Taxonomy Expansion

Udemy's taxonomy currently operates as a controlled vocabulary (it's a controlled list of approved terms) instead of a true taxonomy which includes hierarchical relationships for broader and narrower terms.

For example, it currently uses parenthetical notes to narrow focus instead of using hierarchy to deliver the level of detail needed:

• Design Patterns (software)
• Google Gemini (Bard)
• Microsoft Entra ID (Azure Active Directory)

It also doesn't discern between skills and tools, delivering the same experience to both. In reality, tool searches (e.g. python course or best excel tutorial) are transactional and skill searches are typically informational and career-oriented (“how to learn data analysis”) because the lack of a tool indicates a lack of fundamental understanding of the subject and available solutions.

There are two global standard taxonomies related to learning: ESCO and O*NET. Both have a similar hierarchical structure:

1. Occupations (Frontend Web Engineer, Full-Stack Web Engineer, Senior Data Scientist, etc.)
2. Skills/competencies (Data Analysis, Machine Learning, Project Management, etc.)
3. Knowledge domains (Statistics, Object-Oriented Programming, etc.)
4. Tools/technology (Python, Excel, Tableau, Docker, etc.)
5. Credentials (PMP, CompTIA, etc.)

Strict containment. It's worth noting that subjects can't have strict containment in the hierarchy, as a topic like Python can be relevant to multiple skills/competencies like Data Science and DevOps.

We see this most commonly manifest as pages with giant organic search impression volume but very low click-through rates, and high-ranking pages with little click-through traffic. Broad topics need to be broken into smaller, more-targeted subjects. For example, the “Business Analysis” topic page surfaces for general informational queries (“what is business analysis”) that the page, as a simple course directory, cannot answer.

Related skills, relevant occupations, and career context can answer those queries, while more-direct pages can divert better-targeted traffic. Hierarchy can also aid in vertical navigation among related subjects.

The initial scope of this project is AI-adjacent topics: the 99 topics categorized as AI-type in the Udemy taxonomy, plus any AI-relevant occupations whose Representative Topics currently lack /topic/ pages. This scope was chosen because AI-adjacent pages are among the highest-traffic and fastest-growing in the Udemy topic catalog, and because the AI learning market is highly competitive with learners actively searching for career and skill guidance.

An expanded taxonomy, with a complete Skills Graph for AI-adjacent topics, new occupation landing pages, skill cluster pages, and hierarchy, will allow Udemy to better target niche, occupation-level, and career-intent searches that /topic/ pages cannot capture. This will mitigate zero-click losses from broad informational queries (where course listings satisfy no intent), and increase on-page engagement from the traffic these new surfaces attract, as learners will find more relevant, career-contextualized content than a plain course listing provides.

This work can be divided into two workstreams: updating the taxonomy and the website implementation.

The Skills Graph currently maps occupations to related topics with affinity scores, but coverage is incomplete. For the website's new occupation and skill-cluster pages to surface accurate content, the Skills Graph must be complete for the ~99 AI-adjacent topics.

We must then define the set of occupations adjacent to AI subjects and ensure each selected occupation has a Representative Topic and valid Skills Graph entry.

Then, we need to define skill clusters for cluster pages:

• Propose 3–6 AI-adjacent skill clusters (e.g., “AI Engineering,” “Data Science,” “AI in Business,” “Generative AI Tools”)
• For each cluster, define the member topics (from the taxonomy) and their priority ordering
• Ensure cluster labels are distinct from existing topic labels to avoid content duplication

Last, we need to audit for conflicts with existing one-off pages and determine whether the one-off pages are worth keeping or replacing.

• Occupation Landing Pages. Create new /learn/[occupation-slug] pages (or equivalent path TBD with engineering) for AI-adjacent occupations.
  • Occupation name, description, and career context
  • Top 6–8 skills for this occupation, drawn from Skills Graph (each linking to its /topic/ page)
  • Relevant certifications for this occupation, drawn from Skills Graph and taxonomy cert data
  • A curated course feed filtered to this occupation's Representative Topic and top related topics
  • Internal links to related occupation pages and skill-cluster pages
• Skill-Cluster Pages. Create new /learn/[cluster-slug] or /skills/[cluster-slug] pages for each AI skill cluster.
  • Cluster name and description (what this set of skills is and why it matters)
  • The 6–8 member topics as a navigable list, each with a short description and link to its /topic/ page
  • Occupations that rely on this skill cluster (cross-links to occupation pages)
  • A combined course feed across all member topics
• Certification Hub Pages. Create certification hub pages that group AI-adjacent certification topics by occupation. For example, a “Machine Learning Engineer Certifications” page would aggregate TensorFlow Developer Certificate, AWS Certified Machine Learning, and similar.
  • Overview of certification landscape for this occupation
  • Individual certification cards, each linking to the corresponding /topic/ page
  • Recommended learning sequence (aspiring vs. advancing, drawing from Occupation Stage taxonomy data)
• Page Conflicts. For any existing one-off career or certification pages:
  • Apply noindex to existing pages where the taxonomy page is the preferred canonical destination
  • Implement 301 redirects where the existing page URL has meaningful inbound link equity
  • Do not create a taxonomy page where a conflict exists and cannot be resolved without clear owner sign-off

Phase 1: AI-Adjacent Topics. Launch is scoped to AI-adjacent topics and occupations. This allows the team to validate the taxonomy pipeline, page template quality, and SEO performance before committing engineering and taxonomy resources to the full catalog.

Phase 2: Full Taxonomy Rollout. Once Phase 1 demonstrates meaningful organic traffic and GMV lift, the approach scales to all occupations and the full set of topic categories.