WORK · 04
Rebuilding the AIO2O Company Website as an Official Source AI Answer Engines Can Read and Cite
- Client
- AIO2O
- Role
- Planning, information architecture, solo build (directing AI coding agents), migration design
- Period
- 2026.09 ~ 10
- Website
- o2osolution.ai ↗
- 40 Korean · 35 English, every page paired
- 75 pages
- legacy URLs kept alive (permanent redirects)
- 38
- customer cases with public sources
- 14
- from first commit to production
- ~3 weeks
Challenge
The old website had no structured information for AI answer engines to read, and company facts such as the name, founding date, and product names differed across press releases, decks, and the site itself. In a time when customers and investors ask AI about a company, the company’s own official facts had to exist accurately in one place. This was not a design refresh. It was rebuilding the site as an official source that AI can read and cite accurately.
Approach
1. One sourced record of facts. Company facts live in one data file, and each fact carries a public source such as a press release or a public database. Every sentence on screen comes from that record. Unsourced numbers are blocked at build time.
2. A structure AI can read. Every page carries Organization, FAQ, and BreadcrumbList structured data, the site publishes llms.txt, and robots.txt allows AI crawlers including GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and Naver’s Yeti. Korean and English pages are paired with hreflang.
3. A publishing gate. The build checks for banned expressions, non-public numbers, pending-verification marks, broken links, and missing redirect targets, and stops if any one fails. Wrong facts cannot ship even without a manual review every time.
4. The structure. 40 Korean and 35 English pages, 75 in total: solutions, 14 customer cases with public sources, a newsroom that publishes only approved articles automatically, insight guides and a glossary, videos, careers, and the privacy policy and terms. Everything the company should answer for both AI and people.
5. A switch with no downtime. 38 legacy URLs redirect permanently to new pages with the same content. Other services under the same domain were untouched; only the root domain and www changed. The contact form went live on the new address before the switch, and sitemaps were submitted to Google Search Console and Naver Search Advisor the same day.
Result
About three weeks from first commit to production. The company’s official facts now sit in one place with sources, and AI answer engines and search engines read the same facts. No legacy URL and no contact form broke. The same approach carried straight into the WebforAI landing that followed.
The site in production
Open in a new tab ↗This is the real site, not an image. Scroll inside to see the full structure: company, solutions, customer cases, newsroom, and insights.
Frequently asked questions
Why rebuild the website?
Customers and investors had started asking AI about the company, but the old site had no structured information for AI to read, and the company name, founding date, and product names differed across documents. This was not a redesign. It was putting the company's official facts in one place that AI can read and cite accurately.
How do you keep the facts accurate?
Company facts live in one data file, and every fact carries a public source such as a press release or a public database. If a number has no source or a pending-verification mark remains, the build stops and nothing ships. Every sentence on screen comes from that record.
How did you switch a live site without downtime?
38 legacy URLs redirect permanently to new pages with the same content. Other services under the same domain were left untouched; only the root domain and www changed. The contact form was opened on the new address before the switch, and sitemaps were submitted to the search engines the same day.