---
title: "The AI Visibility Glossary for Google Ads People: 18 Terms and Their Traps"
description: "Tracked prompts sound like keywords. Share of voice sounds like impression share. Here are 18 AI visibility terms translated into Google Ads language, including where the analogy breaks."
url: "https://groas.com/post/the-ai-visibility-glossary-for-people-wh"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/50af5d7e-e3a8-4815-a29b-1c516bfc36b2.png"
published: "2026-10-10T05:20:44.474Z"
modified: "2026-10-10T05:20:44.538Z"
---

October 10, 2026 · 10 min read

# The AI Visibility Glossary for Google Ads People: 18 Terms and Their Traps

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![A library drawer labelled "CLICKS" with red label tape reading "CITATIONS" stuck over it, peeling up: old ad metrics relabelled for AI.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/50af5d7e-e3a8-4815-a29b-1c516bfc36b2.png)

In this article

1. [Tracked prompt](#tracked-prompt)
2. [Mention](#mention)
3. [Citation](#citation)
4. [Recommendation](#recommendation)
5. [Share of voice](#share-of-voice)
6. [Prompt variance and run count](#prompt-variance-and-run-count)
7. [AI Overview](#ai-overview)
8. [AI Mode](#ai-mode)
9. [Answer engine](#answer-engine)
10. [AI Overview ads](#ai-overview-ads)
11. [Grounding and retrieval](#grounding-and-retrieval)
12. [Crawler hit vs verified bot vs live answer fetch](#crawler-hit-vs-verified-bot-vs-live-answer-fetch)
13. [robots.txt, sitemap, llms.txt](#robotstxt-sitemap-llmstxt)
14. [Entity and knowledge graph](#entity-and-knowledge-graph)
15. [Answer-first passage](#answer-first-passage)
16. [Content gap and answer mapping](#content-gap-and-answer-mapping)
17. [Off-site authority and citation building](#off-site-authority-and-citation-building)
18. [AI visibility score](#ai-visibility-score)

I got sent an AI visibility report last month that looked like a 2016 PPC report after a find-and-replace job. Keywords had become tracked prompts. Impression share had become share of voice. Clicks had become citations. The numbers looked familiar enough that I almost trusted them on instinct.

That is the trap. **Most AEO jargon is a renamed PPC concept, right up to the point where the analogy breaks.** A mention is not a citation. A crawler hit is not a live answer fetch. An AI Overview ad is not an AI Overview citation, however close together they appear on the screen. Vendors blur all three.

Here are 18 terms translated into Google Ads language, with the misuse each one invites. Keep your next vendor report open while you read.

## Tracked prompt

A tracked prompt is a fixed buyer question run against ChatGPT, Gemini or AI Overviews on a schedule to check whether your brand appears.

Think _keyword_, minus the auction and most of the stability. The mistake is treating one answer as a ranking. AI answers change from run to run; one mention proves about as much as one impression proving a keyword works. I used to tell clients a keyword needed 100 impressions before I judged it. That rule does not transfer neatly here. **Ask for repeated runs, with the run count beside every claimed win.**

## Mention

A mention is your brand name appearing in an AI answer, with or without a link.

This is the loosest outcome a visibility report can count. It resembles an impression: someone might have seen your name, but that alone does not give them a route to your site. I have seen reports roll mentions and citations into one green chart. That is like adding impressions and clicks together, then calling the total progress. **If the report says “mention,” ask whether the name was linked.**

## Citation

A citation is a linked source URL attached to an AI answer.

It is closer to a _click-eligible placement_ than to a click. A reader can follow the link; the report has not proved that anyone did. The distinction matters because a brand can appear in an answer while the linked source belongs to somebody else. [RedFork’s breakdown of AI answer outcomes](https://www.redforkmarketing.com/learn-article/rank-on-google-but-not-cited-by-ai) separates being considered, named, linked and recommended. **In your report, find the column for linked URLs.** An unlinked name-drop does not send a visit.

## Recommendation

A recommendation is an AI answer naming your brand as a choice for a buyer-style question, such as a best-of or near-me prompt.

This is the tempting _conversion_ analogy, but do not take it literally. A recommendation is still text in an answer, not a sale. Some tools distinguish the recommendation question, which they call GEO, from the linked-source question, which they call AEO ([Topify explains that split](https://topify.ai/blog/aeo-agent-vs-geo-agent-vs-seo-automation)). You can make a recommendation list while the hyperlink points to a competitor page. **Ask what the model recommended, what it linked and what the visitor did next.** Those are three different outcomes.

## Share of voice

AI share of voice is your brand’s mentions divided by all tracked brand mentions across a sampled set of AI answers ([Kitbase defines the measure this way](https://kitbase.dev/blog/ai-share-of-voice/)).

Think _impression share_, but inspect the denominator harder. A report saying you appeared in 70% of runs sounds fine until competitors appeared in 90% of those same runs. Presence rate is not competitive share. The sample also needs repeated runs, multiple engines and brands the tool was not already tracking; otherwise, leaving competitors out makes your share look bigger ([Kitbase explains the sampling problem](https://kitbase.dev/blog/ai-share-of-voice/)). **Watch the trend on a fixed prompt set, not one flattering snapshot.**

## Prompt variance and run count

Prompt variance is how much an answer changes across repeats of the same prompt; run count is the number of repeats behind the result.

If you have managed Google Ads, you know why one auction tells you little. Same keyword, different circumstances. AI answers move too, without an auction log to explain the movement. This matters when a vendor says your chart climbed four points. Was that based on repeated observations or one run per prompt per week? With the latter, you cannot tell improvement from noise. I run buyer prompts in batches of at least five before I let anyone claim a lift. **A chart without its run count is selling certainty it has not earned.**

The next terms are where paid and organic get mixed together. In Google Ads, you learn to spot the auction. In AI answers, a shared header can make two different kinds of placement look like one.

## AI Overview

An AI Overview is Google’s synthesized answer displayed alongside its conventional search results.

It gives you a new place to be quoted or linked, not simply another position on the familiar results page. The mistake is treating an Overview appearance as interchangeable with a ChatGPT appearance. A question can produce an Overview on Google and nothing comparable elsewhere. **Label the engine and answer type before comparing results.**

## AI Mode

AI Mode is Google’s more conversational AI search experience for questions that can develop across turns.

It is not just an Overview with more words. The same query can draw on different sources in AI Mode and an Overview, so a combined “Google AI” win rate can hide where you appeared. That is the reporting equivalent of blending Search and Shopping because both happen on Google. **Ask which experience produced each citation.**

## Answer engine

An answer engine is a system that responds to a question with a composed answer rather than only a list of links.

The useful PPC translation is a change in placement: you are competing to be useful _inside the answer_, not merely visible somewhere on a page. The rough process is to interpret the question, retrieve pages, use relevant passages and cite selected sources ([ZenMedia walks through that pipeline](https://zenmedia.com/blog/how-does-answer-engine-optimization-aeo-work/)). The mistake is assuming every answer engine uses the same sources for the same question. **Do not let a cross-engine total stand in for engine-by-engine results.**

## AI Overview ads

AI Overview ads are paid placements served through Google’s ads system in connection with an AI Overview.

An organic AI Overview citation is selected as a source; you cannot buy it with a bid. Vendors blur the two because both sit under the AI Overview label. I treat it like the old confusion between Shopping ads and organic listings, except the surrounding labels look more alike. **There is no organic citation slot to buy**, so a promise of guaranteed citations is a red flag ([RedFork explains the paid-organic distinction](https://www.redforkmarketing.com/learn-article/rank-on-google-but-not-cited-by-ai)). If you want the organic side, start with [the page already at position 8](https://groas.com/post/stop-writing-ai-overview-content-your-fa), not a new paid line in an organic report. Ask any agency reporting a blended win rate to split ads from citations.

## Grounding and retrieval

Grounding is basing an answer on fetched information; retrieval is the step that selects what information to fetch.

The PPC analogy is eligibility before performance: a page that never enters the candidate set cannot win a citation for that answer. The misuse is hearing “the model already knows your brand” and treating that as proof of a live placement. Training knowledge and a source fetched for this question are not the same thing. Indexing, fetch access and a passage the engine can use all matter when you want the page cited. **Ask whether the page was available and retrieved for the prompt, not whether the model could recall your name.**

## Crawler hit vs verified bot vs live answer fetch

A crawler hit is a request in your logs, a verified bot is a requester whose identity you have checked, and a live answer fetch is a page request made to help construct an answer.

These are not three names for visibility. A request claiming to be GPTBot or ClaudeBot may not be genuine; a genuine crawl may be routine inventory gathering rather than part of an answer. A log entry alone does not prove that a user saw your page cited. If your report celebrates thousands of AI bot hits, ask what was verified and what evidence connects any request to a live answer. The [curl and CDN audit for separating bot hits from answer activity](https://groas.com/post/can-chatgpt-actually-read-your-site-a-45) is a place to inspect that gap on your own site. **Access is not exposure, and exposure is not a citation.**

## robots.txt, sitemap, llms.txt

robots.txt sets crawler access rules, a sitemap lists URLs for discovery, and llms.txt is a Markdown file pointing readers or systems toward selected pages.

They solve different problems. The misuse is treating llms.txt as a block or as evidence that AI answers now feature your site. It blocks nothing. The evidence for visibility lift is thin: no major engine confirms such a lift, Google says it has no ranking effect, and one log analysis found that 97% of llms.txt files received zero requests ([Kitbase compares the files and the evidence](https://kitbase.dev/blog/does-llms-txt-work)). It is cheap to add, not a lever to build a plan around. Keep the [plain-English crawl and rendering glossary](https://groas.com/post/robots-txt-to-rendering-a-technical-glos) beside any technical report. **Fix crawler access and indexation first; treat llms.txt as optional.**

## Entity and knowledge graph

An entity is a recognizable business, product, place or concept; a knowledge graph stores entities and their relationships ([Search Engine Land explains Google’s Knowledge Graph](https://searchengineland.com/guide/knowledge-graph)).

Think _account-level context_, not a magic keyword field. In Google Ads, signals beyond one keyword help determine what happens in an auction. In AI answers, signals beyond one page help establish what your business is and how it relates to a question. The misuse is treating keyword density as entity authority. Repeating your category name 40 times does not make you the category’s defining business. Consistent business details, reviews and third-party references do more to clarify who you are. **When a rival gets named and you do not, check entity clarity before adding another paragraph of keywords.**

## Answer-first passage

An answer-first passage is a short, self-contained answer placed directly below the question it addresses, written so an engine can lift it into a response ([ZenMedia describes the extractable format](https://zenmedia.com/blog/how-does-answer-engine-optimization-aeo-work/)).

Think of the responsive search ad headline that actually matches the query, followed by a landing page that earns the click. The misuse is hiding the answer inside a 1,200-word introduction and calling the page optimized. A direct, usable passage gives the engine something to quote; the explanation beneath it gives a reader a reason to stay. The GEO work discussed in this space also points toward citations, quotations and statistics rather than keyword stuffing. **Write one direct answer under the question, then do the explaining.**

![A PPC report with keywords turned into chat bubbles and impression share relabeled as share of voice](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/05cec6e9-8cfc-476a-8b6b-c752f1db7987.png)

## Content gap and answer mapping

A content gap is a buyer question your site does not answer directly; answer mapping pairs each tracked prompt with the page meant to answer it.

This is the old search-term-to-ad-group sheet with prompts in place of queries. I used to make those sheets at 1am to find spend with no home. The mistake here is treating a page-one ranking as proof that a buyer question has an answer. One cited analysis puts only about 38% of AI Overview citations on top-10 pages, with the rest coming from lower-ranked pages or beyond the top 100 ([RedFork cites the breakdown](https://www.redforkmarketing.com/learn-article/rank-on-google-but-not-cited-by-ai)). Ranking and answer coverage are not the same thing. If ten buyer prompts map to three pages, read those pages: **which questions do they actually answer, and which have no home?**

## Off-site authority and citation building

Off-site authority is what other sites say about your business; citation building is the work of making those outside references accurate and useful.

Think of the parts of an account’s reputation you cannot fix by rewriting one ad. Reviews, lists, forum threads, supplier pages and local profiles can all provide context your own site cannot supply alone. The misuse is selling ten directory submissions as ten AI citations earned. A directory listing no answer retrieves is inventory, not a cited source. Running local accounts taught me to care less about the pile of identical listings and more about whether a reference said something specific a buyer needed. **Look at pages the engines retrieve for your prompts, then fix your facts where they appear.**

![Ink cartoon of an advertiser polishing a Quality Score dial locked behind glass](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/e0c6c210-3c58-45fd-9532-92e525490a29.png)

## AI visibility score

An AI visibility score is a single number that blends selected prompt, engine and appearance data into a headline measure.

This is the term I would retire. Singular, blended, often branded: it can roll mentions, citations and recommendations together until the number goes up without telling you what improved. I used to give clients one account-wide Quality Score average. I was wrong. The average hid the ad groups that needed attention.

Before accepting the big green number, make the vendor unpack it:

- **Citation:** Was there a linked URL, or only a name-drop?
- **Share of voice:** How many runs, which engines and which competitors went into the denominator?
- **AI Overview result:** What was bought through ads, and what was selected as an organic source?

Keep the useful pieces: prompts, mentions, linked citations, competitive share and run counts. Kill the blended score. If groas or anyone else sends you one big green number, make them show the parts underneath or send it back.

## Frequently Asked Questions

### What's the difference between a mention and a citation in AI visibility reports?

A mention is your brand name appearing in an AI answer with or without a link, while a citation is a linked source URL attached to the answer. An unlinked mention gives a reader no route to your site, so in reports you should ask whether the name was actually linked.

### How many times should a prompt be run before you trust an AI visibility result?

At least five repeated runs per prompt, because AI answers change from run to run and one answer proves about as much as one impression proving a keyword works. A chart without its run count beside it cannot distinguish real improvement from noise, so ask vendors to show the run count next to every claimed win.

### Why can AI share of voice numbers be misleading?

Share of voice is your brand's mentions divided by all tracked brand mentions in a sample of AI answers, so appearing in 70% of runs can sound fine while competitors appeared in 90% of those same runs. The sample also needs repeated runs, multiple engines and brands the tool was not already tracking, otherwise your share looks inflated.

### Can you buy an AI Overview citation?

No. AI Overview ads are paid placements served through Google's ads system, but an organic AI Overview citation is selected as a source and cannot be bought with a bid. A promise of guaranteed citations is a red flag, and any agency reporting a blended win rate should be asked to split ads from citations.

### Does adding llms.txt improve AI visibility?

It blocks nothing, and the evidence is thin: no major engine confirms a visibility lift, Google says it has no ranking effect, and one log analysis found 97% of llms.txt files received zero requests. It is cheap to add but optional, so fixing crawler access and indexation should come first.

### Does ranking on page one mean my pages will be cited in AI answers?

Not necessarily. One cited analysis puts only about 38% of AI Overview citations on top-10 pages, with the rest coming from lower-ranked pages or beyond the top 100. If several buyer prompts map to just a few pages, read those pages and check which questions they actually answer and which have no home.

### Should I trust a single AI visibility score in a vendor report?

Be careful, because a single blended score can roll mentions, citations and recommendations together so the number rises without telling you what improved. Ask the vendor to unpack it: was there a linked URL, how many runs and competitors went into the denominator, and what was bought through ads versus selected as an organic source?

### If my logs show lots of AI bot hits, does that mean my site is being cited?

Not on its own. A crawler hit is just a request in your logs, a verified bot is a requester whose identity you have checked, and a live answer fetch is a request made to help construct an answer. A request claiming to be GPTBot may not be genuine, and a log entry alone does not prove a user saw your page cited.

## Pay For Results, Not For Hours

Businesses buy the outcome, agencies resell it, and groas answers for it either way.

[See If You Qualify](https://groas.typeform.com/to/xC1bQNUT)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#webpage","url":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh","name":"The AI Visibility Glossary for Google Ads People: 18 Terms and Their Traps","description":"Tracked prompts sound like keywords. Share of voice sounds like impression share. Here are 18 AI visibility terms translated into Google Ads language, including where the analogy breaks.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/50af5d7e-e3a8-4815-a29b-1c516bfc36b2.png"},"breadcrumb":{"@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#breadcrumb"},"dateModified":"2026-10-10T05:20:44.538Z","mainEntity":{"@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"The AI Visibility Glossary for Google Ads People: 18 Terms and Their Traps","item":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#article","url":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh","mainEntityOfPage":{"@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"The AI Visibility Glossary for Google Ads People: 18 Terms and Their Traps","description":"Tracked prompts sound like keywords. Share of voice sounds like impression share. Here are 18 AI visibility terms translated into Google Ads language, including where the analogy breaks.","datePublished":"2026-10-10T05:20:44.474Z","dateModified":"2026-10-10T05:20:44.538Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/50af5d7e-e3a8-4815-a29b-1c516bfc36b2.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"answer, ads, google ads, google, visibility, share, overview, question","wordCount":2216,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"Ex Goldman Sachs and Ex Stanford Computer Science","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/the-ai-visibility-glossary-for-people-wh#faq","mainEntity":[{"@type":"Question","name":"What's the difference between a mention and a citation in AI visibility reports?","acceptedAnswer":{"@type":"Answer","text":"A mention is your brand name appearing in an AI answer with or without a link, while a citation is a linked source URL attached to the answer. An unlinked mention gives a reader no route to your site, so in reports you should ask whether the name was actually linked."}},{"@type":"Question","name":"How many times should a prompt be run before you trust an AI visibility result?","acceptedAnswer":{"@type":"Answer","text":"At least five repeated runs per prompt, because AI answers change from run to run and one answer proves about as much as one impression proving a keyword works. A chart without its run count beside it cannot distinguish real improvement from noise, so ask vendors to show the run count next to every claimed win."}},{"@type":"Question","name":"Why can AI share of voice numbers be misleading?","acceptedAnswer":{"@type":"Answer","text":"Share of voice is your brand's mentions divided by all tracked brand mentions in a sample of AI answers, so appearing in 70% of runs can sound fine while competitors appeared in 90% of those same runs. The sample also needs repeated runs, multiple engines and brands the tool was not already tracking, otherwise your share looks inflated."}},{"@type":"Question","name":"Can you buy an AI Overview citation?","acceptedAnswer":{"@type":"Answer","text":"No. AI Overview ads are paid placements served through Google's ads system, but an organic AI Overview citation is selected as a source and cannot be bought with a bid. A promise of guaranteed citations is a red flag, and any agency reporting a blended win rate should be asked to split ads from citations."}},{"@type":"Question","name":"Does adding llms.txt improve AI visibility?","acceptedAnswer":{"@type":"Answer","text":"It blocks nothing, and the evidence is thin: no major engine confirms a visibility lift, Google says it has no ranking effect, and one log analysis found 97% of llms.txt files received zero requests. It is cheap to add but optional, so fixing crawler access and indexation should come first."}},{"@type":"Question","name":"Does ranking on page one mean my pages will be cited in AI answers?","acceptedAnswer":{"@type":"Answer","text":"Not necessarily. One cited analysis puts only about 38% of AI Overview citations on top-10 pages, with the rest coming from lower-ranked pages or beyond the top 100. If several buyer prompts map to just a few pages, read those pages and check which questions they actually answer and which have no home."}},{"@type":"Question","name":"Should I trust a single AI visibility score in a vendor report?","acceptedAnswer":{"@type":"Answer","text":"Be careful, because a single blended score can roll mentions, citations and recommendations together so the number rises without telling you what improved. Ask the vendor to unpack it: was there a linked URL, how many runs and competitors went into the denominator, and what was bought through ads versus selected as an organic source?"}},{"@type":"Question","name":"If my logs show lots of AI bot hits, does that mean my site is being cited?","acceptedAnswer":{"@type":"Answer","text":"Not on its own. A crawler hit is just a request in your logs, a verified bot is a requester whose identity you have checked, and a live answer fetch is a request made to help construct an answer. A request claiming to be GPTBot may not be genuine, and a log entry alone does not prove a user saw your page cited."}}]}]}
```