The Glossary Page Every SaaS Company Should Have and Almost Nobody Does

A glossary page is a set of short definitions for the words your buyers use, published on your own domain, written so each definition survives being lifted out of the page on its own. It earns its place in AI search for an unglamorous reason. Retrieval works on passages rather than whole documents, and a definition is already shaped like a passage.

Almost nobody builds one properly. The version most SaaS companies ship is a keyword dump with a table of contents, eighty terms deep, each entry padded to two hundred words because somebody read that thin content gets penalised. That page does not get quoted. The version that does get quoted looks almost too plain, and that is the point.

Why a definition is the easiest thing in the world to retrieve

Every retrieval system has to solve the same problem before it can answer anything. It has to cut your page into pieces small enough to compare against a question. Those pieces are usually a few hundred tokens, and the cut points are guessed from headings, paragraph breaks and length. Get cut badly and your best sentence arrives at the model with its subject missing.

A definition is immune to most of that. It states its subject in the first four words, resolves in one or two sentences, and does not depend on the paragraph above it. Cut it anywhere and it still reads. That property has a name in the trade, and we have written about it at length in the piece on the three stages most teams collapse into one, because being retrievable and being indexed are not the same achievement.

The other half of the effect is repetition across the corpus. When your definition of a term matches the phrasing already circulating in reference material, you are corroborating rather than contradicting. Engines reward that quietly. When your definition is idiosyncratic marketing language, you are asking the model to pick you over four sources that agree with each other.

What a real glossary looks like, using three that already work

The best glossaries in existence are not marketing assets. They were built by people who needed a shared vocabulary to do technical work, which is why they are useful models.

The NIST computer security glossary holds several thousand terms and gives each one a definition plus the specific publication it came from. Every entry is addressable, sourced and dated. It is cited constantly by people who have never heard of NIST as a brand, which is the entire mechanism in one sentence.

The UK government’s content design guidance takes the opposite approach and is equally instructive. It tells writers to use the word the reader uses, not the word the institution prefers. For a glossary that means your entry for the term is headed by what buyers type, and the internal name you use in standups appears as an alias inside the entry.

The US federal plain language guidelines supply the register. Short words. Active voice. The definition before the qualification. Read them once and you will notice how much of your existing copy is written to sound considered rather than to be understood quickly.

	The glossary that gets quoted	The glossary that gets ignored
Entry length	40 to 90 words, one idea	200+ words padded for word count
First sentence	States what the thing is	States why the thing matters to your business
Term selection	Words buyers actually say	Keywords with volume that buyers never say
Disagreement	Says when the industry disagrees and why	Presents one definition as settled fact
Product mention	Absent, or one line at the end	Woven through every entry
Update signal	Each entry carries a reviewed date	Page footer says the current year

Working out which terms your buyers actually use, rather than the ones your team prefers, means reading the raw answers engines already give about your category. Our free AI visibility check pulls those answers verbatim across four engines, so you can see the vocabulary the models already treat as standard before you write a word.

How to choose the terms, which is where most of the value is

Search volume is the wrong input. Most of the terms worth defining have almost none, because the people typing them are already deep in an evaluation and there are not many of them. That is exactly why the competition is thin.

Four sources beat a keyword tool here.

• Your sales call recordings. Every term a prospect asks you to explain is a glossary entry with demand attached. If three people in a quarter asked what you mean by a particular metric, that is a term.
• Your support tickets. Words that appear in tickets and nowhere in your documentation are gaps in the shared vocabulary between you and your customers.
• The terms competitors coined. If a rival invented a category label and it stuck, you need an honest entry for it. Refusing to define your competitor’s word does not make it go away, it just means the definition circulating is theirs.
• What engines already say. Ask four engines to define ten terms in your category. Where they hedge, contradict each other or cite nobody, there is room.

That last one is worth doing properly rather than casually. When an engine gives four different definitions for a term across four runs, no source has won it yet. When all four agree and cite the same page, you are competing against an established answer and should probably spend your effort elsewhere. The mechanics of why one source wins that position are covered in how ChatGPT decides which companies to name.

One page or many, and how to decide without agonising

The instinct is to split everything into individual pages for the traffic. The instinct is often wrong, because forty thin pages compete with each other and dilute whatever authority the set has.

Put it on the shared page when	Give it its own page when
The definition fits in under 90 words	An honest answer needs 300 words or more
It is only meaningful alongside its siblings	Someone would search this term alone at 2am
It has no buying question attached	A buying decision hangs on understanding it
You have fewer than 25 terms total	You already rank for something adjacent
It is a variant spelling or an alias	It has genuinely contested definitions worth arguing

We run about thirty terms on a shared page for most clients, with four or five promoted to standalone pieces. The promoted ones are almost always the terms where the industry genuinely disagrees, because disagreement is the only thing that justifies a thousand words.

The markup, which takes ten minutes and removes a guess

Schema does not make a page rank. What it does here is narrower and more useful. It tells a parser exactly where each definition starts and stops, which is the boundary a chunker would otherwise infer from paragraph breaks. Removing that guess is worth the ten minutes.

The vocabulary you want is DefinedTermSet with a DefinedTerm for each entry. It is one of the few types where the markup maps cleanly onto how the page is actually built, which is rarer than it sounds.

{
  "@context": "https://schema.org",
  "@type": "DefinedTermSet",
  "@id": "https://example.com/glossary/#set",
  "name": "Answer Engine Optimisation Glossary",
  "hasDefinedTerm": [
    {
      "@type": "DefinedTerm",
      "@id": "https://example.com/glossary/#retrieval-augmented-generation",
      "name": "Retrieval augmented generation",
      "alternateName": "RAG",
      "description": "A method where a model fetches passages from an external source at answer time and writes its response from those passages rather than from memory alone.",
      "inDefinedTermSet": "https://example.com/glossary/#set",
      "url": "https://example.com/glossary/#retrieval-augmented-generation"
    }
  ]
}

Three details do the work. Every term gets a stable fragment identifier so it can be linked and cited on its own. The alternateName field carries the acronym and the spelling variants, which is how you resolve the same concept arriving in four different forms. And the description is the same text a human reads, not a separate SEO version. If those two ever diverge you have built a cloaking problem for no gain. The wider question of which schema types are worth the effort at all is handled in what matters and what does not in schema for AEO.

The part everyone skips: say which terms are real

Half the vocabulary in this industry is invented by vendors. Some of it is useful, some is a rebrand of something that had a perfectly good name, and buyers cannot tell the difference. A glossary that marks the difference is immediately more useful than one that does not, and it is the single cheapest trust signal available to you.

The convention worth copying comes from standards work. RFC 2119 exists because engineers kept arguing about whether should meant the same as must, so somebody wrote down the difference and everyone pointed at it afterwards. You can do the same thing at a smaller scale by tagging each entry with where it came from.

Tag	Means	Example in this industry
Standard	Defined in a published specification anybody can read	Structured data types, HTTP status codes, robots directives
Established	No formal spec, but used consistently across independent sources	Retrieval augmented generation, reranking, entity resolution
Contested	Used widely with materially different meanings	Answer engine optimisation, share of voice in AI, brand mention
Vendor	Coined by one company, largely used by that company	Most named frameworks and most scored indexes

Measuring it without kidding yourself

Glossary pages fail conventional measurement. Sessions are low, time on page is short by design, and conversion is near zero. Judge them on those numbers and you will kill the asset in month four.

Three things are worth watching instead.

• Direct quotation. Ask the engines to define your terms and look for your phrasing coming back. Exact sentence matches are the strongest signal that your passage is the one being retrieved.
• Attribution without traffic. Being named as the source in an answer that generates no click is still the outcome you paid for. It is also the outcome your analytics will never show you.
• Downstream lift. Internal links from glossary entries to product and comparison pages tend to move those pages before the glossary itself does anything visible.

Expect nothing for six to ten weeks. Definitions have to be crawled, chunked, embedded and then encountered often enough to be treated as the consensus phrasing. That last part is corroboration and it is slow. If you want the mechanics of why some pages cross that line and others never do, the piece on treating your about page as an entity document covers the same dynamic on a page most teams also underrate.

Five ways this goes wrong

Padding entries to hit a word count. A ninety word definition that gets quoted beats a three hundred word essay that gets skipped. The padding is usually a restatement of the first sentence with more syllables, and a chunker will happily hand the model the padding instead of the definition.

Defining terms nobody asked about. If it is not in a sales call, a ticket, or an engine’s answer about your category, it is there for the keyword. Buyers can tell and so, increasingly, can the models, because nothing else on the web links those terms to your domain.

Making every entry a product pitch. One line at the end of an entry pointing to how you handle the thing is fine. Threading the pitch through the definition makes the passage unquotable, which costs you the only thing the page was built to do.

Never revisiting it. This vocabulary is moving fast. An entry written eighteen months ago that describes how a product used to work is worse than no entry, because stale and confident is the one failure mode buyers punish. Put a reviewed date on each entry and actually honour it.

Hiding it three clicks deep. A glossary linked only from the footer gets crawled late and thinly. Link it from the pages where the terms appear, in the sentence where the term first appears, which is also where a reader wants it.

What to do this week

Pull the last forty sales calls and write down every term somebody asked you to explain. Sort by frequency. Take the top thirty. Write each definition in under ninety words, first sentence states what the thing is, second sentence qualifies it, no product mention. Tag each one standard, established, contested or vendor, and be honest about which of your own words are vendor terms.

Add the DefinedTermSet markup. Link each entry from the first place that term appears elsewhere on the site. Set a calendar reminder for ninety days to check which definitions engines have started echoing back. That is the whole project, and it is a week of work that keeps paying for years, which is not something we get to say often about content.

---

Related reading

• The AEO technical playbook. Where a glossary sits relative to everything else, in the order the work should actually happen.
• Why AI Overviews cite some brands. The citation behaviour a definition page is built to win, examined on its own terms.
• What AEO actually is. The full picture if this is the first piece of ours you have landed on.