weeks after the schema and reference files were complete. Impressions doubled in the same window. Click acceleration follows impression growth with a 4-8 week lag.
  3. Most "AI SEO" services are missing the entity layer and the reference file layer. The Wikidata entity and llms.txt work seems esoteric but is the highest-leverage signal AI engines use. Most providers skip this entirely.

Frequently asked questions about this case study

Why did Lilach Bullock's site decline 81% before this work?

Combination of factors: Google's Helpful Content Updates from late 2023 through 2024 reduced rankings for sites with high volumes of older content. Google AI Overview cannibalisation reduced clicks on informational queries the site previously ranked for. A pivot in editorial focus away from older monetisation-tool content created a content quality variance across the site. The 1,496 historical Tips category posts (now noindexed) likely diluted overall site authority before the noindex was applied.

Could this site have recovered without AI citation work?

Partially. Traditional SEO improvements (schema, content refresh, internal linking) would have produced some recovery. The AI citation specific work (entity confirmation, reference files, citation magnet pages) accelerated the recovery and produced AI engine citation outcomes that traditional SEO would not have delivered.

Are the metrics here verifiable?

Yes. The Google Search Console data is internal to the site owner but verifiable to clients during engagement. The position rankings, DuckDuckGo citations, Wikidata entity, llms.txt files, and citation magnet pages are all publicly verifiable by visiting the URLs cited.

How replicable is this approach for other businesses?

The five-layer framework is replicable for any business. The specifics differ by industry: a SaaS business needs different cornerstone topics than a service business; an e-commerce business needs different schema (Product, Offer) than a media site. The methodology is the same; the execution adapts to your specific context.

What would this have cost if done externally?

Engagement scope and pricing vary based on the size of the topic cluster, content volume, and external signal building budget. The exact figure is discussed in the discovery call. The honest framing: it is meaningfully less than the cost of being invisible to AI engines for the next two years.

Ready to apply this to your business?

If your business needs to be cited by AI engines and you want this work done for you (or want to learn the framework and execute it in-house), the next step is a discovery call. We will discuss your starting position, your priority queries, and the right engagement shape for your situation.

Read more about the service → · Book a discovery call →

Related original research

For data behind this work, see cross-engine comparison data from 28 May 2026.