Skip to content
SeoFreshUp Docs

Glossary

SEO concepts

Core Updates — Major Google algorithm updates published several times per year. They reshape rankings based on E-E-A-T signals. SeoFreshUp tracks the latest official Core Updates feed and uses it in the risk analysis.

Core Web Vitals (CWV) — Three Google performance metrics: LCP (Largest Contentful Paint), INP (Interaction to Next Paint), CLS (Cumulative Layout Shift). Direct ranking factors since 2021.

E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness. Google’s quality framework. Higher E-E-A-T → higher chance of ranking, especially in YMYL niches.

GEO (Generative Engine Optimization) — SEO for AI search engines (ChatGPT, Perplexity, Google AI Overviews). Different from classic SEO: focuses on chunk-friendly content, citability, structured data, llms.txt.

Orphan article — Published article with 0 incoming internal links. Google sees it as isolated. Loses ranking potential.

Dead-end article — Published article with 0 outgoing internal links. Bad UX, doesn’t transmit authority.

Thin content — Articles with too little substance to deserve indexing. Typically <300 words. Penalized by Google’s quality systems.

Cannibalization — Two articles competing for the same search query. Google picks one (often the wrong one) and demotes the others.

Pruning — Deliberately removing low-quality articles to lift the average quality of the remaining ones. Demonstrated SEO benefit (HubSpot’s famous pruning experiment lifted overall traffic by 33 %).

301 / 302 / 410 — HTTP redirect/status codes. SeoFreshUp uses 301 (permanent redirect) when articles are trashed, so visitors and Googlebot land on the parent category instead of a 404.

Canonical — <link rel="canonical"> tag declaring the official URL of a page. Resolves duplicate content issues.

Hreflang — <link rel="alternate" hreflang="..."> tags telling Google which language/region each version of a page is for.

Schema.org / JSON-LD — Structured data markup that helps Google understand the page (Article, Product, FAQPage, Organization, etc.). Often unlocks rich results in SERP.

Plugin-specific terms

Audit — The full pipeline that analyzes every article: traffic + risk + verdict + quality score. Triggered by 📊 Analyse complète.

Verdict — The AI’s recommendation per article: GARDER (Keep), METTRE À JOUR (Update), RETRAVAILLER (Rework), NOINDEX, SUPPRIMER (Delete).

Potentiel — The AI’s estimate of the article’s SEO potential if rewritten well: FORT (Strong), MOYEN (Medium), FAIBLE (Weak), NUL (Zero).

Risque Core Updates — The AI’s estimate of how likely the article is to be hit by an upcoming Core Update: FORT, MOYEN, FAIBLE, NUL.

Score qualité (Quality %) — Composite 0-100 score combining all signals. 100 = perfect, 0 = critical.

Maillage — Internal mesh / internal link graph. The set of <a> tags between your published articles.

Sortants — Outbound links. <a> tags pointing to other domains.

BYOK (Bring Your Own Key) — Pricing model where the customer pays the AI provider (OpenRouter, OpenAI, etc.) directly with their own API key, rather than the plugin reselling AI calls with a markup.

Skill (custom) — A Markdown text snippet (max 32 KB) injected into the AI prompt to teach the model your editorial charter, vocabulary, or constraints.

Info-card — A generated image displayed inline in an article body, showing key data scraped from an external URL (pricing table, feature list). Different from the featured image (which goes at the top).

Planificateur (Scheduler) — The queue system for bulk rewrites + auto-publish at editorial pace. Hard-capped at 2 publications/day.

Drift baseline — A snapshot of SEO-critical elements (titles, meta, schema, headings, content) used to detect changes/regressions over time. Like git for SEO.

AI / LLM terms

LLM (Large Language Model) — The AI behind text generation. Examples: Claude, GPT, Gemini, Llama.

Token — Roughly 4 characters of text. Pricing is per-token. A 1500-word article ≈ 2000 tokens.

Prompt — The instructions sent to the LLM. SeoFreshUp builds detailed prompts behind the scenes (article content + your skill + structured output schema).

Context window — Maximum tokens the LLM can read in one call. gpt-4o-mini: 128k. Claude Haiku 4.5: 200k. Gemini 2.5 Pro: 2M.

Temperature — Randomness control (0 = deterministic, 1 = creative). SeoFreshUp uses 0.1 for analyses (factual) and 0.6 for rewrites (some creativity).

Multimodal — A model that accepts text + images as input. Used by SeoFreshUp’s image gen feature to pass the source screenshot (Firecrawl) as visual reference.

OpenRouter — A unified API gateway that aggregates Claude, GPT, Gemini, Llama, etc. behind a single key + a single API. Reduces vendor lock-in.

Firecrawl — A web scraping API that handles JavaScript rendering (so it can scrape SPA sites that vanilla cURL can’t). Used by SeoFreshUp for URL enrichment in rewrites.

JSON Schema (response_format) — A way to force the LLM to return structured JSON matching a specific schema. SeoFreshUp uses it for risk + verdict analyses to guarantee parseable responses.

Embedding — A vector representation of text used for semantic search and RAG. Not currently used by SeoFreshUp (planned for the docs chatbot).

RAG (Retrieval-Augmented Generation) — A pattern where you pre-index your knowledge base into vectors, retrieve relevant chunks at query time, then feed them to an LLM. Planned for the SeoFreshUp docs chatbot.

What’s next?