> ## Content Index > Fetch the complete content index at: https://www.datadab.com/blog/llms.txt > Use this file to discover other available public pages before exploring further. # Ask the vendor to define it: eighteen terms that mean nothing until they do - URL: https://www.datadab.com/blog/ask-the-vendor-to-define-it/ - Published: 2026-11-06T04:30:00.000Z - Updated: 2026-11-06T04:30:00.000Z - Description: A citation, a mention and a share are three different numbers. Make a vendor say which one is on your dashboard before you read the price. - Author: Amit Ashwini - Tags: GEO *The category named itself four ways in two years, and the measurement words are worse. Half of what a demo costs is arguing about vocabulary.* I sit in on vendor calls for clients, and the same twelve minutes disappear every time. A tool says it improves "AI visibility". The vendor does not define it. The room agrees to a number. So we wrote [a glossary of the terms this category actually uses](https://www.datadab.com/research/ai-visibility-glossary), eighteen entries across four groups: the engines, the optimisation methods, the measurement vocabulary, the strategy vocabulary. This post is not a summary of it. It is the argument for why the glossary exists, and the four words I now make a vendor define before I will look at a price. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-01-hero-1.png) ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-02-the-category-has-four-names-and--1.png) ## The category has four names and no owner AEO, answer engine optimization. GEO, generative engine optimization. LLMO. "AI visibility". These are not the same discipline with different labels. They imply different objects: an engine to answer, a generator to influence, a model to feed. A vendor's choice of term is usually a claim about what they can move, made without saying it. Our guidance for a leadership briefing is deliberately boring: use AEO as the category term, because it is the most recognised in boardrooms in 2026, and use citation share as the headline metric, because it behaves like organic share of voice and your CFO already knows how to read that. For the evaluation itself, do the opposite of boring. Ignore the tool's own branding terms, the "AI visibility suite" language, and read the methods section instead. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-03-citation-is-doing-the-work-of-fo-1.png) ## "Citation" is doing the work of four words The measurement group has four separate entries that vendors collapse into one: AI citation, citation share, brand mention, mention prominence. And a fifth, brand representation, that most platforms treat as an afterthought. A mention is presence. A citation is a source attribution. Prominence is where you were placed in the answer. Share is your slice of a category's total. Ask which one is on the dashboard, and the room changes. A tool reporting "mentions" while your board hears "citations" is running a risk register you have not agreed to. Two platforms in this market use related but proprietary scoring for the same-looking number, which means their scores are not comparable and their demo slides are not evidence. Brand representation is the one I ask about last and care about most. Being cited in a wrong sentence is not a win, and most dashboards count it as one. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-04-rag-grounding-and-passage-retrie-1.png) ## RAG, grounding and passage retrieval are not synonyms Three engine-group terms, three different failures. Retrieval-augmented generation describes a system that fetches text before answering. LLM grounding describes whether the answer stays attached to what was fetched. Passage retrieval describes the specific act of finding the 100-300 word window that answers a question. They fail differently. A page can be reachable and unretrievable, retrieved and ungrounded, or grounded in a sentence you did not write. The reason to care is that each failure has a different fix. A vendor who uses the three terms interchangeably has one playbook, and it is usually the second one. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-05-extractability-is-the-word-that--1.png) ## Extractability is the word that ends an argument AI extractability means whether a machine can pull the meaning out of your page: structure, tables, dates, answers near the top. Accessibility means whether a screen reader or a crawler can reach it. People say "our site is AI-accessible" when they mean robots.txt allows a bot. That is the weaker claim. Allowing the crawl and surviving the parse are separate. The pricing vocabulary is where extractability meets money, and it is worth the second source here: our [pricing and entitlements study](https://www.datadab.com/research/ai-marketing-software-pricing) shows two plans in the same comparison column at $0.67 and $0.02 per engine-query. The difference was not quality. It was that one vendor's word "prompt" meant a query and the other's meant a query per engine. Vocabulary is not pedantry when the unit is undefined. It is the invoice. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-06-the-three-questions-that-close-a-1.png) ## The three questions that close a demo Our own guidance for anyone buying a tool in this category reduces to three, and I have not improved them since writing them. \* *Which prompt volumes are you indexing against?*\* A claim about visibility is a claim about a sample. Ask what the sample is, how many prompts, which engines, refreshed when. \* *How do you score mention prominence rather than just mention?*\* If the answer is "we count appearances", you are buying a count and calling it a ranking. \* *How does your implementation arm work, or do you ship nothing and leave the rewrite to me?*\* Both answers are fine. A vague answer means you will discover the second one at renewal. Two of those three questions cannot be answered by a dashboard, which is why they work. They force a sentence about method, and method is where the differences between vendors actually live. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-07-what-to-do-with-the-words-on-you-1.png) ## What to do with the words on your own pages The strategy group exists because teams get the vocabulary right in the demo and wrong on the site. Lead with the answer, one paragraph, one answer, so a passage retriever has something concrete to take. Use comparison tables and FAQ blocks, which are the structures answer engines parse most reliably. Add the structured data that describes the entity and the article. Refresh dated pages, because a stale date reads as a stale claim. Then hold your own pages to the same definitional discipline you are imposing on vendors. If your pricing page says "AI features" and your docs say "assistant" and your comparison page says "visibility tools", you have three terms for one thing and an engine has to guess. An entity that cannot be resolved from your own site will not be resolved by the web either. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-08-three-terms-decide-whether-you-a-1.png) ## Three terms decide whether you are being sold \* *Mention prominence*\* is where you appear in the answer: first, second or fifth in a recommendation list, named alongside competitors or after them. The first named brand typically draws a disproportionate share of the buyer's attention downstream, so a tool that counts appearances without counting order is reporting the cheapest possible signal. \* *Brand representation*\* is how accurately you are described, which is a separate failure from being absent. A model can name you consistently and describe you wrongly every time, as the enterprise tool for large retailers when you are a mid-market tool for something else. No dashboard I have seen makes that a headline metric, and it is the one that keeps founders awake. \* *Topical authority*\* is cumulative credibility on a specific topic, built by depth over months. Sites that publish one category-defining resource get cited in that category; sites that publish across everything get cited for nothing in particular, which is the same sentence your SEO lead has been saying for a decade with a new noun. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-09-two-of-the-eighteen-terms-are-ou-1.png) ## Two of the eighteen terms are ours, and that is worth saying AI extractability and decision-stage content are DataDab coinages. I would rather flag that than let a reader discover it in a footnote. Extractability, as we define it, is the degree to which an engine can pull structured, citable facts off a page: a single-paragraph answer at the top, key terms defined in the first 200 words, comparison tables and FAQ blocks carrying the substance. Decision-stage content is material written for buyers already comparing, which is a purchase-shaped intent rather than an interested one. Owning a term is not the same as being right about it. The definitions are checkable, the coinage is ours, and any vendor borrowing our vocabulary should be asked to say where their number for the same word comes from. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-10-four-ways-to-use-the-page-itself-1.png) ## Four ways to use the page itself The glossary is organised by reader rather than alphabetically, which is the only reason it is worth linking. \* *If you are evaluating a vendor pitch,*\* ignore the branding terms the tool uses for itself and read the methods. Ask which citations they measure, which passage-retrieval assumptions they make, and how they handle brand representation versus brand mention. \* *If you are briefing leadership,*\* AEO as the category, citation share as the headline metric, decision-stage content as the reason you are doing the work rather than watching it. \* *If you are writing the pages,*\* lead with the answer, use comparison tables and FAQ blocks, add the right structured data, refresh dated pages. \* *If you are buying a tool,*\* the three questions. Prompt volumes, prominence scoring, and what the implementation arm actually ships. If a vendor cannot answer the third question in one sentence, you have found the shape of the relationship before signing anything. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-11-who-this-is-for-1.png) ## Who this is for \* *Anyone evaluating an AI visibility platform:*\* print the four measurement terms and make the vendor say which one is in the headline metric. Watch for the person who answers with a product name. \* *Founders briefing a board:*\* AEO as the category, citation share as the metric, decision-stage content as the reason you are doing the work rather than monitoring it. That is three sentences and no arguments. \* *Writers and SEO leads:*\* your job is the answer-first paragraph and the structure. Every term in the engine group exists to explain why a well-written page still goes uncited. \* *Procurement:*\* undefined units are a financial risk. Add the definition to the contract schedule and it stops being a marketing conversation. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-12-a-note-on-how-confident-we-are-1.png) ## A note on how confident we are This is a vocabulary reference, not a measurement. It describes how terms are used across the tools and pages we examined, and that usage changes faster than any page can keep up. The glossary carries its own source and verification notes; read those rather than trusting my summary of them. Several entries are contested in the market rather than merely varied. Where a term has no settled meaning, the glossary says which camp uses which sense instead of inventing a neutral definition. Our own usage is not neutral either. We sell implementation, so "decision-stage content" is a term we benefit from defining well. Read that entry with the discount it deserves. Nothing here is a claim about how any engine ranks internally. Terms describe what vendors sell and what we can observe in published answers, not how a model decides. ![](https://www.datadab.com/blog/content/images/2026/10/ask-the-vendor-to-define-it-13-wrap-up-1.png) ## Wrap-up Eighteen terms, four of them load-bearing, and a category that has not agreed what it measures. Make the vendor say the words out loud before you read the price. Want the vocabulary in one place, with the sources? [Read the glossary](https://www.datadab.com/research/ai-visibility-glossary), then [bring it to the next vendor call as a checklist](https://www.datadab.com/ai-visibility-diagnostic). Term definitions and the three vendor questions come from our AI visibility glossary. The cost-per-query figures come from our pricing and entitlements study, dated on the page. [Get a second read on your visibility data](https://www.datadab.com/ai-visibility-diagnostic)