The case study is this website: how we got cited by ChatGPT & Perplexity for Miami lead generation.
Every agency now claims it can get you “found by AI.” Almost none of them show their work. So here is ours, run on our own domain, with dates and sources on every number: in September 2026 we logged 40+ inbound leads in a single month — counted across texts, calls, forms and email — and every tracked lead we’ve asked traces back to an AI assistant’s recommendation. AI-referred sessions grew roughly 9× month over month, Microsoft Copilot recorded its first citations of our guides, and we spent $0 in ad spend on the channel. This page is the timeline, the receipts, and what we’d copy first.
Every number, dated and sourced.
We track this channel three ways: Microsoft Clarity’s AI-referral segment and Citations module, analytics referrers (chatgpt.com, perplexity.ai), and — most reliably — asking every lead where they found us. Here’s what each one showed.
| Metric | Number | Period | Source |
|---|---|---|---|
| Inbound leads, one month | 40+ | September 2026 | Our own all-channel count (texts, calls, forms, email) |
| AI-referred sessions, 30 days | 3 → 28 (~9×) | 30d to Sept 1 vs 30d to Sept 16 | Microsoft Clarity, AI Referrals segment |
| Share of all site sessions | 2.76% (93% new visitors) | 30d to Sept 16 | Microsoft Clarity |
| Top AI referrer | chatgpt.com (then Perplexity) | September 2026 | Clarity + GA4 referrer data |
| First Copilot citations | 6 citations, 24% share of authority | recorded Sept 16, 2026 | Clarity Citations module (Copilot + partners) |
| AI-referral close rate | ~40–50% vs ~20% benchmark | August 2026 (4–5 leads, 2 closed) | Our own pipeline |
| Ad spend on this channel | $0 | entire period | — |
| Leads from cold outreach, ever | 0 | through Sept 2026 | Our own pipeline — for contrast |
Honest caveat: tracked clicks undercount this channel badly. Most AI-referred prospects read the assistant’s answer, then text or call without ever clicking — August’s leads mostly arrived with no tracked referrer at all. The lead count is ours; the session and citation data is Microsoft’s.
Twelve weeks from foundation to 40+ leads a month.
- 1July 2, 2026 — the foundationShipped
llms.txtandllms-full.txt(plain-text site summaries AI crawlers read), opened robots.txt to GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot, and put Article + FAQ schema on every guide. No results yet — this is plumbing. - 2August 23 — first tracked AI visitPerplexity sent its first recorded visitor — not to the homepage, but deep-linked straight into a guide about roofing website costs. First signal that answer pages, not brand pages, are what assistants lift.
- 3August 25 — IndexNow + published benchmarksSet up IndexNow (instant Bing indexing — Bing feeds ChatGPT and Copilot), submitted all 178 URLs, and published our lead-cost benchmarks guide — original data no other page could offer.
- 4September 1 — the indexing wake-up callSpot-checked our flagship pages in Google Search Console: three of three had never been crawled — “Discovered, currently not indexed.” Our best citation bait had been invisible to Google-grounded AI answers the whole time. Requested indexing manually, then made re-indexing a standing routine on every deploy.
- 5September 12 — the jumpAI-referred sessions hit ~15× the prior run-rate within two weeks, 5 of the last 7 via chatgpt.com. Assistants were deep-linking our pricing, contact and Spanish pages directly.
- 6September 16 — first Copilot citationsMicrosoft’s Citations module recorded its first data: 6 citations, 24% share of authority — all from one query (“landscape marketing ideas miami fl”) pointing at one listicle with lead costs in it. We cloned that exact pattern into 12 more trade-×-city idea guides the same day.
- 7September 30 — 40+ leads in the monthClosed September with 40+ inbound leads across texts, calls, forms and email — against 4–5 in August. Same site, same phone number, $0 in ad spend on the channel. The compounding came from pages published weeks earlier finally being indexed, cited and recommended.
The five lessons we’d copy first.
Verify indexing — don’t assume it
The single biggest unlock was discovering our best pages had never been crawled. A page Google hasn’t indexed doesn’t exist to Gemini, and a page Bing hasn’t indexed doesn’t exist to ChatGPT or Copilot. Check every important URL in Search Console and Bing Webmaster Tools; use IndexNow for Bing. This costs nothing and most sites skip it.
Numbers get cited; slogans don’t
Our first Copilot citations came from a listicle with dollar figures in it — not our homepage, not our services page. Assistants lift pages that already contain the answer: a price, a comparison, a dated stat. If your page can’t complete the sentence “roofing leads in Miami cost ___,” it won’t be the source.
Assistants deep-link interior pages
Our AI entry pages were pricing, pay-per-lead, contact and Spanish-language pages — rarely the homepage. Every page has to stand alone as an answer, because the assistant decides where the conversation lands, not your navigation.
Freshness is a ranking input
We bump key pages every 2–3 weeks with a genuinely new month of data — not a fake “updated” stamp. Recently-updated pages visibly win answer slots, and each bump is another IndexNow ping.
Don’t trust click tracking to grade the channel
28 tracked AI sessions and 40+ leads in the same month only reconcile one way: most AI-referred buyers never click. They read the recommendation, then text or call. If you grade this channel on referrer traffic alone you will kill your best lead source — ask every lead where they found you.
The contrast that convinced us
Over the same months, our cold outreach — daily, systematic, personalized — produced zero leads. The AI-citation channel produced all of them, at a ~40–50% close rate, because an AI-referred prospect arrives pre-compared. We now treat published, structured, corroborated content as the pipeline, not a side project.
Questions about this case study.
How do you count the 40+ leads in this case study?
It's our own count of inbound leads for September 2026 across every contact method — texts, calls, website forms and email. Most AI-referred prospects text or call rather than fill out a form, so form submissions are the minority of the total and analytics referrer data undercounts the channel. When asked where they found us, tracked leads consistently name an AI assistant; our cold outreach has produced zero leads to date.
How long did it take to get cited by ChatGPT and Perplexity?
About 7–8 weeks from foundation to first measurable results. We shipped llms.txt, AI-crawler access and schema in early July 2026, recorded our first tracked Perplexity visit on August 23, and by mid-September had our first Microsoft Copilot citations and a ~9x month-over-month jump in AI-referred sessions (3 sessions in the 30 days to September 1 vs 28 in the 30 days to September 16, per Microsoft Clarity).
What moved the needle most?
Three things, in order: verifying indexing instead of assuming it — three of three flagship pages had never been crawled by Google until we manually requested it; publishing answer-shaped pages with real numbers in them — our first Copilot citations came from a specific listicle with lead-cost figures, not the homepage; and freshness — updating key pages every 2–3 weeks with a new month of data.
How much did this cost in ad spend?
Zero. The AI-citation channel runs entirely on published content, structured data, indexing hygiene and third-party corroboration. The cost is labor: writing pages that actually answer buyer questions with numbers, and keeping them fresh.
Can a contractor replicate this?
Yes — the mechanics are identical for a roofer or HVAC company: be indexed in Google and Bing, publish pages that answer the questions homeowners ask AI assistants (with prices and specifics), add FAQ and Article schema plus an llms.txt file, allow AI crawlers, stack reviews and directory corroboration, and keep pages fresh. The full method is in our 8-step playbook, and we build it for contractors as a service.
Want AI assistants recommending your company? We’ll show you the gap, free.
The free audit covers whether your key pages are even indexed, what AI assistants currently say when asked about your trade in your city, and the shortest path to being the answer — in writing, within 24 hours. From the agency that ran this experiment on itself.
No card. No contract. Takes one phone call.
The method behind this case study, step by step: