Roofing Contractor — +340% AI Citation Lift in 30 Days
The situation
A regional residential roofing contractor with 3 service locations ranked top-3 on Google Maps in every service area. Yet when prospective customers asked ChatGPT, Claude, Gemini, or Perplexity for “best roofer in [city]” or “roof repair near me,” the brand was not named in 92% of relevant queries. AI engines were defaulting to national lead-gen aggregators (Angi, HomeAdvisor, Networx) and a single regional competitor who had llms.txt deployed.
Pre-deploy baseline: 5 mentions out of 60 queries (8%). Their classic SEO was strong. Their AI discovery foundation was missing.
What we did
1. Pre-deploy baseline scan — 60 queries × 4 AI platforms
Established a measurable starting line. 15 queries per platform across ChatGPT, Claude, Gemini, and Perplexity. Logged mention rate, citation position, and which competitors were winning. Result: 8% mention rate, never cited as #1.
2. Wrote a custom 1,800-word llms.txt
Geographic specificity for each of 3 service areas with explicit 25-mile radii. Declared license number, BBB rating, review count, service categories, common customer questions in the exact phrasing customers type, financing options, warranty terms, and authoritative source pages. AI engines could now reliably answer “is [client] in my area?”
3. Updated robots.txt with explicit AI crawler allow-list
GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Applebot-Extended. Previous robots.txt was implicitly allowing some and explicitly blocking Google-Extended — a common 2025/2026 misconfiguration that excludes a site from Gemini training data updates.
4. Initial schema overlay — LocalBusiness + Service + FAQPage
Deployed JSON-LD on 5 high-value pages (homepage + 3 location pages + 1 services page). FAQPage schema wrapping the existing visible FAQ content in the 40-60 word answer format AI engines preferentially quote.
5. 30-day post-deploy rescan + delta report
Re-ran the identical 60-query battery on day 30. Generated a side-by-side delta report so the client could see exactly which queries flipped, which platforms gained the most, and which competitors lost share.
The results
30-day post-deploy rescan against the identical 60-query battery:
| Platform | Pre-Deploy | 30-Day | Lift |
|---|---|---|---|
| ChatGPT (GPT-4.5) | 1 / 15 | 5 / 15 | +400% |
| Claude 3.7 Sonnet | 0 / 15 | 4 / 15 | +∞ |
| Gemini 3.5 Flash | 2 / 15 | 6 / 15 | +200% |
| Perplexity | 2 / 15 | 6 / 15 | +200% |
| TOTAL | 5 / 60 (8%) | 21 / 60 (35%) | +340% |
In real-business terms, the citation lift translated to an estimated 3-5 additional booked estimates per month within 60 days, based on attribution conversations during sales calls.
What we learned
Three reasons most llms.txt deployments fail — and what we did differently:
- Geographic specificity beats vague service-area declarations. “Serving the United States” gets you nothing. Each of 3 cities + 25-mile radius gets you cited for local queries.
- Question phrasing matters more than marketing copy. AI engines preferentially quote answers that literally match the question. “How much does a new roof cost?” beats “Our Pricing” every time.
- Credential signaling as plain facts. License number, BBB rating, review count, warranty terms — stated as bare facts AI can extract verbatim. Most contractors bury this in legalese.
Want results like this for your business?
Start with the FREE AI Visibility Audit + Fix-It Kit (normally $199). Full visibility score, ready-to-paste fixes, 5-step action plan custom to your site. FREE this month only.
Get My FREE Audit →