A taxonomy is a top-down hierarchical classification scheme of preferred terms, dividing general concepts into progressively more specific ones through parent-child relationships. Three types of logical relationship can connect terms within a taxonomy: broader/narrower, the most common: "a narrower term is a kind of the broader term," like "oak is a tree"; instance-of, where a general category has specific, often proper-named instances, like "Zoom is an instance of an online-meeting application"; and whole/part, like "a page is part of a website."
Most taxonomies place each term in exactly one location, a strict hierarchy, but a polyhierarchy allows a term to appear in more than one place when a concept genuinely belongs in multiple categories, like a piano being both a percussion instrument and a stringed instrument. Polyhierarchy improves findability but is harder to maintain, so it's used sparingly rather than by default. Interestingly, the categories people find easiest to work with aren't at the very top or bottom of a taxonomy. Mid-level categories, concrete enough to picture but general enough to be widely understood, are the most cognitively natural starting point for most people.
"Animal" is too abstract to picture. "Himalayan cat" requires knowledge most people don't have. "Cat" sits in the middle, and that middle is where most people's natural thinking actually lives.
1: A broader/narrower relationship has been misapplied. "A podcast is a kind of article" doesn't hold up, podcasts and articles are two different content types that happen to sit side by side, not one nested inside the other.
2: Should be modeled as an instance-of relationship. "The New York Times" is a specific, named entity, an instance of the general category "News Sources", which is a different logical relationship from broader/narrower and should be modeled that way to keep the taxonomy's relationships meaningful.
3: Inconsistent polyhierarchy. Placing content in two categories isn't wrong on its own, polyhierarchy is a legitimate, if sparingly used, technique, but doing it without a consistent rule for when it applies is what creates the actual problem, since other similar articles may not receive the same dual placement.
The correct fix for problem 1 is placing "Podcasts" as a sibling category to "Articles", both narrower terms under a shared broader category like "Content Types", rather than nesting one inside the other. This preserves the taxonomy's logical integrity: each level should only contain terms that are genuinely kinds of the level above them, and two unrelated content formats sitting at the same level, side by side, reflects their actual relationship far more accurately than forcing one to be a subtype of the other.Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Casts the AI as: You're an information architect specializing in software documentation taxonomies.
Spells out exactly what to analyze and return.
Target audience, Current problem and Business goal tell the AI what your specific situation is, not a generic one.
Output as markdown indented list with: top-level categories, one level of children, and a short scope note under each top-level label.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Casts the AI as: You're an information architect exploring organizational approaches.
Names 3 specific outputs to produce, so the response comes back structured rather than a general summary.
Content, Audience, Content areas and Goals tell the AI what your specific situation is, not a generic one.
For each taxonomy option: the organizing principle, a sample top level, strengths, risks, and when you would choose it over the others.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Casts the AI as: You're an information architect who specializes in the Diátaxis framework.
Asks for a taxonomy that integrates all four Diátaxis types, using the structural approach you choose (Diátaxis-first, feature-first, or hybrid).
Content and Features/Topics tell the AI what your specific situation is, not a generic one.
Output as indented list with: Diátaxis types as top level (or facets), topic categories beneath, and a note on where hybrid pages should live.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Casts the AI as: You are an information architect validating taxonomy balance and quality.
Names 4 specific outputs to produce, so the response comes back structured rather than a general summary.
Total pages, Target audience and Content domain tell the AI what your specific situation is, not a generic one.
Provide: a pass/fail per check, the specific nodes that fail, and a concrete rewrite for each failing label or relationship.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Use this prompt when: Creating primary taxonomy from scratch.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Casts the AI as: You're an information architect refining an existing taxonomy.
Names 3 specific outputs to produce, so the response comes back structured rather than a general summary.
Total content, Audience and Business goal tell the AI what your specific situation is, not a generic one.
Output: