Should we build a glossary so AI assistants use us as the reference?

Every sector has a vocabulary that buyers only half understand. Your clients ask what a schedule of condition is, what separates a lease assignment from a sublet, whether a managed service is the same as outsourcing. Your people answer those questions every week, and the answers are nowhere on your site. When a buyer asks ChatGPT, Perplexity or Google instead, the assistant explains the term, and when it searches the web to do so, it cites whoever explained it best. Usually that is a large firm, a trade body or a regulator. Sometimes it could be you.

Definition questions are where AI answers appear most. In a US study by Pew Research Center published in July 2025, 60% of Google searches that began with a question word such as "what" or "how" produced an AI summary, against 8% of one- or two-word searches.

A glossary is the simplest way to answer those questions on your own site. It also has limits most advice skips: assistants define common terms from memory without citing anyone, hundreds of one-paragraph pages can do more harm than good, and being cited for a definition is a long way from being recommended for the work.

#Do AI assistants actually use glossary definitions?

Sometimes. For a common term, ChatGPT or Claude will usually answer from what the model learned in training, and in our experience often cite nobody. They search and cite more when the term is specialist, UK-specific or recently changed, or when it comes with a practical question such as "do I need one?".

That second group is where a glossary earns its keep. Google's AI Overviews and AI Mode also split a question into related searches, which Google calls query fan-out, and cite pages that answer the parts. Ahrefs found in March 2026 that only 38% of AI Overview citations came from pages in Google's top ten, down from 76% in July 2025. So a precise entry can be cited without ranking first for the main query, though ranking still helps.

Reputation matters too. In an SE Ranking study reported by Search Engine Journal in November 2025, the number of different websites linking to a domain was the strongest predictor of ChatGPT citations. A small firm's definitions can be cited, and are far more likely to be when trade bodies, partners and the trade press link to them.

#Who owns this definition today?

Check before you write. For each term, ask ChatGPT (with search on), Perplexity and Google AI Mode "What is [term]?" in a clean session, and note who each one cites. Answers vary from run to run, so ask at least twice and record what repeats.

Most lists shrink once the officially owned terms come off.

#Do glossary pages help SEO or count as thin content?

They help when each entry has substance, and count as thin when it does not. A one-paragraph definition that stops there offers nothing the assistant could not say itself, and the method in our guide to auditing existing content would score it as shallow.

Google's guide to its generative AI features, updated in July 2026, says there is "no requirement to break your content into tiny pieces for AI to better understand it". It adds that creating separate content for every variation of how people search, mainly to manipulate rankings, breaks its scaled content abuse policy. A glossary of 150 near-empty pages is that pattern.

So test every term: can you write the full entry below, with an example and a real distinction, without padding? If yes, it earns a page. If not, it goes on a hub page with its neighbours. Why ChatGPT ignores our website when we have hundreds of pages explains the wider risk.

#Should each glossary term have its own URL?

Only the terms that pass that test.

For example, a commercial property firm might give "schedule of condition" and "dilapidations" their own pages and put "rent deposit" and "service charge cap" on a lease terms hub.

#What does a good entry look like?

In our experience an entry that gets drawn on has seven parts, and we now add an eighth.

• The term as its title, as people say it, with the acronym if there is one.
• A one-line definition that could stand alone in an assistant's answer. Put it first: in Kevin Indig's analysis of 1.2 million ChatGPT answers, reported by Search Engine Land in February 2026, 44% of citations came from the first 30% of a page.
• A plain explanation: what it is, why it exists, how it works. As long as the term needs.
• How it differs from the neighbouring terms buyers confuse it with.
• An example from your sector, labelled as one.
• Related terms, linked.
• A named author, a date and a link to the authoritative source, such as the statute or professional body.
• When you need a specialist: the point at which a buyer should get help, linked to the matching service or comparison page.

Take a firm of commercial property solicitors writing up "schedule of condition". The one-liner: a record of a property's state at the start of a lease, usually photographs and a written description, used to limit the tenant's repair obligations at the end. Then why tenants want one, how it is prepared and attached to the lease, and what happens without one. Then an example: a tenant taking a five-year lease of an older unit who, without a schedule, could have to return it in better condition than they found it. Then links to "dilapidations", "full repairing and insuring lease" and "break clause", the partner's name and the date. Finally the specialist paragraph: taking a lease on an older building, have the schedule agreed before you sign, with a link to the firm's lease work.

What an article that ChatGPT actually quotes looks like covers the same structure for longer pages.

#How do I get ChatGPT to use my firm's definition of a term?

You cannot make it, but you can make yours the clearest, most current and most referenced version it finds. Assistants that search the web live can pick up a new page within weeks; answers from a model's memory change only when it is retrained, which takes months.

1. Say what generic definitions do not: the UK rule, the practical threshold, the mistake buyers make.
2. Use one wording everywhere. Two versions of a definition on your site give an assistant a reason to trust neither. Our rebuild for arum followed that rule: "Proper structure and markup on every page, a plain summary AI assistants can read, facts stored once so two pages never disagree, and every old address redirected so years of rankings carry over."
3. Cite the authority. In a Princeton and IIT Delhi study presented at KDD 2024, citing sources raised a page's visibility in AI-generated answers by about 30%. It was tested on a research engine, so treat it as directional.
4. Make it readable. ChatGPT search needs OAI-SearchBot allowed, and GPTBot, ClaudeBot and PerplexityBot do not run JavaScript, so keep the definition in the HTML. See how AI crawlers read your website.
5. Get it referenced by trade bodies, partners and journalists who need the definition.

#Does a glossary bring in enquiries or just traffic?

Mostly citations and early-stage visits. A buyer asking what a term means is still working out the problem, and definition prompts rarely produce a supplier recommendation.

Being understood is not the same as being recommended either. At the start of our work with ID-Pal, "AI could describe ID-Pal. It never recommended it." "Across 36 recorded AI answers to the questions ID-Pal's buyers actually ask, ID-Pal was named three times, and only for its brand or its Salesforce integration. AI knew who it was, and had no reason to recommend it." Part of the fix was pages that answer buyers' questions directly, and the result is dated: "UK buyer questions where AI names ID-Pal: none in May, two in August, four in September 2026."

So judge glossary pages by the visits they pass to service pages as well as by their own traffic.

#What schema markup should a glossary page use?

DefinedTerm for each term and DefinedTermSet for the glossary, both core schema.org types. Google shows no rich result for either: neither appears in Google's list of the structured data it supports, updated in June 2026, and Google says its AI features need no special markup. OpenAI and Anthropic have not said whether they read it. Add it for accuracy, and do not expect it to lift rankings.

On a hub page, give each term its own DefinedTerm with an @id pointing to its anchor. Keep the description word for word the same as the visible definition. More in do we need schema markup for AI assistants to understand our site?

#How do we build the glossary?

#Step 1: List the terms buyers ask about

Ask your fee-earners, sales team and support inbox which terms clients ask them to explain, and add the terms in your proposals and engagement letters. Building content around the questions customers ask is the method.

#Step 2: Test, rank and decide page or hub

Run the ownership test, then rank what is left. Tier one is the 25 to 40 terms every buyer meets early, tier two the terms that matter once a project is under way, tier three the specialist ones. Publish tier one first, over two to three months.

#Step 3: Write each entry to the eight-part shape

Give each term to the person who explains it best in person, and have a writer shape their explanation. Never define a term using other undefined terms.

#Step 4: Link it into everything else

• Link each glossary term once per article, at its first use, with the term itself as the anchor text.
• Add a "Terms used on this page" box to long guides.
• End every entry with a link to the deeper guide or service page on its subject.
• Restate definitions elsewhere in the glossary's exact wording.

How to organise our content so AI assistants see us as the expert covers the site-level pattern.

#Step 5: Maintain it on a calendar

Add terms as buyers raise them, review every entry yearly and update at once when a rule changes, with a visible date and a line on what changed. Seer Interactive found in July 2026 that 75% of pages cited by AI assistants had been updated in the previous 12 months, in four consumer and finance industries. See whether ChatGPT prefers newer content.

#How can we tell whether it is working?

Keep a fixed list of 20 to 50 questions: your published definitions plus the buyer questions they should lead to. Ask them in ChatGPT, Perplexity and Google AI Mode several times a month in a clean session, and count a win only when it repeats. Search Console's generative AI reports, live for all sites since 31 August 2026, show your impressions in AI Overviews and AI Mode. How to find out whether ChatGPT is recommending your firm has the full method.

#What to do this month

1. Ask five colleagues who deal with clients for the ten terms they explain most often, and combine the lists.
2. Run the ownership test on your top 40 terms and strike out the ones an official body owns.
3. Decide for each remaining term whether it earns its own page or belongs on a hub.
4. Write the first five entries to the eight-part shape and publish them in a proper glossary section.
5. Link those terms from your existing guides at first use, and add a "Terms used on this page" box to your three longest guides.
6. Put a quarterly review in the diary.

#Frequently asked questions

#How many terms do we need?

Start with 25 to 40 candidates and let the ownership test cut the list. Then grow as buyers raise new terms, with no target count of pages. Forty precise, current, well-linked entries do more than three hundred thin ones.

#Should acronyms have their own entries?

No. Put the acronym in the title of the full term, such as "Full repairing and insuring lease (FRI)", and let one page answer both. Separate entries split the value and create duplicates that assistants have to choose between.

#How is a glossary different from an FAQ page?

An FAQ page answers scattered questions in one place. A glossary gives each term its own page or anchor, so an assistant can match a specific question to a specific passage. Google stopped showing FAQ rich results on 7 May 2026, so neither format earns a special display; both are judged on their answers.

#Does the glossary need to be public?

Yes. An assistant cannot draw on what it cannot read, and neither can Google. Definitions behind a login or in a PDF download do nothing for being named. Keep them on public pages, and check that robots.txt and your CDN let the AI search crawlers in.

#We are a small firm. Will anyone take our definitions seriously?

Yes, if they are better and other sites point to them. Links from other websites were the strongest predictor of ChatGPT citations in the SE Ranking study, so offer your definitions to trade bodies, partners and journalists. Small firms do best on specialist, UK-specific terms that large sites cover vaguely.

If you want to know which terms in your field are still unowned, and whether assistants describe your firm without recommending it, book a free Growth Review: 30 minutes with David Adams and three fixes in writing. Expert content that gets a firm named is part of Reputation That Wins.

#Sources