A definition an assistant can lift is one sentence that names the term, places it in a category, and states what separates it from everything else in that category. “A canonical URL is the URL a site designates as the preferred version of a page that is available at several addresses.” Term, category, distinguishing feature, full stop. That sentence answers “what is a canonical URL” on its own, in any context it is copied into.
This shape is old. Logicians call it definition by genus and differentia: the genus is the broader class, the differentia is what sets this member apart. It has lasted because it is the shortest form that both includes everything the term covers and excludes everything it doesn’t. Those are also the two properties a machine needs from a sentence it might quote as an answer.
This post covers the pattern, the five ways definitions usually fail, before-and-after rewrites of each, how glossary pages should be built, and how to check definitions across a site. It is a narrow case of the general answer-block method in The 40–60 word answer block: a definition is the first sentence of an answer block for a “what is” question. Both sit under Answer Engine Optimization: a page-level guide.
Why does the first sentence carry the definition?
Because the first sentence under a heading is the one most likely to be read alone. A retrieval system that splits a page into passages, or an extractor that takes the opening of a section, gets the heading and whatever follows it. If the definition is in the third sentence, the passage that gets read opens with context the reader didn’t ask for. Chunking: why passages have to stand alone covers why sections are read in isolation.
There is also a reader reason that doesn’t depend on any machine. Someone who searched “what is X” wants the answer before the history, the caveats, or the sales pitch. Leading with the definition serves that reader and the extractor with the same sentence.
So the rule is positional as well as structural: the definition goes first, directly under a heading that names the term. A question heading (“What is a canonical URL?”) is the clearest pairing. Question headings vs. keyword headings covers when to use one.
What is the pattern?
The pattern is: [Term] is a [genus] that [differentia].
- Term. The exact word or phrase, spelled the way people search for it. Don’t swap in a synonym or a pronoun.
- Genus. The nearest useful category. “A URL,” “a crawler,” “a file,” “a pricing model.” Nearest matters: “a thing” and “a concept” are categories too, and they tell the reader nothing.
- Differentia. The feature, or the few features, that separate this term from the other members of the genus. If two terms share a genus, their differentiae must differ.
Then add up to two follow-up sentences, each doing one job:
- Scope or example. One concrete instance, or what the term applies to. “For example, a product available at both
/shoes?color=redand/shoes/red/should name one of them as canonical.” - Contrast. What the term is commonly confused with, and the difference. “It is a hint to search engines, not a redirect: both URLs still load.”
Three sentences is a complete definition section. Everything after that is elaboration, and it should be written so that removing it doesn’t damage the definition.
What are the five ways definitions fail?
Definitions fail in a small number of recognisable ways. Each one makes the sentence harder to lift alone.
| Failure | What it looks like | Why it fails |
|---|---|---|
| Circular | The term, or a form of it, appears in its own definition | Assumes the reader already knows the answer |
| By example only | ”Things like X, Y, and Z” | Gives instances without the category or the boundary |
| By negation | ”It’s not A, and it’s not B” | Says what the term excludes, never what it is |
| Buried | The definition arrives after a preamble | The extracted passage opens with the preamble |
| Pronoun-led | ”It is a…” or “This refers to…” | The sentence names nothing when copied alone |
A sixth failure is a category error rather than a structural one: a genus that is wrong or too broad. “Answer engine optimization is a marketing trend” places the term in a category that tells the reader nothing about what is actually done.
Before and after
All the examples below are written for this post. The “before” versions are typical patterns, not quotes from any site.
Circular.
Before: Crawl budget is the budget a search engine allocates for crawling your site.
After: Crawl budget is the number of URLs a search engine is willing and able to fetch from a site in a given period, set by how much load the server can take and how much the engine wants the content.
The “before” repeats both words of the term and defines neither. The “after” names a genus (a number of URLs) and two differentiae (server capacity and demand).
By example only.
Before: Structured data includes things like product prices, review ratings, and event dates.
After: Structured data is machine-readable markup, usually JSON-LD, that labels the facts on a page, such as a product’s price or an event’s date, using a shared vocabulary.
The examples survive in the “after,” but behind a category and a mechanism.
By negation.
Before: A user-initiated fetcher isn’t a crawler, and it doesn’t build a search index.
After: A user-initiated fetcher is an agent that retrieves a single URL because a person asked an assistant about that page, rather than crawling the web on a schedule.
The negation is still there, as the contrast at the end, but it now follows a positive statement.
Buried.
Before: As AI search has grown, more site owners have started asking how they can control what assistants read. One answer that has been proposed is llms.txt, which is a file…
After: llms.txt is a proposed Markdown file at a site’s root that lists the pages a site owner wants language models to read, with a short summary of each.
The context sentence can follow the definition. It can’t precede it.
Pronoun-led.
Before (under the heading “Boilerplate”): This is the repeated content that appears across many pages.
After: Boilerplate is text repeated near-identically across many pages of a site, such as navigation, footers, and cookie banners.
The heading supplied the term, so the writer assumed the sentence didn’t need to. Once the sentence is read without the heading, “this” names nothing.
How long should a definition be?
As long as the genus and the differentia need, and no longer. Most terms fit in one sentence of ordinary length. If a definition needs a second clause beginning “and also,” the term may be two terms, and each deserves its own sentence.
A practical test is to read the definition sentence alone, with the heading hidden, to someone who doesn’t know the term. If they can say what category it belongs to and how to tell it from its neighbours, it works. If they ask “what’s ‘it’?”, it fails the pronoun test. If they ask “isn’t that just X?”, the differentia is missing.
GEOBee’s AEO lens has a Conciseness signal, worth 5 of 100, that rewards sentences averaging under about 25 words (the three lenses). That is a page-wide average, not a rule for one sentence, but definitions that run much longer usually contain a second idea.
How should glossary pages be built?
A glossary works best when every term is its own addressable unit, with its own heading or definition-list entry, so each one can be found and read separately. Three structural choices matter:
- One unit per term, a heading or a
<dt>with anid, so the term can be linked directly, as in/glossary/#boilerplate. Link to that anchor wherever the term is used elsewhere on the site. - The definition first under each heading, in the pattern above. A glossary is the place where every section is a “what is” answer, so every section should open with one.
- Separate pages for terms that need more than a paragraph. If a term needs examples, a table, and caveats, it is an article. Give it a page and keep a one-sentence definition in the glossary that links to it.
Schema.org has types for this: DefinedTerm, “a word, name, acronym, phrase, etc. with a formal definition,” and DefinedTermSet for the collection. They are optional. Google’s structured data search gallery doesn’t list a feature for them, and Google’s guidance on AI features says no special schema is needed to appear in them. Use them if they describe your page accurately. The visible sentence is still what gets read. This site’s own glossary uses a definition list with one anchored entry per term.
How to check definitions across a site
By hand, the check is quick: for each “what is” heading, read the first sentence alone and test it against the five failures. For a site with hundreds of definition sections, a crawler and a spreadsheet do the reading.
With crawler custom extraction. Add an XPath rule that pulls the first paragraph after each question heading that asks for a definition:
//h2[starts-with(normalize-space(.), 'What is')]/following-sibling::p[1]Add a second rule that pulls the heading text itself: //h2[starts-with(normalize-space(.), 'What is')]. Configure both to return all matches, crawl, and export to a spreadsheet with one row per heading and paragraph pair.
Then flag the failures with formulas. With the heading in column A and the first paragraph in column B:
- Pronoun-led:
=IF(REGEXMATCH(B2, "^(It|This|These|They|That) "), "pronoun", "") - Missing term: extract the term from the heading, then
=IF(ISNUMBER(SEARCH(C2, B2)), "", "term missing"), where column C holds the term. - Long:
=IF(LEN(B2) - LEN(SUBSTITUTE(B2, " ", "")) + 1 > 40, "long", "").
REGEXMATCH is a Google Sheets function. In Excel, use LEFT checks for each pronoun instead. Circularity and wrong genus need a person to read, but the formulas cut the list down to the rows worth reading.
For a single section, paste the heading and its opening into the answer block checker, which runs seven answer-block checks on the first paragraph under a heading, including whether the first sentence names the subject of the heading and whether it opens on a pronoun.
GEOBee’s AEO lens scores the closest page-level property, the Answer block signal worth 20 of 100: a concise direct answer near the top of the page. Its Answer column reports each page as Yes, Partial, or No. It doesn’t parse definitions individually, so the extraction above is still the way to check every term on a glossary.
What a good definition buys you
A definition in the pattern above is a sentence that can be quoted without context and still be correct. That is the property that makes a passage liftable, and What makes a page citable covers the other properties that sit next to it. It doesn’t make any particular assistant choose your sentence. It removes the reasons a reader or a machine would have to look elsewhere for one.
To find the pages on your site with no direct answer near the top, download GEOBee and check the Answer column across your crawl during the 7-day trial.