Someone mentions “visibility” in a brief, and three people nod for three different reasons. AI search moves so fast that new terms arrive before we’ve defined the old ones.
This AEO glossary gives your team one shared reference, covering terms from agentic search to zero-click visibility. All entries come with definitions, and most include research, dos and don’ts, and additional reads.
A to Z Index
AEO · Agentic Search · AI Crawlers · AI Mode · AI Overview · AI Referral Traffic · AI Visibility Pyramid · Answer Engine · Atomic Answer · BLUF · Brand Mention · Chunking · Citable Facts · Citation Gap · Citation Rate · Content Refresh · Conversational Search · E-E-A-T · Entity · Entity Optimization · Extractability · Freshness Signal · GEO · Ghost Citation · Grounding · Hallucination · Hub and Spoke · Inauthentic Mentions · Information Gain · Knowledge Graph · llms.txt · Micro-Answer Block · Non-Determinism · Parametric Memory · Passage Retrieval · People Also Ask · Prompt Tracking · Query Fan-Out · RAG · Recommendation Rate · Retrievability · Scaled Content Abuse · Semantic Drift · Semantic Search · Share of Model · Source Diversity · Structured Data · Triple · Zero-Click Visibility
Terms by Topic
Topic | Terms |
|---|---|
How AI Search Builds Answers | Agentic Search · Answer Engine · Chunking · Conversational Search · Entity · Grounding · Hallucination · Knowledge Graph · Non-Determinism · Parametric Memory · Passage Retrieval · Query Fan-Out · RAG · Retrievability · Semantic Search · Triple |
What to Measure | AI Referral Traffic · Brand Mention · Citation Gap · Citation Rate · Ghost Citation · Prompt Tracking · Recommendation Rate · Share of Model · Source Diversity · Zero-Click Visibility |
How to Write and Structure Pages | Atomic Answer · BLUF · Citable Facts · Content Refresh · E-E-A-T · Entity Optimization · Extractability · Freshness Signal · Hub and Spoke · Information Gain · Micro-Answer Block · Semantic Drift |
Google Features and Policies | AI Mode · AI Overview · Inauthentic Mentions · People Also Ask · Scaled Content Abuse |
Disciplines, Frameworks, and Site Setup | AEO · AI Crawlers · AI Visibility Pyramid · GEO · llms.txt · Structured Data |
A
AEO (Answer Engine Optimization)
Answer engine optimization (AEO) is the practice of writing and structuring pages so AI answer engines can find a passage and cite it. The goal is for your brand or page to appear inside the generated answer, alongside your rankings in traditional search.
Further reading: SEO vs. AEO: A Field Guide for B2B SaaS Content Marketers
Agentic Search
Agentic search is search in which an AI agent researches for a user. Instead of running one query, it runs many (Query Fan-Out), compares the results, and synthesizes the findings.
Some agents also take actions on websites for the user, like submitting expenses or making a purchase.
For your AEO work: Check your server logs to see which pages AI agents visit. They identify themselves by name. For example, Google added Google-Agent in March 2026 for its agents that act on a user’s request. The AI Crawlers entry lists the other bots you'll see in your logs, like training and search crawlers.
Further reading: Everything You Need to Know about Agentic Search (Conductor)
AI Crawlers
AI crawlers are the bots AI companies send to fetch web pages. Platforms like OpenAI, Anthropic, and Perplexity run three kinds of crawlers:
Training crawlers (GPTBot, ClaudeBot).
Search crawlers that surface pages in AI search (OAI-SearchBot, Claude-SearchBot, PerplexityBot).
User-triggered fetchers that load a page when a user asks (ChatGPT-User, Claude-User, Perplexity-User).
Site owners steer crawlers with robots.txt, a file that tells bots which pages they may visit. Not every bot honors these instructions. OpenAI says robots.txt rules may not apply to its user-triggered fetcher, and Perplexity says its fetcher generally ignores them. Anthropic says all its bots honor robots.txt.
For your AEO work: Check robots.txt so you aren’t blocking search bots by mistake. Each bot has its own setting, so you can block training bots (GPTBot, ClaudeBot) and still appear in AI search. For Gemini training, the name to block is Google-Extended, which doesn’t affect Google Search.
Further reading: The AI Bots That \~140 Million Websites Block the Most (Ahrefs)
AI Mode
AI Mode is Google’s AI search experience for questions that need more exploration, reasoning, or comparison than just a list of links. It works as a conversation: you can ask with text, voice, images, or files, ask follow-up questions, and get answers that can include interactive diagrams, charts, and simulations.
Behind each answer, AI Mode uses Query Fan-Out. Google says it divides your question into subtopics and searches for all of them at the same time across multiple data sources.
AI Overview | AI Mode | |
|---|---|---|
Where it appears | On the regular results page | In its own space: google.com/ai, or the AI Mode button in Search |
When it appears | Only when Google judges it adds to classic Search, so often not at all | Whenever the user chooses it |
What you get | A snapshot of key information with links | A back-and-forth conversation, with follow-ups and interactive visuals |
Best for | Getting the gist of a topic | Exploring, reasoning through, or comparing options |
Query fan-out | May use it | Uses it |
Google notes that AI Mode and AI Overviews may use different AI models and techniques, so the answers and links they show vary. From an AI Overview, a user can continue into AI Mode with a follow-up question.
AI Overview (AIO)
An AI Overview (AIO) is the AI-generated summary Google shows on some search results pages: a snapshot of key information, with links to pages for digging deeper. Google shows an AIO only when doing so adds something to the regular results, so many searches don’t get one. A user who wants to go further can ask a follow-up and continue in AI Mode. The table in the AI Mode entry compares AI Overviews and AI Mode.
For your AEO work: Search your target queries. Note which ones show an overview and whether it includes your page. A page qualifies as a link if it’s indexed and eligible to show with a snippet in Google Search, Google says.
Further reading: AI Overviews Are Eating Your Search Traffic: Here’s How to Adapt
AI Referral Traffic
AI referral traffic is the visits your site gets when someone clicks a link in an AI assistant’s answer. Google Analytics 4 groups these visits in a default channel called “AI Assistant” (medium ai-assistant), for referrers such as ChatGPT, Gemini, Deepseek, Copilot, and Grok. The channel excludes clicks from Google’s own AI Overviews and AI Mode, which GA4 counts as Organic Search.
Referral numbers might understate how much buyers rely on AI. In Tracey Wallace’s Contentment newsletter (October 16, 2026), Klaviyo reported that 40% of its prospects use LLMs to evaluate their product, yet AI shows up as the referrer less than 2% of the time. People find Klaviyo in an AI answer, then go straight to their homepage or search for the brand on Google.
For your AEO work: Watch the GA4 AI Assistant channel, but don’t judge AI’s impact on that number alone. Like Klaviyo, also track direct and branded search traffic, and add a question to your signup flow asking whether people used an AI assistant to evaluate you.
AI Visibility Pyramid
The AI Visibility Pyramid is Animalz’s framework for building a brand’s presence in AI search. The base layer is traditional SEO: relevant content for your audience, which makes you visible in classic search.
The middle layer is context and content strategy. You learn how AI answers mention your brand and your competitors, then create citation-worthy content to change those answers.
The top layer, trust and authority signals, extends your content’s reach and credibility through reviews, PR, third-party mentions, and directory listings.
For your AEO work: Build from the base up, since each layer depends on the one below it. Use the pyramid to see which layer needs more attention. Once the base is in place, ship one improvement at each layer every month. For example, a high-intent page at the base, a citation-worthy asset in the middle, and a review, PR, or directory push at the top.
Further reading: AI Visibility Pyramid: How to Improve Your Presence in AI Search
Answer Engine
An answer engine is a search tool that answers a question directly instead of returning a list of pages that might contain the answer. Examples include AI assistants such as ChatGPT, Perplexity, Gemini, and Claude, and the AI features inside Google Search (AI Overviews and AI Mode).
For your AEO work: Answer engines don’t all find pages the same way. Google’s AI features retrieve pages through Google Search (see Grounding), while ChatGPT, Claude, and Perplexity send their own search crawlers (see AI Crawlers). Check how your brand shows up in each engine your buyers use, not only in Google.
Atomic Answer
An atomic answer is a short passage that answers one question and still makes sense when it’s read without the rest of the page.
It matters for AI search because answer engines work with passages, not whole pages. Google says its AI features retrieve relevant pages and then “review the specific information from those retrieved pages” to build a response. Microsoft says assistants like Copilot break content into “smaller, structured pieces,” then assemble those pieces into answers. A passage that stands on its own can be used as is.
Example: For the question “What is a 301 redirect?,” the answer “It permanently sends users to a new URL” doesn’t make sense on its own. The answer “A 301 redirect permanently sends users and search engines from an old URL to a new one” does.
Further reading: 20 Techniques That Get You Cited in Answer Engines
B
BLUF (Bottom Line Up Front)
BLUF (Bottom Line Up Front) is the practice of putting the main point at the start of a piece of writing, before any other background or detail. It comes from US Army writing standards, which call for “putting the main point at the beginning of the correspondence.”
BLUF helps AI search as well as readers. Before an AI tool writes an answer, it picks the passages that best match the question. A 2025 study by Zeng et al. found that many retrieval models are on average 15.6% worse at spotting a match when the answer sits later in a passage.
Example: A section on Google’s March 2024 core update that opens with three paragraphs on the history of search updates buries the point. Opening with “Google’s March 2024 core update targets scaled content abuse across mass-produced, low-value pages” BLUFs it.
For your AEO work: Answer the heading’s question in the section’s first sentence, then add background and caveats.
Further reading: BLUF: The Military Standard That Can Make Your Writing More Powerful
Brand Mention
A brand mention (sometimes called an AI mention) is when an AI answer names your brand in the response text, such as ChatGPT listing your product among top tools in a category. Visibility tools like Ahrefs Brand Radar count one brand mention per response that names your brand. They track mentions separately from citations, where the answer links to your page as a source.
Brand mentions and citations often happen without each other. An answer engine might name your brand from third-party coverage without linking to your site, or link to your page as a source without naming your brand in the answer text (a ghost citation).
A 2026 study by Semrush and Kevin Indig counted 3,981 domain appearances across ChatGPT, Gemini, Google AI Overviews, and Google AI Mode: 25.1% named the brand without a source link, 61.7% were ghost citations, and only 13.2% included both a mention and a source link.
Example: An answer to “What tools track AI visibility?” that says “Ahrefs and Semrush both track AI visibility” gives each brand a mention. If the answer also links to an Ahrefs help page as its source, Ahrefs gets a citation too.
For your AEO work: Report brand mentions next to citations and your citation rate. Together, those numbers show whether answer engines name your brand while citing someone else’s page, or cite your pages without naming you.
C
Chunking
Chunking is the process AI search systems use to split a document into smaller pieces, called chunks, so the system can retrieve only the pieces that match a question.
Chunking happens on the system’s side. Google Cloud’s RAG Engine, for example, splits each document into chunks of a set size, with some overlap between neighboring chunks. When a question comes in, the system pulls up the best-matching chunks, the step covered in Passage Retrieval.
In content marketing, “chunking” also refers to breaking web pages into bite-sized pieces for AI. Google says “there’s no requirement to break your content into tiny pieces for AI to better understand it.”
Example: In a 3,000-word email marketing guide, a retrieval system stores a long paragraph that mixes deliverability rules with subject-line tips as one unfocused chunk. Splitting those two topics into separate paragraphs under separate headings gives the system a clean chunk for each query.
For your AEO work: Don’t split a strong page into fragments for AI. Organize the page for your readers with clear sections and headings, and write key passages as atomic answers.
Further reading: Moving from a Google-shaped Web to an Agent-shaped Web: A Refutation of Misinformation about Chunking (iPullRank)
Citable Facts
Citable facts are specific, checkable facts, figures, and quotes, with their sources named, that an answer engine can lift into an answer and attribute.
A research paper on generative engine optimization tested whether citable facts make pages more visible in AI answers. The researchers rewrote web pages in different ways and checked which versions a generative search engine showed more often. Adding statistics, quotes from credible sources, and citations to reliable sources did best, lifting visibility in the engine’s responses by up to 40%. (Adding more keywords did little.)
Example: “According to Ahrefs’ 2025 study of 17 million citations, AI assistants cite URLs that are, on average, 25.7% fresher than organic Google results” gives an engine a checkable stat and a source. “AI assistants prefer newer pages” has nothing to cite.
For your AEO work: Replace vague claims with named sources, exact figures, and direct quotes where you have them.
Citation Gap
A citation gap is a query where your page ranks in Google’s organic search results, yet AI answers for that same query don’t cite the page. High organic rankings don’t guarantee AI citations: in an August 2025 Ahrefs study of 15,000 long-tail prompts, only 12% of links cited by ChatGPT, Gemini, and Copilot also appeared in Google’s top 10 organic results for the same prompt.
Example: A page ranks third in Google’s organic results for “content decay,” but when you ask Perplexity “What is content decay?,” it cites three competitors and leaves the page out.
For your AEO work: Take the queries your pages already rank for in Google and check them in ChatGPT, Perplexity, and Google’s AI Overviews. Note where they cite competitors but not your ranking page. Refresh those pages first: Google already ranks them, so a content refresh builds on visibility the pages have earned.
Further reading: SEO vs. AEO: A Field Guide for B2B SaaS Content Marketers
Citation Rate
Citation rate is the percentage of relevant AI answers that link to your domain or page as a source. AI visibility tools such as Profound, Ahrefs Brand Radar, and geoSurge track citations alongside brand mentions and share of model.
Tools calculate citation numbers in different ways. geoSurge calculates citation rate as the share of AI answers about your topics that cite at least one URL from your domain. If 30 of 100 tracked answers cite your site, your citation rate is 30%.
Ahrefs Brand Radar reports citations as a count instead, next to a separate count of brand mentions. Ahrefs counts one citation for a domain per answer, so an answer that cites five of your pages adds one citation for your domain, not five.
Share-of-voice metrics measure something else again: how your visibility compares with your competitors’ (see Share of Model).
For your AEO work: Before you compare citation numbers from two tools, check what each one counts: the share of answers that cite you, a count of citations, or your share against competitors. Write that definition next to the number in your dashboard.
Further reading: 20 Techniques That Get You Cited in Answer Engines
Content Refresh
A content refresh is the practice of updating a published page’s facts, examples, and structure on the existing URL instead of writing a new piece.
Content refreshes help pages stay visible in both Google and AI answers. Google runs “query deserves freshness” systems to show fresher content for queries where users expect current information.
AI answers mostly cite recently updated pages, even on everyday topics: in a 2026 Seer Interactive study of non-branded questions, 75% of cited pages had been updated in the last year, and more than a quarter of those were first published over two years ago (see Freshness Signal). In Tracey Wallace’s Contentment newsletter (October 16, 2026), Klaviyo reported that refreshed assets get 154% more citations than net-new assets.
At Animalz, we call a small update a micro-refresh: new stats, working links, and tweaks based on how the page performs in AI search.
Example: Changing the publish date on a guide without editing the text doesn’t count as a refresh. Replacing 2023 figures with current data, adding a direct answer under each H2, and then updating the visible date does.
For your AEO work: Refresh decaying pages that already rank before you commission new articles on similar topics. Our free tool Revive finds them for you: it analyzes 12 months of your Google Analytics traffic and ranks the posts worth refreshing by potential impact. We recommend regular micro-refreshes, plus less frequent deep refreshes that rewrite weak sections.
Further reading: Content Refresh Strategy: How to Update Old Content for SEO and AI Search
Conversational Search
Conversational search is a way of searching where you ask a question in full sentences, get one written answer, and then ask follow-up questions that build on what came before.
Google’s AI Mode is one product built for conversational search. In AI Mode, you type a question into the “Ask anything” bar, get an AI-powered response with links to explore further, and then type a follow-up into the same bar instead of starting a new search.
For your AEO work: In Google’s AI Mode, ask one of your target queries, note which pages the answer links to, then ask the follow-up a buyer would likely ask next and check the links again. A page that appears in the first answer but not in the follow-up covers the opening question but not the next one, which shows you what to add.
E
E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness)
E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is Google’s shorthand for the qualities its ranking systems look for when they decide which content is most helpful. Of the four, Google says trust matters most.
E-E-A-T itself isn’t a specific ranking factor, Google says. Instead, Google’s systems use a mix of factors to spot content that shows those qualities. Bylines are one example Google points to: a byline that leads to background on the author and the areas they write about makes it clear who created the content. Google also warns that fabricating creator profiles, with AI-generated headshots, made-up names, or false credentials, is a form of deception.
The same thinking carries over to AI search. Google’s guide to AI features says the foundational SEO best practices for Search, including helpful, reliable, people-first content, also apply to AI Overviews and AI Mode.
For your AEO work: Add an author byline with real credentials to each article, include first-hand experience (such as original test results) in the body copy, and cite outside validation from primary sources or third-party reviews. These signals show readers and answer engines that a qualified expert wrote the page.
Further reading: Answer Engine Optimization (AEO): The Complete Guide for 2026 (CXL)
Entity
An entity is a specific person, place, organization, product, or concept that a search engine or language model can recognize in text.
Google’s Natural Language API sorts entities into two groups: proper nouns, which name one specific thing (a public figure or a landmark), and common nouns, which name a general kind of thing (a restaurant or a stadium).
Google’s Knowledge Graph is built on entities. When Google introduced the Knowledge Graph in 2012, the company described it as a model that “understands real-world entities and their relationships to one another.”
Google uses the query “taj mahal” to show why that matters: the words could mean the monument, a Grammy Award-winning musician, or a casino in Atlantic City, and the Knowledge Graph lets Google tell them apart.
Example: A page that says only “Animalz” leaves a search engine to guess whether it means a company or a misspelling of “animals.” A page that says “Animalz, a B2B content marketing agency” points to one organization.
Entity Optimization
Entity optimization is the practice of naming and describing your brand and products the same way everywhere they appear, so search engines and AI systems recognize them as specific entities. The goal is for those systems to connect your site and your profiles elsewhere to the same entry in knowledge bases like Google’s Knowledge Graph.
Structured data, often called schema markup, is one way to make your organization easier to recognize. Google’s Organization structured data guide says adding markup to your home page can help “disambiguate your organization in search results,” which means telling your organization apart from others with similar names. The markup can also shape which logo Google shows in search results and in your knowledge panel.
The same markup can link to your profiles on other sites, such as a social media or review site. Google asks you to use the same name in the markup that you use for your site name.
Example: Calling a product “Acme” on the homepage, “Acme BI” in the markup, and “our reporting tool” on G2 leaves engines unsure whether these are one product or three. Calling it “Acme Analytics, a B2B product analytics platform” in all three places gives them one entity to recognize.
For your AEO work: Use one brand name, product description, and category on your homepage and directory profiles, and link to those profiles from the Organization markup on your homepage. Matching those details helps answer engines attribute every mention and review to the same business.
Extractability
Extractability is how well a passage keeps its meaning when an AI system lifts it out of the page and uses it in an answer, without any of the surrounding text.
AuthorityTech defines AI extractability as the degree to which a content section “can be parsed, compressed, and reused by AI systems without losing its core meaning.” One check AuthorityTech scores is whether a passage names its subject or leans on pronouns.
Extractability differs from an atomic answer. An atomic answer is a type of passage: one you write to answer one question on its own. Extractability is a quality every passage has to some degree: how well it holds up once it’s lifted out of the page.
Example: “The Machine Relations Index measures cross-platform citation share” holds up on its own. “It measures this across platforms” doesn’t.
For your AEO work: Hide each H2 and H3 heading, read the section’s first two sentences, and replace any “it,” “this,” or “they” that no longer points to a clear subject with the specific noun. Answer engines can pull a paragraph without its heading, so naming the subject in the text keeps the passage usable on its own.
Further reading: 20 Techniques That Get You Cited in Answer Engines
F
Freshness Signal
A freshness signal is any cue that shows a page is current: a dateModified value in structured data, a visible last-updated date, the sitemap’s lastmod field, and recent figures within the text.
A 2026 Seer Interactive study looked at 47,097 citations in ChatGPT, Gemini, and Perplexity answers to non-branded questions for four brands: a pet retailer, a vacation rental marketplace, a retail energy provider, and a commercial bank. Seer dated each cited page from signals such as its structured data and sitemap, and found that 75% of the pages had been updated in the last year. The study shows when cited pages were last updated, not that updating a page causes citations.
For your AEO work: When you update a page’s substance, update the visible date, dateModified schema, and sitemap lastmod together so crawlers and readers see the same timestamp.
G
GEO (Generative Engine Optimization)
GEO (Generative Engine Optimization) is the practice of improving how often your content appears in AI-generated answers. The name comes from a 2023 research paper, “GEO: Generative Engine Optimization,” which introduced GEO to help content creators improve “their content visibility in generative engine responses.”
Marketers and SEOs use GEO, AEO, and LLMO (Large Language Model Optimization) for overlapping work. Nobody has settled whether the three names mean different things.
Ghost Citation
A ghost citation is an AI answer that links to your page as a source without naming your brand in the answer text. A ghost citation is the opposite of an unlinked brand mention, where the answer names your brand but doesn’t link to you.
Kevin Indig coined the term in a 2026 study with Semrush of 115 prompts across four AI engines. Informational queries (“what is,” “explain,” “how does”) produced the most ghost citations: brands that appeared in those answers were cited as a source 89.3% of the time but named only 18% of the time. Comparative queries (“best,” “vs,” “recommend”) produced 2.4 times as many brand mentions as informational ones.
Usage varies: Indig’s own Growth Memo glossary uses the term for any answer that draws on your content without a visible link back. Promptwatch uses it for a mention without a link.
For your AEO work: Look for informational pages that answer engines cite without mentioning you. On those pages, add clear references to your brand and product, or a comparison section featuring specific tools and companies. Tying your insights to specific products gives the answer engine a reason to name your brand instead of only linking to the page.
Further reading: The Ghost Citations Study (Semrush)
Grounding
Grounding is connecting an AI model’s answer to verifiable outside sources, instead of relying only on what the model learned in training.
Google’s AI optimization guide describes grounding, which it also calls retrieval-augmented generation (RAG), as relying on its Search ranking systems “to retrieve relevant, up-to-date web pages from our Search index.” Google’s systems then review those pages, write the response, and link to the pages they drew on.
Grounding lowers the risk of hallucination but doesn’t remove it. Retrieval can also add its own errors. A January 2026 survey of research on RAG systems notes that pulling in outside pages introduces new kinds of hallucination, which come from how the system’s search and writing steps interact.
H
Hallucination
A hallucination is false content that an AI model generates and presents as fact.
Vectara’s Hallucination Leaderboard measures how often models add content that isn’t in a document when they summarize it. On October 22, 2026, Ant Group’s Finix S1 32B had the lowest rate, adding content to 1.8% of its summaries. Several models did so in more than 20%, including OpenAI’s o3-pro at 23.3%, and half of all models did so in at least 9.6%.
OpenAI’s 2025 research argues that models hallucinate because standard training and testing reward a confident guess over admitting uncertainty.
Whether rates are falling is hard to say: OpenAI says GPT-5 hallucinates less than its earlier models, but Vectara changed its test documents between versions, so its numbers can’t be compared across years.
For your AEO work: State product facts, pricing, and key claims in plain sentences on your pricing page, product pages, and About page, and link each claim to a primary source. Clear statements on those pages give AI systems less to guess when they describe your company.
Hub and Spoke
A hub and spoke is a group of pages on one broad topic. The hub links out to separate spoke pages, each covering one subtopic, with every spoke linking back to the hub.
Animalz’s article on hubs and pillars describes the setup this way: “The hub is a table of contents that links out to many articles that cover subtopics.” The hub’s job is to send readers to the spokes, not to keep them on the hub.
A pillar does the opposite and puts every subtopic on one long page: fold a hub and its spokes into a single page, and you have a pillar.
At Animalz, we use hub and spoke mostly to target one competitive keyword, since a single blog post per keyword usually isn’t enough to rank.
For AEO, pick the structure by how people search the subtopics:
Hub and spoke, when each subtopic is a search people run on its own. Each spoke can rank and be cited for its own fan-out search.
Pillar, when the subtopics belong together and people rarely search them on their own. A pillar can still be cited: Google says its systems can pick out the relevant part of a long page.
Intent constellation, when you’re answering one question and its follow-ups. Animalz’s intent constellation is a single page that answers the main question plus the follow-up questions people ask about it.
Example: One long page on content marketing ROI that covers calculating ROI, industry benchmarks, and attribution models is a pillar, not a hub. An overview of content marketing ROI that links to three separate articles on those subtopics, each linking back, is a hub and spoke.
For your AEO work: Link the hub to every spoke, and link each spoke back to the hub and to the other spokes. Those links make the set easy to move through and give each spoke its own chance to rank while it sends readers back to the hub.
Further reading: Hubs vs. Pillars: What’s the Difference?
I
Inauthentic Mentions
Inauthentic mentions are brand placements bought on third-party sites to sway AI answers, usually without disclosure. Inauthentic mentions differ from an earned brand mention, which comes from genuine reviews, PR, or independent coverage.
In her October 2026 Substack post, Lily Ray explains how some AI visibility platforms and agencies sell inauthentic mentions: placements in “best X” listicles on third-party sites, sold per placement, often starting around $135 to $250 each, and packages of “AI brand mentions,” such as five mentions for $1,500. Ray calls paid brand mentions “the 2026 version of paid link building.”
Google warns against the practice in its AI optimization guide: “Seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.” Its AI features draw on the same systems that favor high-quality content and block spam, so mentions bought on low-quality pages are unlikely to carry much weight.
Example: A vendor that pays a respected niche blog for a spot in its “Best B2B Payroll Tools” list, with no sponsored label, has bought an inauthentic mention, even though the blog looks independent. A vendor that an independent HR analyst includes in a benchmark report after testing the product has earned a mention.
For your AEO work: Put your budget into PR, customer reviews, and original research that other sites cite on merit, and skip paid listicle packages pitched for AI visibility.
Information Gain
Information gain is the new information a page adds beyond what readers can already find in other results on the same topic. SEOs took the term from a Google patent that scores a document by how much new information it adds to the pages a user has already seen. Google hasn’t said whether it uses the patent.
Information gain matters in AI search because an answer engine pulls from multiple sources at once. Google’s AI optimization guide tells site owners to write “non-commodity content:” unique expert or firsthand takes that go beyond common knowledge. Google says that kind of content will likely shape your site’s presence in generative AI search more, in the long run, than any of the guide’s other suggestions.
As Melissa Popp of RicketyRoo puts it in Zyppy’s survey of 130+ search experts, “If AI search already has ten pages saying roughly the same thing about a topic, page eleven needs to bring something those ten didn’t.”
Example: A guide on content decay that repeats the same definition and generic fixes as the top ten search results adds no information gain. Publishing original benchmark data on how fast different article types lose traffic gives answer engines new facts to cite.
For your AEO work: Before approving a brief, name the specific data point, firsthand example, or argument the draft will add that isn’t on the ten pages already ranking in Google’s organic results. Adding a fact or perspective that answer engines can’t pull from existing pages makes your article worth citing.
Further reading: Information Gain: The SEO Theory that AI Made Mandatory
K
Knowledge Graph
A knowledge graph is a database of real-world entities, such as people, places, and organizations, and the relationships between them. Google’s Knowledge Graph is the best-known example: Google uses it to understand which real-world things a query or page is about.
A knowledge graph links entities to each other instead of only storing facts about them. In Google’s example, Marie Curie is linked to her husband, Pierre Curie, and to her two children. Each link is a relationship between two entities, which can be written as a triple.
Searchers see Google’s Knowledge Graph in the knowledge panel, a summary of key facts about the entity they searched for. Organization markup on a home page feeds into it in two ways, according to Google’s Organization structured data guide:
Some properties help Google tell your organization apart from other organizations.
Others can influence what searchers see, such as “which logo is shown in Search results and your knowledge panel.”
L
llms.txt
llms.txt is a proposed Markdown file that gives large language models a concise guide to a site’s content. Placed at the site root or under a subpath, the file offers brief background notes and links to key pages. The proposal was first written in 2024 and is published at llmstxt.org.
Google says Google Search, including AI Overviews and AI Mode, doesn’t use llms.txt files. A June 2026 Search Central update adds that the files “won’t negatively or positively impact your visibility or rankings.”
SEO practitioners are similarly unmoved. Zyppy’s survey asked 130+ search experts how much each of 13 factors affects whether Google AI Overviews and AI Mode cite a page, on a scale from -3 to +3. They rated llms.txt at +0.05, the lowest of the 13.
Survey respondent Przemysław Charchan says he “collected 7 independent studies on llms.txt, covering nearly 500,000 domains, and all of them found that AI bots do not check this file on their own.”
In 90-day server-log tests summarized by Optimizely, Limy recorded 408 llms.txt requests out of more than 500 million AI bot visits, while Otterly logged 84 across 62,100+.
For your AEO work: Skip building or maintaining llms.txt for search visibility, and spend that engineering time on crawlable HTML and clear on-page structure instead.
M
Micro-Answer Block
A micro-answer block is a compact, self-contained page element that gives answer engines a ready-made unit to pull into a response. Examples include a definition box, a short step list, or a pros-and-cons table.
An atomic answer is a passage you write to answer one question on its own, and extractability is how well any passage holds up once it’s lifted out of the page. A micro-answer block is the formatted container that holds an answer on the page.
Example: Burying a definition of usage-based pricing three paragraphs into a narrative intro doesn’t give answer engines a clean block to pull. Placing a two-sentence definition box at the top of the section does.
Further reading: SEO vs. AEO: A field guide for B2B SaaS content marketers
N
Non-Determinism
Non-determinism is the fact that AI assistants can return different outputs to the same prompt across separate runs. OpenAI notes that its chat completions are non-deterministic by default.
The variation is large when assistants recommend brands. In a SparkToro study, 600 volunteers ran 12 prompts asking for brand or product recommendations through ChatGPT, Claude, and Google’s AI tools a combined 2,961 times. For those prompts, the odds of getting the same list of brands twice were less than 1 in 100, and the odds of getting the same list in the same order were about 1 in 1,000.
Because answers vary this much, Prompt Tracking runs each prompt many times and reports how often a brand appears.
For your AEO work: Treat a single screenshot of an AI answer as an anecdote, and judge visibility from repeated samples of the same prompt over time. The next run of that prompt will likely name a different set of brands.
P
Parametric Memory
Parametric memory is what a language model knows from its training data, stored inside the model itself. Non-parametric memory, by contrast, is information the system looks up in an outside source when it answers.
Parametric memory is one of the hardest AEO factors to change quickly. In Zyppy’s survey of 130+ search experts on Google AI Overviews and AI Mode, a brand’s presence in the model’s memory ranked third of the 13 factors the experts rated. You can win live retrieval page by page, but a model’s stored knowledge builds slowly, based on how the wider web describes your brand.
For your AEO work: Publish original data and clear answers on your own pages, and use one consistent company description in your press boilerplate, guest post bios, and third-party review profiles. Repeating those facts across the web makes it more likely that models encounter the same description of your brand, both in training data and in live retrieval.
Passage Retrieval
Passage retrieval is the step where a search system or answer engine pulls specific sections from a source page to select the text for an answer.
In AI Overviews and AI Mode, for example, Google’s ranking systems fetch relevant URLs from the Search index and then review the specific information on those pages to generate a response with source links. Google says its systems can tell several topics apart on one page and use the relevant piece, so writers don’t need to break content into tiny segments (chunking).
Passage retrieval might still favor text near the top of a page, though the evidence so far comes from a small sample. In CXL’s analysis of 100 Google AI Overview citations, 55% of cited passages came from the first 30% of the source page (where BLUF puts the main point), 24% from the middle section (30% to 60%), and 21% from the bottom 40%. Several citations near the bottom came from FAQ blocks, where each question-and-answer pair stood on its own as an atomic answer.
People Also Ask (PAA)
People Also Ask (PAA) is the box of related questions on a Google search results page. Each question expands to show a short answer and a link to the page it came from. That answer can appear as a featured snippet, which flips a regular search result: the text comes first, with the page link below it.
Google’s systems pick which page to feature without any markup from the site, and clicking a snippet takes the user straight to the section of the page that supplied the passage.
For your AEO work: Search your target query in Google and add the People Also Ask questions that fit the topic to your content brief as H2 subheadings or FAQ items. Under each heading, write an atomic answer. That gives Google a self-contained passage to pull into the snippet box, and it covers the follow-up questions answer engines address alongside the main query.
Prompt Tracking
Prompt tracking is the practice of running a fixed set of prompts through answer engines on a schedule and recording which brands and sources appear in the responses. Prompts are to AI visibility what keywords are to SEO.
Prompt tracking tools typically rerun prompts daily, and each tool counts different signals:
Profound runs tracked prompts in engines such as ChatGPT, Gemini, and Google AI Overviews. It reports the share of responses that include a brand, which feeds metrics like share of model.
Semrush’s Prompt Tracking bases its visibility score on citations in ChatGPT and Google AI Mode instead, running from 0% when a domain never appears among the top cited sources to 100% when the site holds the first spot for every tracked prompt.
Prompt tracking is more reliable for measuring how often your brand shows up than where it ranks in a single answer. The same prompt rarely returns the same list of brands twice (non-determinism).
In a SparkToro study, when volunteers ran 142 differently worded queries for one product category 994 times, the top brands still appeared in 55% to 77% of answers. SparkToro concluded that visibility percentage across dozens to hundreds of prompts is a decent proxy for how often a brand shows up, while rank position in an AI answer isn’t.
For your AEO work: Track a set of prompts in a visibility tool, and report the percentage of responses that include your brand across the whole list. Buyers rarely phrase questions the same way and AI models shuffle recommendation order, so presence across a prompt set is a steadier metric than average position.
Q
Query Fan-Out
Query fan-out is when an AI search engine breaks one question into several related searches, runs them at the same time, and uses the results together to write one answer.
Google says AI Overviews and AI Mode may use it, “issuing multiple related searches across subtopics and data sources.” In Google’s AI optimization guide, a search for “how to fix a lawn that’s full of weeds” might fan out into “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in lawn.”
Engines fan out at different rates. In an analysis of 5 million fan-outs by Peec AI, an AI visibility tracking company, Perplexity ran an average of 1.4 fan-out searches per prompt, ChatGPT 2.1, and Grok 6.8. Peec AI also found that ChatGPT adds brand names and words like “best,” “top,” and “reviews” to its fan-out searches.
Google warns that creating separate content for fan-out queries “primarily to manipulate rankings or generative AI responses in Google Search violates Google’s scaled content abuse spam policy.”
Example: Spinning up separate thin pages for “programmatic SEO examples,” “programmatic SEO risks,” and “programmatic SEO vs. traditional SEO” risks a spam violation. One “What is programmatic SEO?” page that covers all three can meet those fan-out searches on a single URL.
For your AEO work: Answer the sub-questions a topic raises on the same page as the main question, rather than giving each one a page of its own. At Animalz, we call these pages “intent constellations.”
Further reading: How AI Search Platforms Expand Queries with Fan-Out and Why It Skews Intent (iPullRank)
R
RAG (Retrieval-Augmented Generation)
Retrieval-augmented generation (RAG) is an approach that retrieves passages from an outside source, such as the web or a document index, then uses them to generate an answer. Google says RAG improves the quality, accuracy, and freshness of AI responses. The approach comes from a 2020 paper by Lewis et al.
Example: Instead of answering from training data alone, ChatGPT search retrieves web results first and uses them to generate a response with source links.
Recommendation Rate
Recommendation rate is a metric for how often AI answers actively recommend a brand, as opposed to only mentioning it. The term comes from people who track AI visibility, not from the AI platforms themselves. Recommendation rate is a stricter test than a brand mention or share of model: an answer can name a brand without endorsing it.
Recommendations seem to drive visits, even without a link. In a Similarweb study of US desktop users (July to December 2025), the brand ChatGPT recommended got more visits in the following seven days than its competitor in all six brand pairs tracked. The gap ranged from Skyscanner (9.5% of users visited, against 7.6% for Kayak) to Capital One (14.2%, against 3.8% for American Express).
Example: Asked “What’s the best email platform for ecommerce?,” an answer that lists eight platforms gives each one a mention. An answer that says one platform is the best fit for ecommerce stores gives that platform a recommendation.
For your AEO work: In your tracked buying prompts (such as “best X for Y”), mark each answer that names your brand as either a recommendation or a passing mention. Then open the sources cited on the prompts where a competitor gets the recommendation: those pages show what the engine relies on when it endorses someone else.
Retrievability
Retrievability is how easily a retrieval system can find a document across the queries it may be asked, a measure introduced by Azzopardi and Vinay (2008).
In AEO, retrievability is how likely an answer engine is to pull your page into the shortlist of sources it considers for a question. A page that never makes the shortlist can’t be cited, however good its answer.
You can’t see that shortlist, but you can check two things that feed into it:
Whether AI crawlers can reach the page (see AI Crawlers).
For Google’s AI features, whether the page ranks in Google Search, since those features retrieve pages through Search’s ranking systems (see Grounding).
Further reading: AI Visibility Pyramid: How to Improve Your Presence in AI Search
S
Scaled Content Abuse
Scaled content abuse is Google’s spam policy against generating many pages “for the primary purpose of manipulating search rankings” instead of helping users. Google announced the policy in March 2024 for mass-produced pages created by automation, humans, or a mix of both.
The scaled content abuse policy lists “Using generative AI tools or other similar tools to generate many pages without adding value for users” as an example. Google’s AI optimization guide specifically names building separate pages for Query Fan-Out variations to manipulate rankings or generative AI responses as a violation.
Semantic Drift
Semantic drift is a change in what a word means, or how people use it. In SEO and AEO, it describes how an evergreen page can lose relevance as the language around its topic moves on, even when the page is still accurate.
Example: A search guide that still calls Google’s feature “SGE” after Google renamed it “AI Overviews” reads as outdated. An updated guide uses “AI Overviews” and mentions the former name once for readers who search the old term.
For your AEO work: When you refresh an evergreen page, search its key terms in Google, note the names the top results use now, and swap any outdated term on the page for the current one. That keeps the page findable in the words your readers search with today.
Semantic Search
Semantic search is a search technique that matches a query on its contextual meaning and intent, not only on its keywords. For example, “how to reduce customer churn” can surface a page titled “retention strategies that work,” even though the two phrases share no words.
A page can match searches that never use its exact keywords, as long as the page clearly covers the idea the searcher is asking about. This matters for AEO because AI search works on the same idea: the retrieval models that pick passages compare meanings, not only words (see Passage Retrieval). One clear answer can match many phrasings of a question, so you don’t need a separate section for each phrasing.
Share of Model (AI Share of Voice)
Share of model is how often AI assistants name your brand compared with your competitors, across the questions people ask about your category. AI share of voice is another name for the same metric.
A prompt tracking tool measures share of model. The tool runs a fixed set of prompts through AI assistants on a schedule and counts how often your brand and your competitors appear.
Share of model has no standard definition, so tools measure it in different ways. Anglera’s definition counts the share of answers that mention or cite your brand. Semrush’s Brand Performance report looks at your share of all brand mentions instead. Because the methods differ, two tools’ numbers for the same brand don’t compare.
Example: A brand appears in 6 of 10 AI answers about project management tools, and those answers name competitors 24 times. Counted by answers, the brand’s share of model is 60%. Counted by mentions, it’s 20% (6 of 30).
For your AEO work: Keep one tracking tool and one fixed prompt set. Next to the share-of-model number in your reports, write how many prompts you track, which engines the tool tests, and whether it counts answers or mentions.
Source Diversity
Source diversity is the variety of domains and content types that an answer engine cites in a single response. Google says its AI Overviews and AI Mode use query fan-out to show “a wider and more diverse set of helpful links” than a classic web search. In a 2025 Pew Research Center study of 12,593 Google AI summaries, 88% cited three or more sources, and only 1% cited a single source.
Source diversity across queries is more concentrated. A 2026 audit by Allaham and Diakopoulos of 712 queries across ChatGPT, Copilot, Gemini, and Perplexity found that the four engines cite a narrow set of domains repeatedly, plus a large number of other sites only once or twice.
For your AEO work: For each of your target prompts, list the domains that ChatGPT, Perplexity, and Google’s AI Overviews cite, and group them by type: review sites, forums, news outlets, and vendor pages. The domains that recur across prompts are the sources engines lean on in your category, so they’re the first places to earn coverage.
Structured Data (Schema Markup)
Structured data, often called schema markup, is code you add to a page in a standardized format that tells search engines what the page’s content is. On a recipe page, for example, the markup labels which text is the ingredients, the cooking time, and the calories.
Structured data is often called essential for AI search, but the evidence is mixed. Google’s AI optimization guide says it “isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” Yet Google also says structured data matters for rich-result eligibility, and lists accurate markup among the SEO basics for AI Overviews and AI Mode.
Outside Google, Microsoft has said that schema helps the models behind Copilot understand content.
Tests so far show little effect on citations. In a 2026 Ahrefs study, 1,885 pages that added schema saw no clear change in ChatGPT or AI Mode citations. And in a 2025 searchVIU test, ChatGPT, Claude, Perplexity, Gemini, and AI Mode all missed a product price that appeared only in a page’s schema markup.
FAQ markup is a type of structured data most often recommended for AEO. It labels the questions and answers on a page, but it no longer earns a rich result in Google Search: Google stopped showing FAQ rich results on May 7, 2026, and removed the feature’s documentation in June 2026.
For your AEO work: Run each key page through Google’s Rich Results Test. Check that the markup states the same author, organization, product details, and last-updated date a reader sees on the page, and fix any mismatch.
T
Triple
A triple is a fact stated in three parts: a subject, a predicate, and an object. Knowledge graphs use triples to store the relationships between entities. The W3C’s RDF specification, a standard model for this kind of data, describes RDF graphs as “sets of subject-predicate-object triples.”
Example: “Backed by years of experience, our team helps companies grow through a range of offerings” hides both entities and what the company does. “Animalz provides content marketing services for B2B SaaS companies” states the subject (“Animalz”), predicate (“provides”), and object (“content marketing services”) directly.
For your AEO work: Read the key sentences on your homepage and product pages against the pattern. Each should name both entities and state plainly how they relate.
Z
Zero-Click Visibility
Zero-click visibility is the exposure a brand gets inside an answer, such as an AI Overview or a ChatGPT response, when the user doesn’t click through to the brand’s site. It’s an umbrella term: a brand mention, a recommendation, and a citation nobody clicks all count as zero-click visibility. It’s broader than zero-click search, which describes a search session that ends without a click to the open web.
On Google, users click less when an AI summary appears. In Pew Research Center’s 2025 study of 900 US adults’ Google searches, users clicked a search result link in 8% of visits when a Google AI summary appeared, against 15% when no summary appeared.
Exposure without a click can still lead to a visit later (see Recommendation Rate). Those visits often arrive through a branded search: in a Similarweb study across finance, travel, and beauty, over half of the site visits that followed a ChatGPT recommendation started with one.
You can measure parts of zero-click visibility. Search Console’s Generative AI performance report shows how often links to your site appeared in AI Overviews and AI Mode, and AI visibility tools track brand mentions and citations across assistants.
Example: Asked which agencies help B2B companies with AEO, ChatGPT names Animalz. The user doesn’t click through, but the brand registers. That’s zero-click visibility.
Further reading: AI Overviews Are Eating Your Search Traffic: Here’s How to Adapt
Check a Page against This Glossary
To audit an article, answer these 11 questions. Each “no” points to the glossary entry that explains the fix. For questions 4 and 6, search the article’s main topic in Google first.
Does each section answer its heading’s question in the first sentence? (BLUF)
Does each section’s opening still make sense with the heading hidden? (Extractability)
Are claims backed by named and linked sources, exact figures, or direct quotes? (Citable Facts)
Does the page add a data point, example, or argument the top ten results don’t have? (Information Gain)
Does the article have a byline with real credentials and first-hand experience? (E-E-A-T)
Does the page answer the follow-up questions that Google’s People Also Ask box and AI Mode raise for the topic? (Query Fan-Out, People Also Ask)
Are the facts, figures, and platform features on the page still current? (Content Refresh)
Do the visible date, the schema date, and the sitemap date match? (Freshness Signal)
Does the page use the terms readers search with today? (Semantic Drift)
Does the markup match what a reader sees on the page? (Structured Data)
If answer engines cite the page, does it name your brand and product? (Ghost Citation)
This glossary tests its own advice: each entry opens with a definition an answer engine can lift, names its sources, and gets checked on a schedule for anything that needs a refresh.
Curious whether it works? Ask ChatGPT or Google what one of these terms means and see whose page it cites. If it’s ours, the checklist did its job. If it’s someone else’s, you know which entry we’re fixing next.
Want the same treatment for your content? We run AEO programs for B2B SaaS companies. And we’re always up for a call, whether you want help or just want to debate which of these terms will still be around next year.