Overview
Traditional website production usually begins with a site map, page list, editorial management sheet, CMS design, and category structure. The information is organized first, and then reflected into the website. That order is still useful, especially for large sites and stakeholder-heavy projects.
But with generative AI and Codex, part of that order is beginning to change. Time Columns tested this by building Time Glossary as an SEO entry structure. The experiment organized 572 terms into 26 glossary pages, generated HTML, checked internal links, updated the sitemap, and pushed the result to GitHub.
The interesting part was not only the speed. We built the web pages first, then let AI read the HTML and extract categories, terms, meanings, and URLs back into Excel-style data. The conventional route moves from Excel or specifications to the website. This experiment moved from the website back to Excel-style structure.
Traditional Website Production Workflow
Until now, many website projects treated the website as the output of already-organized information. Teams first designed page lists, site maps, categories, editorial sheets, and CMS fields, then moved into design, writing, coding, CMS registration, and publishing.
That order exists for a reason. If the site structure is not decided early, information placement becomes vague and future updates become harder. In projects with multiple people, the team also needs a shared understanding of responsibility and review flow.
So planning is not wasted work. It is still necessary. The problem is that the planning stage can become too heavy. A project may have spreadsheets and structure diagrams, while the pages that readers and search engines can actually reach do not exist yet.
This Time, We Built the Website First
For Time Glossary, we changed the order. We started with a practical term list that had already appeared in real website operations, SEO, DX, cloud, database, security, and AI work. AI categorized those terms as an SEO entry structure, wrote short explanations, generated category pages, integrated links, and reflected the pages in the sitemap.
At that point, there was no completed Excel management sheet placed before production. The website came first. After the HTML existed, AI read the created pages and extracted categories, terms, meanings, and URLs back into Excel-style structured data.
In other words, the public website existed first, and the management data came afterward. Instead of preparing a management table in order to build the website, we built the website and then let AI create management data from it.
Workflow Comparison
The difference is easier to see as a workflow comparison. The conventional route starts with management tables and specifications, then reflects them into the website. This experiment created usable HTML first, then let AI read the structure and turn it back into management data.
| Perspective | Traditional flow | This experiment |
|---|---|---|
| Starting point | Excel, specifications, CMS design | A practical term list |
| First output | Management table or site map | Publishable HTML pages |
| When structure is refined | Before production | After production, with AI |
| Main bottleneck | Review, rework, and handoff waiting time | Checking AI output and publication judgment |
| Relationship with spreadsheets | Spreadsheet to website | Website back to spreadsheet-style data |
This does not mean every site should be built this way. But for small owned-media sites, glossaries, FAQs, and column sites, creating the visible structure first and refining it afterward can be more practical than waiting for a perfect management sheet.
AI Compressed the Waiting Time Between Processes
Generative AI did not only make writing faster. What mattered here was the reduction of handoffs between production steps.
Even a glossary normally involves several separated tasks: organizing terms, researching meaning, designing categories, writing short definitions, generating pages, checking links, updating the sitemap, publishing, and producing management data. When those tasks are split across people or tools, review time, rework, and context transfer appear between each step.
AI made the workflow more continuous. Human review remained necessary, and AI output still had to be checked before publication. But the work did not stop between every small process. That is why this is not just time saving. It is compression of the production workflow itself.
HTML Can Be Structured Afterward
In the conventional view, HTML is mainly a display layer. Information is managed in Excel, a CMS, or a database, and HTML is the final form shown to the reader.
With generative AI, HTML can also become a source of structured data. Headings, body text, links, URLs, categories, and descriptions are not only for human readers. They are also material that AI can read and reorganize.
That means a site can be created first, then turned back into category lists, term lists, internal-link candidates, and maintenance data. The website is no longer only a published surface. It can become a knowledge structure that is reorganized after publication.
This Was Not Full Delegation to AI
The experiment worked not because everything was delegated blindly to AI. AI handled classification, short definitions, HTML generation, internal-link checks, sitemap reflection, and the export back into spreadsheet-style data.
The human side still held the important decisions: which terms were worth covering, why the glossary should exist, how it should work as an SEO entry structure, and whether it fit the broader Time Columns context. If that part is left entirely to AI, the output can become thin, scattered, or disconnected from the business.
The practical division was simple: humans decide the material and direction; AI handles organization, generation, conversion, and restructuring. At this stage, that division feels more realistic than full automation.
Summary
Generative AI is beginning to change website production from a single fixed order into multiple possible orders. In this experiment, Time Columns built Time Glossary as web pages first, then extracted categories, terms, meanings, and URLs from the HTML back into Excel-style data.
The traditional flow goes from Excel or specifications to the website. This experiment went from the website back to structured management data. That is not only a speed improvement. It means the website itself can become a knowledge structure that AI can read, reorganize, and maintain.
Large websites and projects that require strict data modeling will still need careful planning before production. But for smaller owned-media sites, glossaries, FAQs, and column sites, building first and restructuring afterward is becoming realistic. Generative AI is not only making website production faster. It is beginning to reverse part of the workflow itself.