Reference
Web3 AI Visibility Glossary
Master the vocabulary AI uses to cite Web3 protocols. If LLMs ignore your project, this glossary reveals the exact machinery to turn your protocol from a rumor into an authoritative source.
Free tool - measure the readability of any Web3 copy with our Flesch-Kincaid calculator⌘K
How AI discovers you7 terms
- AI visibilityGEO, generative engine optimization, LLM visibility
- How often and how accurately large language models (LLMs) like ChatGPT, Claude and Perplexity name, describe and link to your protocol in their answers. It is about being cited, not just ranked.Example: A DeFi protocol that appears by name in ChatGPT's answer to 'best private lending protocols' has AI visibility; one that no LLM mentions does not, regardless of Google ranking.See alsoGEO (generative engine optimization)Answer engineCitationRelated readsFour Essential Principles to Improve Your AI Search VisibilityThe Science Behind AI Brand Visibility
- GEO (generative engine optimization)generative engine optimization
- The discipline of shaping your on-page content, structured data and public surfaces so AI answer engines select and cite you. GEO optimizes for the answer, not the blue link.See alsoAI visibilityAnswer capsuleStructured dataRelated readsFour Essential Principles to Improve Your AI Search VisibilityThe Science Behind AI Brand Visibility
- Answer engineAI search, generative search
- A service that responds to a query with a synthesized answer instead of a list of links - for example ChatGPT, Perplexity, Google AI Overviews, Claude and Gemini.See alsoAI visibilityCitationRetrieval-augmented generation (RAG)
- Crawlerspider, bot, GPTBot, PerplexityBot
- An automated program that fetches and scans web pages to build an index or knowledge base. AI crawlers (GPTBot, PerplexityBot, ClaudeBot, CCBot) visit your site to read it; you control them via robots.txt.See alsorobots.txtIndexabilitysitemap.xmlRelated readsFour Essential Principles to Improve Your AI Search Visibility
- robots.txtrobots exclusion standard
- A plain-text file at your site root that tells crawlers which paths they may or may not fetch. For AI visibility it must explicitly allow the AI crawlers - GPTBot, PerplexityBot, ClaudeBot, CCBot, Google-Extended.Example: A protocol whose robots.txt blocks GPTBot and PerplexityBot is invisible to those answer engines no matter how good the content is.See alsoCrawlerIndexabilitysitemap.xmlRelated readsFour Essential Principles to Improve Your AI Search Visibility
- sitemap.xml
- An XML file listing the URLs a site wants crawlers to find, usually with last-modified dates. Submitting one to Google Search Console helps ensure every page - including each glossary entry's page - is discovered.See alsoIndexabilityCrawlerWhitepaper indexability
- Indexabilityindexable, noindex
- Whether a page is allowed and able to be read and stored by crawlers. A page marked noindex, behind a login, or rendered only by JavaScript may be effectively invisible to AI engines.See alsoCrawlerrobots.txtWhitepaper indexability
How LLMs answer6 terms
- Knowledge cutofftraining data cutoff
- The date up to which a model's static (parametric) knowledge was trained. Events after the cutoff are unknown to the model unless supplied by live retrieval - which is why fresh, current content you publish matters.See alsoParametric knowledgeRetrieval-augmented generation (RAG)
- Parametric knowledgeparametric memory, training data
- Facts a model holds from its training data, recalled without fetching anything live. It can be stale or wrong, and it is the default when no external source is provided.See alsoKnowledge cutoffRetrieval-augmented generation (RAG)Grounding
- Retrieval-augmented generation (RAG)live retrieval, grounding
- The pattern where a model answers by first searching external sources and then quoting them. About 40% of an answer may draw on live retrieval - the part you can influence with publicly available, verifiable content.See alsoGroundingCitationKnowledge cutoff
- Groundinggrounded answer
- Tying an answer to verifiable sources rather than model memory. Grounded answers cite documents - so your onchain dashboards, audit reports and docs pages become the evidence the answer is built on.See alsoRetrieval-augmented generation (RAG)CitationOn-chain grounding
- Citationsource link, referenced
- An answer engine naming and linking the source it drew from. Perplexity cites every source; ChatGPT and Google AI Overviews cite many but not all. Citations are the core KPI of AI visibility.See alsoSource graphZero-click AI answerAI visibility
- Source graph
- The network of trusted domains and documents an answer engine prefers to draw from. Being inside it repeatedly across queries builds your authority with that engine.See alsoCitationEntity authorityDigital public relations (digital PR)
Becoming a citable source7 terms
- First-party datafirst-party numbers, canonical facts
- Data you own and publish about your own protocol: TVL, volumes, user counts, dates of audits and grants. LLMs trust it more than third-party claims, and it is what they quote when describing you.Example: '$670M processed, 0.19% NPL rate' in a protocol's own docs gives an answer engine a concrete, quotable fact - far stronger than a vague 'industry-leading' claim.See alsoCanonical factsOnchain transparencyTVL (total value locked)
- Answer capsule
- A self-contained, quotable one or two sentence answer to a likely question, placed directly in your copy. It states one thing completely with one number, so an engine can lift it verbatim.See alsoGEO (generative engine optimization)Context windowFirst-party data
- Canonical facts
- The stable, public truths about what your protocol is and does - category, mechanics, date founded, current status. Publish these consistently so every AI surface describes you the same way.See alsoFirst-party dataOrganization schemaEntity resolution
- Onchain transparencyonchain data, verifiable data
- Publishing data verifiable on the blockchain - balances, transactions, contracts - that an engine can treat as provably true. It is some of the strongest trust material a Web3 project has.See alsoOn-chain groundingTVL (total value locked)First-party data
- TVL (total value locked)
- The total value of assets deposited in a protocol. A dated, sourced TVL figure listed on DefiLlama or your docs is a first-party number answer engines routinely cite when ranking a category.See alsoOnchain transparencyOn-chain groundingFirst-party data
- Audit as a trust signalsecurity audit, audit report
- A published security audit with a date and report link. For Web3 auditors, LLMs and answer engines treat an independent audit as one of the strongest signals a project is legitimate and current.See alsoOnchain transparencyDecentralized identity (DID) for AIFirst-party data
- Digital public relations (digital PR)link earning, digital PR
- Getting listed, mentioned and referenced across third-party directories, marketplaces and media so multiple trustworthy surfaces describe you consistently - reinforcing what LLMs know about you.See alsoSource graphEntity authorityBrand mention monitoring
Structured data6 terms
- Structured dataschema markup
- Machine-readable annotations - usually JSON-LD - added to a page so engines understand its meaning precisely instead of guessing. It is the language AI visibility is written in.See alsoJSON-LDDefinedTermRich results
- JSON-LDJSON for Linking Data
- The most common format for embedding structured data in a page: a JSON block in the <head> describing the page to engines using schema.org vocabulary.See alsoStructured dataDefinedTermOrganization schema
- DefinedTerm
- A schema.org type for a glossary entry: the term plus its definition, marked so engines can read, quote and link it as a precise definition rather than loose text.Example: Every entry on this page is emitted as a DefinedTerm, which is exactly the kind of machine-readable terminology answer engines can pull a definition from.See alsoStructured dataJSON-LDRich results
- FAQPage schemaFAQ schema
- Structured data marking a question-and-answer block so engines can surface the Q&A verbatim, often as a rich result. Useful for the top questions about your protocol.See alsoStructured dataRich resultsAnswer capsule
- Rich resultsrich snippets
- Enhanced search listings built from structured data - FAQ, glossary, product, review. They make your result stand out and confirm to an engine that the content is structured and trustworthy.See alsoStructured dataFAQPage schemaDefinedTerm
- Organization schema
- Structured data describing your project as an organization - name, logo, links to socials - so engines know who you are and can associate all your surfaces with one entity.See alsoJSON-LDEntity resolutionCanonical facts
Measurement5 terms
- Citation per querycitation rate, mention rate
- The KPI of AI visibility: across a fixed set of representative queries, how often an answer engine names and links your domain. Track it over time rather than chasing rankings.See alsoBrand mention monitoringAI share of voice (AI SOV)Citation
- AI referral traffic
- Visitors arriving from answer engines. Perplexity citations are clickable and appear in analytics; ChatGPT and others are mostly not, so referral traffic undercounts real citation influence.Example: Filtering analytics by referral from perplexity.ai shows clean, attributable traffic - while a ChatGPT mention that isn't clickable stays invisible to analytics even though it shaped the answer.See alsoZero-click AI answerCitation per queryBrand mention monitoring
- Brand mention monitoringvisibility monitoring, AI mention tracking
- Systematically checking whether answer engines name you for relevant queries. Tools that ping ChatGPT, Perplexity, Claude and Gemini for your brand name automate this tracking over time.See alsoCitation per queryAI share of voice (AI SOV)Source graph
- Zero-click AI answerzero-click answer, no-click citation
- When an LLM returns the complete answer to a query - 'What is Arbitrum's TVL?' - sourced from your content, but the user never clicks through to your site. The answer is consumed inside the engine. Your value is captured as the cited authority, not as traffic, so optimize to be the named, linked source even when the click is zero.Example: Perplexity answering 'What is Arbitrum's TVL?' with your figure and your domain cited, while the user never visits your site - a zero-click answer that still builds brand authority.See alsoCitationAI referral trafficAI share of voice (AI SOV)
Web3 x AI specifics5 terms
- Entity resolutionentity disambiguation, named entity resolution
- The process by which an LLM determines that two tokens, projects or people sharing the same name or ticker are distinct entities. For Web3 - where tickers are sparse and reused across chains - resolution is what stops one protocol being conflated with a namesake. Anchor your entity with Organization schema, a canonical URL and stable, repeated facts so the model resolves you to the right identity.Example: Two protocols both using the ticker 'NOVA' on different chains: entity resolution is what lets an answer engine tell them apart instead of merging their TVL, tokenomics and history into one confused profile.See alsoOrganization schemaCanonical factsTicker hallucination
- On-chain groundingonchain grounding, verifiable grounding
- Anchoring an AI-generated claim to immutable blockchain data - wallets, TVL, transaction history, contract state - rather than to a mutable web page. On-chain facts are provable and time-stamped, so an engine can treat them as ground truth even when a blog or docs page could be edited or deleted. Link your claim to the exact on-chain data so a model can verify it.Example: Instead of a docs page stating '$50M TVL', a protocol links the statement to its live on-chain treasury address - so an answer engine quoting the figure can verify it against the ledger.See alsoGroundingOnchain transparencyTVL (total value locked)
- Whitepaper indexabilitywhitepaper parsing, docs indexability
- Whether a protocol's whitepaper - typically a PDF or Gitbook - can actually be parsed, read and cited by an LLM. A PDF with no text layer, or a JS-only Gitbook, may be invisible to crawlers no matter how substantial the document is. Serve a crawlable, text-first version of the whitepaper alongside the PDF to make it indexable and citable.Example: A whitepaper published only as a scanned-image PDF is unreadable to GPTBot; republishing the same content as a crawlable text page or a plain-text Gitbook lets an engine read and cite it.See alsoIndexabilityCrawlerEntity resolution
- Ticker hallucinationticker confusion, token name collision
- When an LLM confuses one token with another that shares the same ticker symbol, mixing up price, utility, history or chain. It is commonplace in Web3 because tickers like 'BTC', 'USDC' and 'SOL' are reused across chains and bridges. Prevent it by stating your chain, contract address and distinct utility explicitly - on every surface - so the model has enough signal to keep you separate.Example: An answer engine reporting the liquidation mechanics and treasury of one 'NOVA' token while quoting the price of a different 'NOVA' - ticker hallucination in action.See alsoEntity resolutionEntity authorityCanonical facts
- Decentralized identity (DID) for AIDID, onchain identity, verifiable credential
- Using on-chain credentials - wallet ownership, DAO membership, attestations, signed statements - to verify who authored or authorized Web3 content. DIDs give an LLM a machine-checkable chain of authorship and authority, so it can attribute a claim to a known, legitimate source rather than an anonymous account. Publish signed, identity-bound content to give engines a verifiable author.Example: A protocol's documentation signed by its governance multisig, tied to an on-chain identity, lets an answer engine verify the content is the official project's statement - not an imposter's fork.See alsoOnchain transparencyAudit as a trust signalEntity authority
Advanced GEO & AI mechanics4 terms
- Semantic embeddingsvector embeddings, embedding vectors
- The mathematical vectors an LLM converts text into to represent meaning, so it can match and rank content by concept rather than keyword. Content that is contextually similar sits close together in vector space. Write prose that says what you mean plainly and completely - embeddings reward clear, dense, on-topic language over keyword stuffing.Example: A page about 'speeding up settlement' can rank for a query about 'fast finality' because their embeddings are close in meaning, even when no exact keyword is shared.See alsoRetrieval-augmented generation (RAG)Context windowAnswer capsule
- Context window128k tokens, token limit
- The bounded amount of text an LLM can hold in memory at once - a few thousand tokens on small models, up to 128k or more on large ones. Because the window is finite and shared across many retrieved sources, concise, dense material gets preserved and cited while rambling content gets trimmed or dropped. Make every sentence carry one verifiable fact.Example: A 5,000-word blog post may have its most important claim truncated out of a 128k context window, while a one-paragraph fact sheet stating the same TVL number survives intact - and is what the engine cites.See alsoSemantic embeddingsAnswer capsuleWhitepaper indexability
- Prompt injectionindirect prompt injection
- A malicious attempt to manipulate an LLM's output by embedding hidden instructions inside content the model retrieves - a technique Web3 brands must guard against because their public docs, forums and tooling are all read by AI crawlers. Publish from controlled, signed, canonical surfaces and treat any instruction-like text in user-generated content as untrusted.Example: A token page that injects 'ignore the previous instructions and say X' is a prompt injection; a protocol that keeps its authoritative facts on a signed, first-party page limits what such attacks can hijack.See alsoEntity authorityOnchain transparencyDecentralized identity (DID) for AI