Glossary
The AEO and retrieval glossary
The terms behind Seeklyn’s analyzers and citation checks, defined in plain English and matched to the code that produces them. AEO, BM25, the citation funnel, query fan-out — each term links to a full explainer with the mechanism, the evidence grade, and where it appears in the product. The AI Search Intelligence metrics (mention rate, citation share of voice, source position and more) are defined in the dashboard’s ⓘ tips and on its Methodology page.
Concepts
- Answer Engine Optimization (AEO)Answer Engine Optimization (AEO) is a 6-lever practice for structuring content so AI answer engines — ChatGPT, Google AI Overviews, Perplexity, and Gemini — cite it as a source rather than just rank a link.
- Generative Engine Optimization (GEO)Generative Engine Optimization (GEO) is a 6-lever practice for structuring content so AI answer engines — ChatGPT, Google AI Overviews, Perplexity, and Gemini — cite it as a source in their synthesized answers.
- Answer engineAn answer engine synthesizes a direct answer from multiple web sources and cites them inline, instead of returning a ranked list of 10 blue links.
- Query fan-outQuery fan-out is the process by which an AI answer engine silently expands 1 user prompt into multiple sub-queries before retrieving passages.
- llms.txtAhrefs (2026) studied 137,000 sites and found 97% of sites that added an llms.txt file saw zero change in AI-bot traffic — no measurable citation effect.
- Schema markup and AI citationsAhrefs (2026) measured the effect of schema markup (JSON-LD) on AI citation rates across a large sample and found it null: the measured range was −4.6% to +2.4%, straddling zero.
- SEO vs AEOSEO (Search Engine Optimization) earns a ranked link in Google's 10 blue links; AEO (Answer Engine Optimization) earns a citation inside an AI-generated answer from engines like ChatGPT, Perplexity, or Google AI Overviews.
- Zero-click searchA zero-click search is a search session that ends on the results page — roughly 95% of queries fanned out by AI prompts carry zero traditional search volume according to Seer Interactive (2026).
- Retrieval-Augmented Generation (RAG)Retrieval-Augmented Generation (RAG) is a 2-step method used by AI answer engines: first, retrieve the most relevant passages from an index (typically using BM25 keyword matching, dense vector search, or both); second, generate a grounded answer that cites those passages.
- LLM visibilityLLM visibility is the sum of 2 layers: retrievability (whether the engine's index surfaces your page) and citability (whether your passages are structured to be lifted as standalone answers).
Content analyzer
- Citation funnel (Retrieval → Citation → Overall)The citation funnel is Seeklyn's 3-stage content score, each rated 0–100: Retrieval measures whether a retrieval index can find your passages for the target query, Citation measures whether the retrieved passages are actually quotable, and Overall is the mean of the two.
- Passage citabilityA citable passage answers a question in its first sentence, stands alone without prior context, runs 100–300 words, and carries at least 1 concrete fact with a named source.
- BM25 retrievalBM25 (Best Match 25, developed by Robertson & Spärck Jones, 1994) is the lexical ranking function at the core of production search indexes — including those used by AI answer engines.
- Content chunking (passages)Content chunking is the process of splitting a page into 100–300-word passages so each one can be individually retrieved and scored.
- Passage extractability (the 8 flags)Passage extractability is Seeklyn's 8-flag rubric for measuring whether an AI engine can lift a passage verbatim as a self-contained answer.
- Retrieval (AEO)Retrieval is the first of 3 stages in Seeklyn's citation funnel — a 0–100 score measuring whether your passages surface for the target query using BM25 keyword matching.
- Anaphoric opening (ANAPHORA_OPEN)8 trigger words — it, this, that, these, those, they, such, and “as mentioned” — fire the ANAPHORA_OPEN flag when they open a sentence.
- Answer-first structure (NO_ANSWER_FIRST)Answer-first structure places a direct, concrete answer in the first sentence of a passage — the first sentence must be declarative, non-anaphoric, hedge-free, and contain at least 1 fact token (digit, percent, unit, or 4-digit year).
- Hedging language (HEDGE)Seeklyn's HEDGE flag fires on 6 epistemic qualifiers when any of them appears as a word token in a sentence, marking that sentence's verdict as "hedged" rather than "liftable".
- Run-on sentence (RUN_ON)A sentence exceeding 30 words (strings.Fields count) fires Seeklyn's RUN_ON flag, because packing multiple ideas into 1 sentence prevents AI engines from extracting a single atomic fact.
- Fact density (LOW_FACT_DENSITY)Seeklyn's LOW_FACT_DENSITY flag fires when fewer than 1 fact token appears per 100 words — a fact token being any digit, decimal, percentage, 4-digit year, or dollar amount matched by the pattern \b(\d+([.,]\d+)?%?|\d{4}|percent|\$\d+)\b.
- Source attribution (NO_ATTRIBUTION)Source attribution means placing a named source — or at minimum a 4-digit year — in the same passage as any numeric or factual claim.
- Entity density (ENTITY_SPARSE)Seeklyn fires ENTITY_SPARSE when a passage of 50 or more words names fewer than 2 distinct entities — brands, people, products, places, and organizations.
- Keyword stuffing (KEYWORD_STUFFING)Seeklyn fires KEYWORD_STUFFING on passages of 20 or more words when any content word ≥4 chars exceeds 8% of tokens, unique-token ratio drops below 50%, or a target keyword exceeds 6% of tokens — whichever fires first.
- Evidence gradeEvery rule and flag in Seeklyn carries 1 of 3 evidence grades — strong (peer-reviewed research or official search-engine documentation), moderate (correlational studies or vendor data), or weak (plausible but unconfirmed).
- Evidence gapSeeklyn detects up to 6 evidence gaps per passage across 5 gap types — vague, comparative, freshness, hedge, and source — and emits a typed, actionable prompt for each.
- Crawl gateThe crawl gate is 3 hard prerequisites every page must pass before an AI answer engine can cite it.
- noindex / nosnippetnoindex and nosnippet are 2 meta robots directives with opposite scopes: noindex removes a page from all search engines entirely; nosnippet blocks only the text extracts that engines show and AI systems quote.
- AI crawlersSeeklyn checks for 14 AI crawler user-agents: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-Web, anthropic-ai, PerplexityBot, Perplexity-User, Google-Extended, Applebot-Extended, CCBot, Bytespider, Amazonbot, and Meta-ExternalAgent.
Headline analyzer
- Headline CTR ScoreSeeklyn's headline CTR score (0–100) estimates how well a headline earns the click by blending 5 sub-scores: word balance 34%, headline type 20%, length 20%, sentiment 13%, and readability 13%.
- Headline ScoreA headline score is a single 0–100 number that equals 0.5 × CTR score + 0.5 × AEO score, rounded to the nearest integer.
- Word BalanceWord balance is a 0–100 sub-score equal to the mean of 4 vocabulary bucket scores: common words (target 20–30% of tokens), uncommon words (10–20%), emotional words (10–15%), and power words (target 1–3 count).
- Power WordsSeeklyn targets 1–3 power words per headline: 0 scores 0 (no persuasive signal), 1–3 scores 100, and more than 3 scores only 30 — a clickbait penalty that rewards restraint.
- Emotional WordsEmotional words are headline tokens whose LabMT hedonometer valence falls below 4.0 (negative end) or above 6.0 (positive end).
- Headline TypeHeadline type is the structural format class Seeklyn assigns to a title — how-to, list, question, superlative, you-directed, or generic — and each type carries a 0–100 CTR sub-score.
- Headline LengthIdeal headline length is 6–9 words and 50–60 characters.
- Headline Sentiment AnalysisHeadline sentiment analysis sums AFINN lexicon valences across headline tokens to produce a polarity label (positive, negative, or neutral) and a 0–100 strength score.
- SkimmabilitySkimmability is a binary headline check: the primary topic — target keyword or first recognised entity — must fall in the first or last 3 tokens of the title.
- Headline clarityHeadline clarity is a 3-condition check: a headline is "clear" when it contains 12 words or fewer, reads at Flesch-Kincaid grade 12 or below, and uses 3 or fewer power words.
- Clickbait (headline honesty check)Seeklyn's clickbait check detects 2 specific honesty-eroding markers: an ALL-CAPS alphabetic word of 4 or more letters (e.g.
- AEO Readiness ScoreThe AEO Readiness Score is Seeklyn's headline panel that awards a letter grade (A+ for ≥85, A for ≥75, B for ≥65, C otherwise) and a modelled percentile estimate (formula: min(99, max(1, round(0.85 × score + 14)))) derived from the overall headline score.
- Flesch-Kincaid Reading GradeFlesch-Kincaid grade is a readability formula that estimates the US school grade level needed to understand a piece of text: FK grade = 0.39 × (words ÷ sentences) + 11.8 × (syllables ÷ words) − 15.59.
- Google Title RewriteGoogle rewrites roughly 13–20% of SERP titles (Google Search Central, 2021).
- Safer title variantA safer title variant is the trimmed alternative Seeklyn generates when a headline exceeds 60 characters — the threshold above which Google commonly rewrites SERP titles.
- YouTube Title LengthThe best YouTube title length is about 30–50 characters.
- YouTube Title TruncationYouTube title truncation is where the interface visually cuts a title with an ellipsis (…) when it runs past the space a surface allows — roughly 40–50 characters on mobile feeds, search, and TV, and about 60 on desktop search.
- Keyword Front-LoadingKeyword front-loading is placing your primary keyword or hook in the first 3–5 words — inside the first ~40 characters — so it survives truncation and is the first thing a viewer and a ranking system read.
- Emoji in TitlesEmoji in titles significantly reduce long-form YouTube performance.
- Google RSA 30-Character HeadlineA Google Responsive Search Ad (RSA) accepts 3 to 15 headline assets, and each headline is capped at exactly 30 characters — 15 in double-width languages like Japanese, Korean, or Chinese, where each character counts as two.
- Ad Headline TruncationAd headline truncation is where a platform accepts a longer headline than it actually shows, then clips the rest at render time.
- CTA in Ad HeadlineA CTA in an ad headline is an explicit action verb — Get, Start, Save, Try, Download — placed in the headline itself rather than only on the button, telling the reader exactly what to do next.
Am I cited?
- Am I Cited? (observed vs inferred)Seeklyn checks AI citations by reading the actual AI answers and the sources they cite, tracked in real time across ChatGPT, Google AI Overviews, Perplexity and more.
- Citation likelihood (inferred)Citation likelihood is Seeklyn's forecast of how likely a page is to be cited by AI answer engines, a score between 0 and 1 expressed in 3 bands: likely (≥0.65), possible (≥0.35), and unlikely (<0.35).
- Citation match levels (exact, domain, none)Seeklyn classifies every AI-engine citation check at 3 match levels: exact (same normalized URL), domain (same registrable domain, different page), and none (no presence).
- Citation rankCitation rank is the 1-based position of your page in an AI engine's cited-sources list: rank 1 means your page was the first source listed, rank 2 the second, and so on.
- Brand mention (answer-text visibility)Seeklyn tracks 4 brand-visibility states per engine check: cited + mentioned, cited only, mentioned only, and neither.
- Competitor citationsCompetitor citations are the domains an AI engine cites for the same query — sources that appeared in the engine's top-5 reference list while your page was absent, or that ranked above you when present.