Case study: our own site, before and after a GEO rebuild (measured)
Short answer: we rebuilt our own website for AI search (GEO) on 15 November 2026 and measured the result with the same script before and after. Every page now has machine-readable facts: 11 AI and search crawlers named and allowed in robots.txt, an llms.txt summary, JSON-LD on 35 pages, FAQPage markup on 18 pages (107 questions), a 35-URL sitemap, and canonical tags on all 35 sitemap pages. What we have not measured yet: whether ChatGPT, Perplexity, Google AI Overviews or Gemini mention or cite us, and any traffic or ranking change. Those results are pending, and this page will be updated with dated numbers once they exist.
This is a case study of our own site, not a client project. We publish it because it is the one GEO project where we can show real before/after data.
What was measured, and how
- Before: the previous version of this site on its earlier host, checked live on 15 November 2026 between 18:50 and 19:05 KST, plus the saved snapshot of its files.
- After: the new site at yunju-company.pages.dev, checked live on 15 November 2026 at 20:21 KST (before this page was added).
- Method: plain HTTPS requests with a browser user agent; JSON-LD read from the server-delivered HTML; robots.txt parsed per crawler group; every URL in the sitemap plus internal links crawled. No JavaScript rendering.
Before and after
| Check | Before (old site) | After (new site) |
|---|---|---|
| HTML pages reachable | 11 (English only) | 39, all HTTP 200 (English and Korean) |
| robots.txt | 404 (no file) | 200, text/plain |
| AI crawlers named in robots.txt | 0 | 11 named with Allow: / (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended, Applebot-Extended, CCBot, Bingbot, Yeti) |
| Sitemap line in robots.txt | none | 1 |
| sitemap.xml | 404 | 200, 35 URLs |
| /llms.txt | 404 | 200, 4,848 bytes |
| /llms-full.txt | 404 | 200, 62,228 bytes |
| Pages with JSON-LD | 0 of 11 | 35 of 39 (0 invalid blocks) |
| Distinct schema.org types | 0 | 28 (incl. Organization, ProfessionalService, Person, Service, Offer, OfferCatalog, UnitPriceSpecification, FAQPage, BlogPosting, BreadcrumbList, WebSite) |
| FAQ markup | FAQ text on the home and /faq/ pages, no FAQPage | FAQPage on 18 pages, 107 questions |
| Canonical tag | 0 pages | 35 pages |
| hreflang (en / ko / x-default) | 0 pages | 26 pages |
| Open Graph image / Twitter card | 0 / 0 pages | 35 / 35 pages |
| Meta description | 11 of 11 | 39 of 39 |
| Home page title | "Korean Proofreading, Localization & QA" (old offer) | "GEO audits, UGC-style video ads & n8n automation" (current services) |
The four pages without JSON-LD or a canonical tag are old portfolio sample files under /portfolio/samples/, kept as they were and not listed in the sitemap.
What we changed
- Crawler access: a robots.txt that names each AI search and training crawler and points to the sitemap.
- A summary for language models: llms.txt (who we are, services, prices, links) and llms-full.txt (full text of key pages).
- Entity and offer markup: Organization and ProfessionalService on every page; Service and Offer with prices on each service page; an OfferCatalog on the pricing page.
- Answer-first pages: each service page opens with a short answer and ends with an FAQ marked up as FAQPage.
- Consistency: absolute canonical URLs, English/Korean hreflang pairs, a sitemap, and Open Graph images on every sitemap page.
- Guides: four practical guides in English and Korean.
Results in AI answers: pending
On 15 November 2026, before the rebuild, we ran two buyer questions ("GEO audit agency Korea", "n8n automation freelancer Korea") through a web-search engine that writes a synthesized answer. Yunju Company was not named in either answer. We have not re-run them yet: a new site has to be crawled and indexed first. We also have no traffic, ranking, or citation numbers to report. When we re-check, we will add the date, the exact questions, the engines used, and the raw results to this page, including a "still not mentioned" result if that is what we find.
What this shows, and what it does not
It shows the technical part of a GEO audit applied end to end, measured with a repeatable script. It does not show that these changes cause AI assistants to recommend a business. Nobody can promise that, and we don't.
Frequently asked questions
Is this a client project?
No. It is our own website. We have no client GEO case studies yet.
Did the changes get you cited by ChatGPT or Perplexity?
Not measured yet. The site was just rebuilt and has to be indexed first. Results are pending and will be added with dates.
Can you run the same before/after on my site?
Yes. The GEO audit ($149, 48 hours) uses the same checks plus a buyer-question panel across AI engines.
Start: email yjs121256@gmail.com with your site (or product link), what you need and your deadline. You get a fixed written quote, usually within one Korean business day.