AI Is Reversing the Website Production Workflow

Overview

Traditional website production usually begins with spreadsheets, specifications, CMS design, database planning, and editorial management sheets. After that, the information is reflected into HTML or a CMS.

In other words, the usual order is: organize the information first, decide the structure, then build the website.

This time, Time Columns tested the exact opposite flow with generative AI and Codex.

The goal was not simply to create a glossary page. It was to place a glossary as an SEO asset inside the site.

By publishing glossary content and allowing search engines to index it, the site gains additional search-entry paths.

The full site expanded to 67 HTML pages. From the first glossary build to the latest additions, including implementation, validation, sitemap updates, GitHub push, and Cloudflare Pages deployment, the work took about 95 minutes.

In a traditional project, this kind of work might be split across a director, SEO specialist, editor, coder, CMS operator, reviewer, and deployment owner. That separation alone can turn a small content expansion into a multi-day process.

Here, AI compressed the workflow by processing it end to end without waiting for human handoffs between each role.

Traditional Website Production Workflow

A normal website project starts with information organization.

Teams create page lists in Excel, prepare site maps, manage drafts, define CMS fields, and design categories and tags. Then the content is reflected into HTML or a CMS.

That means the database or specification comes first, and the website comes afterward.

This workflow is still necessary, especially for large websites and stakeholder-heavy projects.

But in practice, each role creates coordination cost. The manager, director, designer, writer, engineer, infrastructure owner, website vendor, and internal reviewers all need to align.

The more roles are separated, the more time is spent waiting, checking, revising, and transferring context between processes.

Why Projects Stop Before Production

Planning matters. But many website projects spend too much time before pages actually exist.

Teams discuss how categories should be divided, where articles should live, how terms should be organized, how much should be CMS-managed, who writes drafts, and how updates will be maintained.

Those discussions are important, but if the project stays there too long, the site itself does not grow.

This is especially common in small and midsize business website renewals or owned-media projects, where the client side may not fully understand the terminology, process, or technical implications.

As a result, projects can move forward with vague goals, design-only renewals, weak SEO, difficult update flows, and operations that stop soon after launch.

For Time Columns, we changed the order. Instead of preparing a perfect spreadsheet or CMS design first, we built the website content first and then let AI organize the structure afterward.

Glossary Placement as SEO

What we built was not just a glossary expansion.

It was the placement of a glossary as an SEO structure.

The purpose was not to explain terms for its own sake. The purpose was to publish glossary content inside the site, have it indexed by search engines, and create more entry paths from search.

The source material included terms around website operations, SEO, DX, generative AI, AppSheet, cloud infrastructure, databases, security, and marketing.

Some of those terms came from past website renewal work and vendor-control situations, where a shared vocabulary was needed just to make the project conversation possible.

Traditionally, we would organize the list in Excel, define categories, write explanations, and then reflect everything into a CMS or static HTML.

This time, the flow was reversed.

We gave AI the practical term list, asked it to categorize the content as an SEO entry structure, and generated the site pages first.

After that, we extracted categories, terms, meanings, and URLs from the generated HTML and exported them back into a spreadsheet-like structure.

Build the website first
↓
Let AI understand the structure
↓
Export it back into Excel-style data

This is the opposite of the usual Excel-to-CMS-to-HTML production sequence.

Even the Meaning Research Was Delegated to AI

The glossary definitions were not researched and written one by one by a human.

The term list itself was provided by us, but AI handled the meaning organization, rewriting, and compression into short explanations.

That made sense because the goal was not to write a textbook, a legal document, or an academic paper. The goal was to create an SEO-oriented knowledge structure.

For high-risk domains, strict human review is necessary. But these were mostly practical terms around web production, SEO, DX, cloud services, marketing, access analytics, and operations.

The purpose was to give business operators and project members a quick entry point: enough context to understand what a term roughly means and how it connects to website operations.

So the priority was not long-form perfection. The priority was category structure, short definitions, sitewide knowledge clusters, related paths to columns, and clearer topical signals for search engines.

Generative AI tends to avoid extreme or overly aggressive claims. For basic glossary definitions, that tendency can actually be useful.

However, fully outsourcing term selection to AI made the output too thin and scattered. The workflow worked because the term list came from actual business practice.

Humans decided what was worth covering. AI organized, explained, and implemented it.

A Glossary Is a Search-Entry Structure

The goal of placing a glossary was not simply to define words.

The goal was to create search-entry paths by putting glossary content inside the site and allowing search engines to index it.

Terms around website operations, SEO, DX, cloud, generative AI, AppSheet, access analytics, and production workflows can each become small search entrances.

Those entrances allow people who do not yet know Time LLC to discover the site through specific terminology.

But putting every term on a single page would be weak.

If all 572 terms were placed into one page, the topic would become too broad. Search engines would have a harder time understanding what that page is about, and readers would have a harder time finding the term they need.

So the glossary was divided into category pages: website operations, SEO and search, generative AI and ChatGPT, DX and business improvement, infrastructure and networks, database design, website production projects, production tools, and more.

By splitting terms into category pages, each page can carry a clearer topic.

The categorization, page splitting, definition cleanup, HTML generation, and sitemap reflection were all handled by AI.

The human role was to provide the practical term list and decide the direction: turn it into a search-entry structure for this site.

AI Removed the Handoffs Between Processes

The most important point is not that AI wrote text.

The important point is that AI processed the workflow end to end, without requiring human handoffs between each production role.

Normally, even placing a glossary into a site involves many separated tasks: organize terms, define categories, research meanings, edit copy, decide SEO page splits, create HTML, check links, update the sitemap, publish the site, and produce a management spreadsheet.

When those tasks are split across people, waiting time, review time, rework, and context alignment all appear.

The speed came not only from making each task faster, but from removing the transfer cost between tasks.

The flow became continuous: pass the term list, categorize it as SEO structure, generate definitions, build HTML, integrate existing pages, check links and anchors, update the sitemap, export the HTML structure back to Excel-style data, push to GitHub, and deploy through Cloudflare Pages.

That is how a 572-term, 26-page glossary could be implemented, checked, and published in about 95 minutes.

This is not just faster production. It is compression of the production process itself.

Turning HTML Back Into Excel-Style Data

One of the most interesting parts of the experiment was that the completed HTML could be exported back into Excel-style structured data.

The conventional route is: create the term list in Excel, then publish it to the website.

This experiment used the opposite route.

Build the website
↓
Let AI read the HTML structure
↓
Extract category, term, meaning, and URL
↓
Rebuild it as spreadsheet-style data

In other words, the website became a structure that could later behave like a database.

Website production often assumes that clean management tables must come first. But with generative AI, a practical alternative appears: publish a usable site first, then extract and refine structure afterward.

Strict data modeling is still necessary in many cases. But for small owned media sites, glossaries, and column sites, building first and restructuring later is becoming realistic.

What Changes in Website Production

Generative AI is not only a writing tool.

In this experiment, AI handled categorization, duplicate checks, definition cleanup, wording normalization, HTML generation, existing-page integration, internal-link checks, anchor checks, sitemap updates, and Excel-style export.

Those tasks used to belong to different people and different stages.

AI does not judge everything correctly. Humans still need to decide the direction, provide the right material, and check whether the result can be published.

But a large part of the website production workflow is already being compressed.

This works especially well for structured content such as glossaries, columns, article lists, FAQs, service pages, case-study pages, and recruitment pages.

The bigger shift is that websites can become editable knowledge structures, not just final public pages.

Summary

In this experiment, Time Columns placed an SEO glossary inside the site and expanded Time Glossary to 572 terms across 26 pages.

The entire site reached 67 HTML pages at that point.

From the initial glossary build to the latest additions, the work took about 95 minutes based on commit timing.

Traditionally, this kind of project would be divided across roles and stages, often taking days because of coordination and review cycles.

With generative AI and Codex, the meaning research, categorization, definition writing, HTML generation, link checks, sitemap update, Excel-style export, GitHub push, and Cloudflare Pages deployment were processed as one continuous workflow.

That does not mean humans should throw everything at AI without direction.

This worked because the term list came from actual practice, and the human side defined the goal: use the glossary as an SEO entry structure.

Humans hold the material and judgment criteria. AI handles execution and restructuring.

That division may change website production, SEO, and knowledge management more than simple article generation ever did.

Generative AI is not just making website production faster. It is beginning to reverse the order of the workflow itself.