CASE STUDY
GEO & AEO PRACTICE
By 2025 customer research was moving into AI assistants, and Cox’s most important pages were invisible to the crawlers assembling the answers. I built the GEO and AEO practice from zero: a 10 to 15% citation lift and Cox positioned as the most-cited telecom in unbranded AI search.
CLIENT
Cox Communications
YEAR
2025 to 2026
ROLE
Practice lead, Director of UX, BBDO Chicago
DELIVERABLES
Landscape POV · Noscript workflow · llms.txt guide · 37-item QA protocol · Citation benchmarking · Competitive analysis
THE SITUATION
By 2025, a meaningful share of customer research had moved out of search results pages and into AI assistants. When someone asked ChatGPT, Perplexity, or Google’s AI Overviews which internet provider to choose, the answer was assembled from whatever those systems could crawl, parse, and cite. Cox’s most important pages were rendered client-side in JavaScript, which meant the crawlers building those answers often saw nothing at all. The category had the same blind spot, and nobody had a playbook, because the discipline barely had a name yet.
THE APPROACH
01
Establish the landscape and the stakes.
I developed an AI Search Landscape POV for Cox leadership mapping how each major answer engine discovers, parses, and cites content, and where Cox stood relative to competitors. It reframed AI search from a curiosity into a measurable competitive battleground, covering zero-click behavior, GEO and AEO mechanics, and agentic commerce.
02
Solve the JavaScript visibility problem.
I designed a noscript content generation workflow for Cox’s JS-heavy pages, producing crawler-accessible, semantically structured fallback content so AI systems could read the pages that mattered. The single biggest technical liability became addressable, prioritized work.
03
Implement emerging standards early.
I authored Cox’s llms.txt implementation guide before most of the category had heard of the standard, a first-mover position on how models discover brand content. As the evidence matured through 2026, I demoted llms.txt from strategy to hygiene in my client guidance; the audit and rendering work is what moves citations, and saying so plainly is part of the practice.
04
Make quality verifiable.
Optimization without verification is guesswork, so I built a 37-item QA audit protocol on a Screaming Frog dual-crawl methodology: crawling every page with and without JavaScript rendering and diffing exactly what AI crawlers see against what humans see. Every fix became testable.
05
Benchmark the competition.
I led a competitive analysis of telecom AI search readiness, auditing crawlability, structured data, and crawler governance across T-Mobile, Verizon, AT&T, Comcast, and Cox, converting an abstract goal into a scoreboard.
06
Measure what actually matters.
I stood up monthly unbranded citation benchmarking: tracking which brands the answer engines cite when a customer asks the question without naming anyone. Unbranded queries are where consideration is won, and they became the practice’s north-star metric.
WITH JAVASCRIPT
What a person sees: full navigation, product detail, pricing, and the structured copy that answers the question.
WITHOUT JAVASCRIPT
What the AI crawler often sees: an empty shell, missing content, and nothing to cite in the answer about you.
FIG 1 · The dual-crawl diff. The same URL crawled twice, human render versus crawler render, with the missing content exposed.
OUTCOMES
#1
Most-cited telecom in unbranded AI search queries
10-15%
Citation lift versus competitors in AI answer engines
37
Items in the repeatable dual-crawl QA protocol
Beyond the metrics, the engagement produced a durable capability: a documented methodology (landscape POV, noscript workflow, implementation guides, QA protocol, benchmark program) that converted a one-off request into ongoing client work, and a practice area that did not exist at the agency before and is now part of how BBDO Chicago positions itself on AI.
Interactive · what the crawler saw before, and after the noscript workflow shipped.
WHAT THIS DEMONSTRATES
Building in ambiguity. There was no playbook for GEO and AEO. The work required understanding how LLM-based systems actually retrieve and cite content, translating that into technical requirements a client engineering team could execute, and wrapping it in measurement rigorous enough to defend the results. It is the same muscle as any zero-to-one product problem: define the opportunity, design the system, instrument the outcome.
NEXT PROJECT
