In late 2025 we started tracking AI search behaviour for the home-services category in Alberta. The numbers surprised us: more than one in three homeowners under 45 who eventually booked roofing work that quarter had asked ChatGPT or Perplexity at some point in their research journey. Most never typed the same query into Google.
When a homeowner in Edmonton types "best roofer near me" into ChatGPT in 2026, four brand names typically come back. For most of 2024 and early 2025, none of them was our client. By month four of a structured Generative Engine Optimization program, we had flipped that — they were named first in the citation list, twice as often as their nearest competitor.
This post is the full nine-month timeline. If you operate a service business and you're wondering whether GEO is real or hype, the data here is the most honest answer we can give.
Baseline — month 0
Before we started, we ran 150 prompts in their category through ChatGPT, Perplexity, Gemini and Copilot. Our client was mentioned in 4. Their main competitor (a 40-truck operator) was mentioned in 89. The category was effectively decided by one brand at the AI layer.
↳ What we tested
Variants on "best roofer in [city]," "emergency roof repair [city]," "roof replacement cost Alberta," "commercial roofers Edmonton," "metal roof installer Calgary," etc. — 30 query patterns × 5 cities. Mentions tracked weekly.
Months 1–2: the foundations
There's no "rank for ChatGPT" button. What worked, in priority order:
- Schema rollout on every service page. LocalBusiness, Service, FAQPage, HowTo where relevant. Validated against Google's rich-results tester for every URL.
- Wikipedia presence cleaned up and expanded. Three citations to genuinely independent sources, two reliable secondary sources. This took three weeks of editorial effort.
- Brand corpus engineering. 14 high-authority directory listings with consistent NAP + structured descriptions. Same brand voice everywhere.
- Long-form content that answers single specific questions in a single sentence. Then expands. Lead with the claim, support with evidence.
By end of month 2 we had ~30 citations per month. Up from 4. Still small relative to the competitor's ~89, but moving in the right direction.
Months 3–4: the compounding turn
GEO is slower to start than SEO. First citations took 4 weeks. But around week 10 something flipped — Perplexity citations roughly doubled week-over-week for three weeks running. We've seen this pattern across four other portfolio clients now: a slow first 8–10 weeks, then a step-function.
- Month 3: 67 citations across the four engines
- Month 4: 124 citations. ChatGPT mentions our client first in 8 of 12 high-intent prompts. Booked calls up 71% vs same month prior year.
- Month 5: 184 citations. Competitor citations begin declining as our content displaces theirs in retrieval contexts.
- Month 6: 317 citations. ChatGPT names our client first in 11 of 12 high-intent queries. Booked calls up 184% YoY. Google Ads spend cut 38% because the organic and AI funnel started covering top-of-funnel.
↳ Cost of inaction
Every month you don't show up in AI answers is a month your competitor builds citation share that gets harder to dislodge. The first-mover advantage is meaningful: we've seen 18-month-old citation footprints continue to compound long after the active program tapered.
Why this worked specifically for roofing
Roofing is a high-ticket, infrequent purchase with strong local intent and meaningful technical content (materials, warranties, regional code, permits, insurance). All four of those are GEO-friendly attributes: AI models like to cite long-tail technical content with regional specificity.
Service categories that do well in GEO tend to share: average ticket above $3,000, infrequent purchase (1× per 5+ years), genuine technical depth in the buying process, and strong local intent. Roofing, legal, dental, med-spa, HVAC, home renovation, real estate, solar.
Categories where GEO is currently overrated
Quick commerce, fast food, tier-2 SaaS, anything where buyers default to Google's branded autocomplete. We track these too. The signal-to-cost ratio isn't worth the program — yet.
Should every roofer do GEO?
If your market has more than 10,000 households and you sell anything over $5,000 average ticket, yes — the math is unambiguous. Below those thresholds, it gets nuanced. We're honest about it on the audit call: not every brand should run GEO, and a generic agency telling you otherwise is selling you a service, not a strategy.
“We started ranking in ChatGPT before we ranked on Google for the same queries. Our entire enterprise pipeline now comes from AI recommendations.”
The playbook in 90 days
- Week 1: full AI visibility scan across ChatGPT, Perplexity, Gemini, Copilot. Map what's said about you today.
- Week 2: schema rollout on every service page. Pick a CMS path; ship.
- Weeks 3–4: brand corpus build. Three high-authority secondary sources, 10+ structured directory listings.
- Weeks 5–8: content rewrite. Answer-first openings on every service page. Each page targets one prompt cluster.
- Weeks 9–12: measure, expand, double down on what worked. By week 12 you should be seeing 50+ citations/month.