To guide you through the world of web performance, we share our own definitions of technical jargon here in French. They differ from what you find in most online glossaries and lexicons for a good reason: we've aimed to make them as understandable as possible for newcomers. Happy reading!
A
A/B Testing
Tool that allows comparing the effectiveness of several versions of the same web page on real users, particularly regarding conversion rates. Its technical complexity has led to the emergence of specialized players whose solutions are paid: AB Tasty, Kameleoon… On the web performance side, these solutions are worth watching: their scripts, often blocking to avoid any "flicker" effect (fleeting display of the original version), can significantly delay page rendering.
Accessibility
Refers to the ability of a website to be optimally usable by all users regardless of their browser, location, or physical and psychological abilities. The WCAG guidelines, which are the standard, focus on people with disabilities and seniors, who often experience difficulties navigating the internet.
AEO
Answer Engine Optimization: optimizing content for search engine direct answer formats, such as featured snippets, "Questions people also ask" blocks, or voice search. The goal is no longer to get a click, but to be the answer: clearly posed questions, followed by concise answers, increase the chances of appearing.
AI Agent
A program based on a language model that doesn't just respond: it acts on behalf of a user by browsing web pages, calling tools, or filling out forms. Code assistants (Claude Code, Cursor) and "agentic" browsers are examples. For a site, this means new non-human visitors, sensitive to the clarity of markup, accessibility, and page stability: an agent navigates the accessibility tree like a screen reader.
AI Overviews
AI-generated answers from Google, displayed at the top of some results pages and accompanied by links to the sources used. They reduce the number of clicks on informational queries. Google indicates that no specific file or markup is necessary to appear: the fundamentals of SEO (useful content, crawlable and fast pages, structured data) are what matter. See also GEO and AEO.
Ajax
Acronym for "Asynchronous JavaScript and XML," it is a technology that allows a web browser to exchange data with a server without having to reload the current page. It is the basis for tools like Infinite Scroll, but also for SPAs that function like an application.
API
Acronym for "Application Programming Interface," it refers to the ability for one application to give access to all or part of its functionalities to a third-party application. The goal is to facilitate its use and integration into an existing workflow. For example, you watch series on your smart TV thanks to Netflix's API.
APM
Acronym for "Application Performance Management," it refers to a tool installed on a web server to monitor its performance. To do this, it connects to key components of the operating system (processor, RAM, system kernel, etc.) and specific applications (SQL databases, web server, PHP executable, cache system, etc.).
ARIA
Acronym for "Accessible Rich Internet Applications," it refers to markup intended to add an additional semantic layer to web pages. By supplementing HTML tags with specific attributes, ARIA informs browsers and especially assistive technologies (screen readers in particular) about the role and expected behavior of each interface element.
Asynchronous
Asynchronous loading is loading that occurs in the background, without interrupting the browser's processing of the rest of the page. On a script tag, the async attribute allows downloading in parallel with HTML parsing, and then execution as soon as the file is ready. It is the opposite of synchronous loading.
AVIF
A modern image format, derived from the AV1 video codec, which offers significantly smaller files than JPEG and often than WebP at equal quality. It is supported by all recent browsers. Serving images in AVIF, with a fallback to WebP or JPEG if necessary, is one of the simplest levers for improving LCP.
B
Back-end
Part of a site accessible only to the site owners and managers, as opposed to the front-end which is accessible to all internet users. This concerns administration and all scripts and tools executed by the web server.
Base64
An encoding system that allows data, often binary, to be represented as text composed of 64 ASCII characters. Content encoded this way is on average 33% larger. On the web, it is mainly used to embed resources inline (images, fonts, etc.).
bfcache
Short for “back/forward cache.” Browser mechanism that keeps a full copy of visited pages in memory: when a visitor uses the back/forward buttons, the page reloads instantly, without any reloading. Certain technical practices (HTTP headers, scripts, etc.) can deprive a site of this valuable mechanism.
Blocking Time
Period during which a web page is no longer interactive due to excessive use of the device's resources by JavaScript. This results in freezes and/or excessive latency.
Brotli
Modern compression algorithm used by web servers to reduce the size of text-encoded resources (HTML, CSS, JavaScript, etc.), and thus their download time by browsers. Newer than Gzip, it offers a better compression ratio. Its compression is relatively slow, making it particularly suitable for static resources (CSS, JavaScript, fonts, etc.) that can be compressed once and for all, and less so for dynamic content.
C
Cache
Principle of keeping a ready-to-use copy of a content that is costly to produce, in order to serve it faster on subsequent occasions. It is found at all levels of the web: browser cache (which avoids re-downloading resources), HTML page cache on the server side (which avoids regenerating each page on each visit), object cache like Redis (which relieves the database), or even CDN cache. It is one of the most powerful levers for web performance, with TTFB at the forefront.
Cache-Control
HTTP header that tells browsers, CDNs, and proxies whether a resource can be cached, for how long (max-age, s-maxage), and under what conditions (no-cache, private, stale-while-revalidate). It is what decides whether a returning visitor re-downloads the stylesheet or retrieves it from their cache. An incorrect value costs either performance (re-downloaded resources) or freshness (stale content served).
Object Cache
Cache that keeps in memory (Redis, Memcached) the results of database queries and recurring calculations of an application like WordPress. It speeds up pages that cannot be cached entirely: shopping cart, customer area, administration.
CDN
Acronym for “Content Delivery Network,” this tool interfaces between the hosting and the visitors' browsers to improve a site's loading times. It relies in particular on copying the static resources from the origin server to each of the provider's data centers around the world. Cloudflare is the most popular. Learn more about CDNs.
Critical Rendering Path
Translation of "Critical Rendering Path": the set of essential steps and resources needed by the browser to display the first screen of a page (blocking HTML, CSS, and JavaScript in particular). The shorter and lighter this path is, the faster the page appears (FCP, LCP). Optimizing it involves reducing, lightening, or deferring anything not essential for this initial display.
CLS
Acronym for "Cumulative Layout Shift", this performance indicator measures a page's ability to provide an interface whose elements remain visually stable over time. It is one of the 3 Core Web Vitals. A good CLS is below 0.1, a bad one above 0.25.
CMP
Acronym for "Consent Management Platform". This is the tool that allows website owners to collect visitors' consent during their first visit regarding tracking and cookie placement. It is an essential tool for complying with GDPR on any site that processes personal data. Feel free to consult our CMP performance comparison.
CMS
Acronym for "Content Management System", it is a tool that allows a website and its content to be managed through an online administration interface. The best known is WordPress, but there are hundreds, sometimes specialized in certain types of content, like PrestaShop and Magento for e-commerce.
Code splitting
Splitting a site's JavaScript into multiple files loaded on demand, page by page or feature by feature, rather than into one large file. It reduces the amount of script downloaded and executed on the first load, benefiting TBT and INP.
Concatenation
A technique for optimizing loading times that involves grouping multiple CSS or JavaScript files into a single one. Commonly used to reduce the number of HTTP requests for a page, this technique has lost some of its relevance since HTTP/2, which multiplexes requests. It can even hinder caching granularity. Synonym: combination.
Content-Signal
Directive to be placed in the robots.txt file to indicate the permitted uses of a site's content by AI robots: indexing for search, responding to a user's query, training models. Proposed by Cloudflare, it does not technically block anything: it is a statement of intent, like the rest of robots.txt. For example, it allows a media outlet to refuse training while remaining visible in assistant responses.
Core Web Vitals
Core Web Vitals, or Essential Web Signals in French, group 3 metrics intended to measure a page's performance in terms of speed and responsiveness: LCP, INP, and CLS, with INP having replaced FID in March 2024. They are part of the Web Vitals, which include other measures like FCP or TTFB. These are the performance indicators taken into account by Google in its algorithm and its tools: PageSpeed Insights, Lighthouse, etc.
Crawl budget
Refers to the volume of resources that Google allocates to crawling a site: number of pages crawled and frequency of visits by its indexing bots. The faster a site responds (especially TTFB), the more effectively this budget is used: more pages are crawled, and therefore indexed. In French, we speak of "budget d’exploration".
CrUX
Acronym for "Chrome User Experience Report", this tool developed by Google allows you to know the real performance level of a page or a site. To do this, it collects performance indicators, including Core Web Vitals, during visitors' navigation on browsers that support it. The field data thus collected is accessible via Search Console and in the "Discover your users' experience" section of the PageSpeed Insights reports.
CSP
Acronym for "Content Security Policy", which translates in French to "politique de sécurité des contenus". It is an http header that aims to tell the browser the authorized origins for different types of resources: CSS, JavaScript, images, iframes, etc. Implementing a CSP improves a site's security but also allows for an audit that is very useful for web performance.
CSR
Acronym for "Client Side Rendering", which translates in French to "rendu côté client". In the SPA ecosystem, this means that the JavaScript framework is responsible for rendering the components of a web application in the browser. This practice has a major impact on performance, both in terms of speed and responsiveness. The alternative technique is SSR.
CSS
Acronym for "Cascading Style Sheets", which translates in French to "feuilles de style en cascade". It is the language used by web browsers to format pages in HTML and XML: display of text, position and color of elements, animations... It is overseen by the W3C.
Critical CSS
Style sheet portion strictly necessary for displaying what is visible above the fold. Integrated directly into the HTML (inline), it allows for an initial render without waiting for the complete style sheets to download, which are then loaded without blocking. It improves FCP and LCP, provided it is kept up-to-date with each design change.
D
Accessibility Statement
Mandatory page for sites subject to the accessibility obligation, indicating the compliance level with RGAA (fully, partially, or non-compliant), the percentage of criteria met, inaccessible content, a contact method, and recourse options. It is based on an audit and updated after each significant site change.
DNS
Acronym for "Domain Name System," this protocol is used by browsers to direct visitors of a page to the appropriate server. It associates each domain name with an IP address using data provided by its owner (DNS records). A global cache infrastructure ensures its responsiveness but, in return, generates propagation delays in case of server changes.
DOM
The DOM (Document Object Model) is the browser's representation of a page's structure, with each HTML tag becoming a "node." It evolves dynamically based on node additions, deletions, and modifications made in JavaScript.
Field data and lab data
Two ways to measure performance. Field data (RUM, CrUX) comes from real visitors, with their devices and connections: this is what Google uses for Core Web Vitals. Lab data (Lighthouse, PageSpeed Insights, WebPageTest) comes from a simulated test under fixed conditions: reproducible, it is used to diagnose and verify a fix, but does not always reflect the real experience.
E
EAA (European Accessibility Act)
European directive on the accessibility of products and services, applicable in France since June 28, 2025. It extends the digital accessibility obligation to many private companies, particularly e-commerce sites and banking services, beyond the public sector alone.
Early Hints (103)
Intermediate HTTP response, code 103, that the server or CDN sends before the page itself to tell the browser which resources to preload or which connections to open. The browser takes advantage of the page generation time to start downloading critical resources: a direct gain on LCP when TTFB is high.
Web eco-design
Approach aimed at reducing the environmental impact of a digital service from its design stage: lighter pages, fewer requests and scripts, suitable hosting, truly useful features. It largely aligns with web performance. In France, it is based on the RGESN.
Edge computing
Execution of code or caching as close as possible to the user, within a CDN's points of presence, rather than on the origin server. It allows pages, redirects, or personalized variations to be served in milliseconds, regardless of the distance to the server.
E-E-A-T
Experience, Expertise, Authoritativeness, Trustworthiness: a quality evaluation framework used by Google to judge content and its author. It is particularly important for expertise, health, or finance content: for the same text, an article signed, sourced, and published on an authoritative site will be better regarded than anonymous content.
Embed
Translates to French as "embedded content," and materializes within a web page by including content from a third-party site. Social networks like Instagram, X (formerly Twitter), or YouTube all offer these types of integrations.
F
Fallback
A fallback can be translated into French as "solution of last resort" or "by default." On the web, this consists of offering an alternative in case a feature or resource format is not supported by the browser.
FCP
Acronym for "First Contentful Paint," this performance indicator measures the time it takes for the first visible element of the page to appear (image, text, etc.). It is heavily impacted by any resources that block page rendering (scripts and stylesheets).
Context window
Amount of text, measured in tokens, that a language model can process at once: the question, the provided documents, and its answer. A web page heavy with markup, scripts, or repeated menus consumes this window at the expense of useful content, hence the benefit of serving agents a lightweight version of the page (see Markdown).
Fetchpriority
HTML attribute that allows for prioritizing (value "high") or de-prioritizing (value "low") resource loading in the waterfall. It is the most effective way to modify the browser's default behavior, preferable to preloads. Applied to an image that constitutes the LCP, it improves the metric by an average of 15%.
FID
Acronym for "First Input Delay," this performance metric measured the delay between a user's first interaction (click, typing, etc.) and the start of its processing by the browser. It was part of the Core Web Vitals until the introduction of INP in March 2024. A good FID was under 100 ms, a bad one over 300 ms.
Breadcrumb
Visual representation of the current page's position within a site. Each element in the hierarchy is clickable to allow a quick return to the homepage and parent pages (category(ies) for an article, for example). In English, it's called "breadcrumb."
font-display
CSS property that controls the behavior of a web font during its loading: immediately display a fallback font and then replace it (swap), hide the text for a short time (block), or abandon the font if it arrives too late (optional). It balances invisible text (FOIT), visible font changes (FOUT), and layout shifts (CLS).
FOUT and FOIT
Two effects of web font loading. FOUT ("Flash of Unstyled Text"): the text first displays in a fallback font, then changes appearance, which can cause a shift (CLS). FOIT ("Flash of Invisible Text"): the text remains invisible until the font arrives, delaying reading and sometimes LCP. A fallback font with adjusted metrics reduces both.
Front-end
The front-end refers to the part of a site executed in the visitor's browser: the HTML, stylesheets, and JavaScript that produce the page as it is displayed. It is the opposite of the back-end, executed on the server, which generates data and pages before transmitting them.
FTP
Acronym for "File Transfer Protocol," this protocol allows data to be transferred to a web server's storage space: images, PHP files, etc. It is only accessible to the site owner through login credentials. It is now largely replaced by its secure variant SFTP, which encrypts exchanges via SSH.
Full stack
Refers to a developer who masters technical environments on the server side (PHP language, SQL syntax, system administration, etc.) and on the browser side (HTML, CSS, JavaScript, etc.). These profiles are rare and sought after by many companies for their great versatility and ability to work on both the front-end and back-end.
G
GEO
Generative Engine Optimization: optimizing content for generative engines like Google's AI Overviews, ChatGPT Search, Perplexity, or Gemini. These engines do not rank pages; they select entities and facts from their sources to compose a summary response. Being cited therefore requires factual, structured, and easily attributable content, even more so than good ranking.
Gzip
Compression algorithm historically used by web servers to reduce the size of text-encoded resources (HTML, CSS, JavaScript, etc.), and thus their download time by browsers.
H
Headless (Headless CMS)
Architecture in which the CMS only manages content and exposes it via an API, with the display being handled by a separate application (Next.js, Nuxt, Astro, etc.). It offers great technical freedom but shifts the responsibility for performance, SEO, and accessibility to this application, which is often more complex to maintain.
.htaccess
System file consulted by Apache and LiteSpeed web servers before serving a page or resource. It is commonly used to implement URL rewriting, manage 301 redirects, add security headers, or optimize browser caching of resources.
hreflang
Attribute declaring the linguistic equivalents of a page ('this page also exists in English, at this address'), so that the search engine serves the correct version to the right audience. Without it, the different languages of the same content can compete with each other in the results, or even be displayed to the wrong visitor.
HTML
Acronym for 'Hypertext Markup Language,' it is the foundational language of every web page. It consists of multiple tags, each with its own role: paragraphs, images, tables, sections, etc. Attributes allow developers to modify their behavior as needed.
HTTP
Acronym for 'Hypertext Transfer Protocol,' it is the communication language between browsers and web servers: the former requests a resource, the latter delivers it. Its version 2 (2015) introduced request multiplexing, meaning the ability to process multiple requests simultaneously over the same connection. Version 3, based on the QUIC protocol, further speeds up connection establishment, especially on mobile networks.
HTTP/2 and HTTP/3
Recent versions of the HTTP protocol. HTTP/2 transmits all resources of a page over a single connection, in parallel, with compressed headers. HTTP/3 is based on QUIC, over UDP: it establishes connections faster and is more resilient to packet loss, especially on mobile. Both are generally enabled by recent CDNs and hosting providers.
Hydration
Step in the loading of a SPA during which JavaScript makes the components already displayed on the page interactive, usually after server-side rendering (SSR). Resource-intensive in terms of CPU, it is one of the main performance bottlenecks for web applications, particularly INP.
I
Generative AI
Family of artificial intelligence models capable of producing text, images, code, or sound from a prompt. ChatGPT, Claude, Gemini, and Mistral are based on generative language models. Their use as a response engine is changing how sites are discovered: this is the subject of GEO.
Icon Font
Set of icons grouped within a Webfont in the form of characters. The format is very popular because it offers vector rendering and is easy to use, but it requires a significant volume of CSS and Webfonts in return. The most popular is Font Awesome.
Iframe
HTML element that allows displaying a web page within another web page in a space with defined dimensions. It is the preferred solution for displaying most advertising banners and social media embeds.
Responsive Images
Technique that serves each device an image adapted to the actual size of its display, thanks to the srcset and sizes attributes or the picture element. A smartphone no longer downloads the image intended for a desktop screen: a significant gain in page weight and mobile LCP.
Infinite Scrolling
UI element alternative to traditional multi-page pagination. With "infinite scroll," the next contents in the list are dynamically loaded into the current page as it scrolls. Ajax technology is used for this.
Inline
Code referred to as "inline" is inserted directly within the original page, as opposed to being called from an external file. In HTML, the most common inline tags are <style></style> for CSS, <script></script> for JavaScript, or <svg></svg> for images. It is also possible to use base64 encoding to embed any resource inline.
INP
Acronym for "Interaction to Next Paint," this performance indicator measures a page's responsiveness following a user action such as a click, touch, or keyboard input. It was integrated into the Core Web Vitals in March 2024 to provide more comprehensive insight into latencies, particularly those related to JavaScript. A good INP should be less than 200 ms, and it is considered poor beyond 500 ms.
J
JavaScript
Programming language used by website publishers to make their pages more dynamic, add an interactivity layer, display advertisements, or track their traffic. Code interpretation is costly for browsers: it often generates "Blocking Time" and degrades the INP.
Token
Text segmentation unit by a language model: a common word, a word part, or punctuation. Models are limited and billed by tokens. A full HTML page often contains several times more tokens than its text alone: converting it to Markdown for an AI agent saves a large portion of these tokens.
JSON
Acronym for "JavaScript Object Notation," it refers to a syntax for exchanging data between different tools. Its great simplicity makes it easy to generate and read by computer languages, including PHP and JavaScript. On the internet, it is widely used by APIs and tools for website publishers.
Is your site as fast as your visitors expect?
JSON-LD
Format recommended by Google for structured data: a script block, invisible to the visitor, that describes the page content according to the Schema.org vocabulary (organization, article, product, FAQ, breadcrumb trail, etc.). It allows search engines and AIs to unambiguously understand who publishes what. Unlike microdata, it does not interfere with the page's HTML: it is easier to maintain and weighs only a few kilobytes.
L
Lazy-loading
Lazy-loading refers to triggering the loading of certain page resources (images, videos, embedded content, etc.) only when visitors are likely to see them. It is implemented using JavaScript tools or the native HTML attribute "loading=lazy".
Lazy-rendering
Technique that excludes part of the DOM from the initial rendering of a page using the CSS property content-visibility. Rendering times are thus reduced, ensuring gains in speed (FCP, LCP, etc.) and potentially interactivity (reduction of Blocking Time). The only constraint is being able to anticipate the height of the excluded elements.
LCP
Acronym for "Largest Contentful Paint," this performance indicator measures the time it takes for the main visible element of the page to load. It is one of the 3 Core Web Vitals. A good LCP is under 2.5 seconds, a bad one over 4 seconds.
Lighthouse
Web audit tool developed by Google's teams, focusing on 4 aspects: Performance, Best Practices, Accessibility, and SEO. Any page can be tested to get its score (from 0 to 100), on desktop and mobile. The latter mode simulates degraded network and processing power access conditions: scores are generally lower.
Fold line
The lower boundary of the area of a page visible on the screen without scrolling ("above the fold" in English, a legacy from print media). Everything displayed above should be loaded with priority, as it is generally where the LCP element is located; what is below can be deferred, notably through lazy-loading.
LLM (large language model)
From English “Large Language Model”: an artificial intelligence model trained on very large volumes of text to understand and produce language. GPT (OpenAI), Claude (Anthropic), Gemini (Google), Llama (Meta), or Mistral are LLMs. They power conversational assistants, response engines, and AI agents.
llms.txt
A Markdown file placed at the root of a site (/llms.txt), inspired by robots.txt, intended to present language models with a structured summary of the site and its main pages. It is not a standard but a proposal: Google Search does not use it, and in practice, AI bots rarely request it. Lighthouse checks it when it exists (a title, links, a response in less than two seconds) without penalizing its absence. It is mainly useful for technical documentation; it does not replace structured data or a solid “About” page.
Long Tasks
Translated into French as “long tasks,” they refer to JavaScript executions that monopolize the processor for more than 50 milliseconds at a time. During this time, the browser cannot respond to any interaction: the page appears frozen. Identifying and splitting them is a major lever for improving INP.
M
Main thread
The browser's single execution thread that runs JavaScript, calculates styles and layout, and responds to interactions. When it is busy with a long task, the page no longer responds to clicks or input: this is what TBT and INP measure. Lightening, splitting, or offloading JavaScript frees up the main thread.
Markdown
A lightweight markup language that formats text with a few symbols (# for a heading, * for a list, [text](link) for a link) instead of HTML tags. It has become the preferred reading format for language models: an AI agent requesting a page in Markdown (Accept header: text/markdown) receives the content without menus, scripts, or styles, which is 80% to over 99% fewer tokens. A site can serve this version through content negotiation, without duplicating its pages.
MCP (Model Context Protocol)
An open protocol published by Anthropic in late 2024 that standardizes how an AI assistant connects to external tools and data sources: an “MCP server” exposes functions (read a document, query a database, run a PageSpeed test…) that the assistant can call itself. Since then, adopted by most AI publishers, it allows an agent to act on a service without custom integration. For a site or application, offering an MCP server is equivalent to providing an API designed for agents.
Media Queries
Media Queries are a CSS tool that allows you to target specific devices and browsers based on their characteristics. They are used notably to create "Responsive Web Design" sites based on screen sizes, but also to detect the type of pointer (mouse, touchscreen, etc.) or to display a specific interface when dark mode is activated.
Microdata
Refers to markup intended to present page data in a structured way. It can be integrated directly into HTML tags (microdata format) or via a script in JSON-LD format, recommended by Google. Google relies heavily on microdata to generate the "Rich Snippets" displayed on its results pages (star ratings, recipes, step-by-step guides, etc.). Synonyms: structured data, microdata.
Minification
Optimization technique aimed at reducing the size of CSS, JavaScript, and HTML files by removing all non-essential characters: spaces, line breaks, comments, optional syntax elements, etc.
Mojibake
Display of unreadable characters (a series of inconsistent symbols like "é" instead of "é") that occurs when text is decoded with a character encoding different from the one used to save it, typically UTF-8 content interpreted as ISO-8859-1. The term comes from Japanese and literally means "character transformation."
Synthetic monitoring
Performance monitoring through automatic tests launched at regular intervals by bots, under fixed conditions (device, network, location). A complement to RUM, it detects regressions as soon as they are deployed, before they are visible in visitor data.
N
Content negotiation
HTTP protocol mechanism by which the same address returns multiple representations of a resource depending on what the client requests in its headers: language, image format (AVIF, WebP), compression encoding, or, more recently, Markdown version for AI agents. It requires sending the Vary header so that caches and CDNs do not serve the wrong version to the wrong visitor.
O
Link obfuscation
SEO optimization technique aimed at hiding a hyperlink from indexing robots while ensuring transparent operation in browsers. This is generally done using JavaScript and base64 encoding.
OCSP Stapling
Translated as "OCSP Stapling" in French, this is a protocol that adds supplementary data during an HTTPS exchange between a server and a browser. This data allows the browser to verify the validity of the SSL certificate without having to query the issuing authority itself. As this check can delay the establishment of the connection, avoiding it improves TTFB. Note: the OCSP protocol is being gradually abandoned in favor of revocation lists (CRLs), with Let's Encrypt having stopped supporting it in 2025.
OPcache
PHP extension that keeps already compiled PHP code in memory, avoiding the need to reread and recompile it for each request. When enabled and properly configured, it significantly reduces the page generation time for WordPress, PrestaShop, or Drupal, and thus the TTFB.
P
Page builder
CMS extension that allows visual page building through drag-and-drop, without writing code: Elementor, Divi, or WPBakery on the WordPress side. While highly appreciated for their simplicity, page builders often generate heavy code in return (large DOM, superfluous CSS and JavaScript) which penalizes web performance if not optimized.
PageSpeed Insights
Performance measurement tool developed by Google's teams and focused on Core Web Vitals. You can run synthetic tests and access field data (CrUX) for any page, on both Desktop and mobile.
Percentile (p75)
Value below which a given percentage of measurements falls. Google evaluates Core Web Vitals at the 75th percentile of visits: a "good" LCP must be so for at least three out of four visits. This rule prevents a flattering average from masking a degraded experience for a significant portion of visitors.
PHP
Acronym for "Hypertext Preprocessor," it refers to a programming language used to run a majority of websites. Free and Open-Source, it is at the heart of most current CMS platforms: WordPress, PrestaShop, Magento, Drupal... It is regularly updated to improve performance.
Polyfill
JavaScript tool that emulates recent browser features on older browsers that do not support them natively. This family of tools has a significant negative impact on page performance.
Prompt
Instruction written in natural language and addressed to a generative AI model to obtain a response or action. The quality of the prompt (context, constraints, examples) determines the quality of the result.
Proxy
Tool that interfaces between a web browser and a web server to provide new functionalities. Most proxies improve performance and/or security levels. CDNs like Cloudflare are technically proxies.
PWA
Acronym for "Progressive Web App," it refers to a website that uses a service worker to offer features similar to a native application: installation on the home screen, offline functionality, push notifications, etc. The interface and experience thus approach what mobile devices natively offer.
Q
QUIC
A transport protocol designed by Google and standardized by the IETF, upon which HTTP/3 is based. It integrates TLS encryption and reduces the number of round trips needed to open a connection; it also allows a connection to be maintained during a network change (switching from Wi-Fi to 4G).
R
RAG (Retrieval-Augmented Generation)
From the English "Retrieval-Augmented Generation": a technique that precedes a language model's response with a search in documents (web pages, knowledge base), whose excerpts are provided to the model. This is how answer engines that cite their sources work (Perplexity, AI Overviews, ChatGPT with search): a fast, well-structured, and explorable page is more likely to be selected as a source.
301 Redirect
A response from a web server indicating to the browser and search engines that a page has permanently changed its address. Essential during a redesign to preserve SEO, it does have a cost: each redirect adds a network round trip. Redirect chains (A to B, then B to C) are particularly detrimental to TTFB.
Resource Hints
A family of HTML attributes that allow you to tell the browser which resources or origins to prepare in advance: preconnect to establish a connection, "preload" to load a critical resource, "prefetch" to anticipate the next navigation, etc. Used wisely, they improve speed metrics like LCP; used excessively, they produce the opposite effect.
Resource
A resource is a file called by the browser (via what is called an HTTP request) in order to display a web page. Any JavaScript, CSS, font, or image file is thus a resource from the perspective of a web page.
Blocking Resource
A file (stylesheet, synchronous script, font) that the browser must download and process before it can display the page. Each blocking resource delays FCP and LCP. They are limited by deferring scripts (defer, async), integrating critical CSS, and loading the rest asynchronously.
Third-Party Resource
A resource called by a page from a different domain name. This is very often JavaScript and fonts, but also sometimes CSS or images.
RGAA
General Accessibility Improvement Framework: a French method for evaluating the accessibility of sites and applications, based on WCAG. Its version 4.1.2 has 106 criteria divided into 13 themes (images, colors, forms, navigation, etc.). It serves as the basis for audits and accessibility statements.
RGESN
General Digital Service Eco-design Framework, published by Arcep and Arcom. It brings together criteria relating to the strategy, interface, content, code, and hosting of a digital service.
AI Bots
Crawlers sent by AI publishers. Two families are to be distinguished: training bots (GPTBot, ClaudeBot, Google-Extended…), which collect content to train models, and inference bots, which consult a page when a user asks a question (ChatGPT-User, Claude-User, Perplexity-User…). Blocking the former can protect a media outlet; blocking the latter makes the site invisible in assistant responses. They are controlled by robots.txt and, at the network level, by tools like those from Cloudflare.
Robots.txt
This text file, placed at the root of a site, aims to facilitate crawling by search engines. It guides indexing robots by allowing and blocking certain sections, as well as by providing links to xml sitemaps.
RUM
Acronym for "Real User Monitoring," which corresponds in French to what is called "field data." It involves collecting performance metrics from real users to generate reports in dashboards. The best-known RUM data source is CrUX, based on data from Google Chrome browser users. But there are dozens of paid tools that collect data via JavaScript tags, including historical APMs.
S
Schema.org
A common vocabulary, created by Google, Microsoft, Yahoo, and Yandex, which defines the types and properties of structured data (Organization, Article, Product, FAQPage, BreadcrumbList…). It is most often expressed in JSON-LD. See also Microdata.
Search Console
Google's free tool that allows site owners to track their presence in the search engine: indexed pages, queries generating clicks, errors encountered… Its "Core Web Vitals" report, fed by CrUX data, indicates whether pages meet the thresholds set for Core Web Vitals.
SEO
Acronym for "Search Engine Optimization," the term refers to the discipline aimed at improving a site's visibility in search engines. Historically very focused on positions in search results pages, SEO has become more complex over the evolution of Google's algorithm. Synonym: natural referencing.
SERP
Acronym for "Search Engine Results Page", it refers to the results page displayed by a search engine following a query. Ranking well there is the goal of SEO. Web performance contributes to this: with equal relevance and quality, Google favors sites that offer the best browsing experience.
Web server
Computer on which a website's data is hosted, and accessible to everyone via an internet connection. Specific software ensures its operation: Apache, Nginx, or LiteSpeed for serving files, MySQL for managing databases, PHP for executing scripts... Its owner can connect to it via FTP or SSH to manage it. See What is considered good hosting.
Service worker
JavaScript script that installs itself in the browser during a site visit and continues to run in the background. It can intercept requests to serve resources from a local cache, allow offline operation, or receive push notifications. It is the technical component at the heart of PWAs.
SLA
Acronym for "Service Level Agreement". A contractual commitment by a service provider regarding the quality of its service, generally expressed as an availability rate: an SLA of 99.9% allows for less than 9 hours of downtime per year. It is a key criterion in choosing hosting.
SPA
Acronym for "Single Page App", it refers to a type of website built in JavaScript that behaves like an application. Components displayed on the current page are dynamically updated via Ajax calls based on user interactions.
Speculation Rules
Browser API that allows anticipating the loading or even rendering of pages based on user actions. The browser uses its own signals to identify relevance and the opportune moment to initiate the process, generally on link hover. This is a very effective feature for improving the performance of multi-page browsing sessions.
Speed Index
Synthetic performance indicator that measures the time needed for elements to display within the viewport. As the information is visual and not technical, analysis tools rely on video recordings. Due to its complexity, ratings vary depending on the tool used.
SSG (static site generation)
From the English "Static Site Generation": pages are generated in HTML once and for all, at the time of publication, then served as-is. This is the fastest rendering mode (very low TTFB, full caching), suitable for content that changes little. See also SSR and CSR.
SSR
Acronym for “Server Side Rendering,” which translates to “server-side rendering” in French. This refers to performing pre-rendering of dynamic pages on the server. The technique is used with JavaScript frameworks like React, Angular, or Vue.js to improve performance, particularly LCP. Components are already present on load before being made interactive via JavaScript afterward (this is “hydration”).
Stack
A stack refers to a set of complementary technologies that allow a web application to function. We generally distinguish between the front-end stack (frameworks, tools, and languages...) and the back-end stack (web server, SSR tools...).
Stale-while-revalidate
A Cache-Control header directive that allows a cache to immediately serve a slightly stale copy while fetching a new one in the background. The visitor never waits for regeneration: an effective technique for keeping TTFB low on frequently changing content.
SVG
A vector image format particularly suited for logos, icons, and pictograms. It displays sharply regardless of screen pixel density.
Synchronous
A synchronous load is a load that executes immediately upon being called, interrupting document parsing and page rendering. This is the default behavior of a script tag without attributes: the browser suspends everything, downloads the file, executes it, and then resumes. Each synchronous script therefore delays display.
T
Tag manager
A tool, like Google Tag Manager, that allows adding measurement, advertising, or tracking scripts without modifying the site’s code. Convenient for marketing teams, it is also one of the primary sources of performance degradation: each added tag slows down the page and occupies the main thread. It requires strict governance.
Cache success rate
The proportion of requests served directly by a cache (page, CDN, object) without querying the origin server; also known as “cache hit ratio.” The higher it is, the lower the TTFB and the lighter the server load. It often drops due to tracking URL parameters (utm_source, gclid) or cookies that make each page unique.
TBT
Acronym for “Total Blocking Time,” this performance metric measures the sum of Blocking Time periods that occur during the loading of a web page. It is often used in synthetic tests as an approximation of INP, as the latter can only be measured via field data.
Template
A template is a set of HTML files used to manage the layout of a website. Some parts are common to the entire site (header, menus, footer, etc.) while others are specific to each type of page: homepage, content archives, articles, product sheets, contact page, etc.
TCP
Acronym for "Transmission Control Protocol," it refers to the data exchange protocol at the core of how the Internet works. It's what allows web browsers to download resources from a server when you visit a web page. It relies on IP addresses for this. Note: With HTTP/3, browsers now prefer QUIC, a newer protocol based on UDP that speeds up connection establishment.
TLS
Acronym for "Transport Layer Security," this protocol is responsible for ensuring secure exchanges between a browser and a server. It secures "https" pages by verifying the authenticity of the digital certificate provided by the server.
Tree shaking
Optimization performed during JavaScript compilation, which eliminates imported but unused code. It reduces the weight of files sent to the browser, especially with large libraries where only a small part is used.
TTFB
Acronym for "Time To First Byte," this performance indicator measures the time it takes for a server to respond to the browser's first request. It depends on the server configuration, the presence or absence of cache, but also on the visitor's connection (bandwidth, latency).
Tofu
In typography, it refers to a character that cannot be displayed because the glyph used is not available in the font file, leading to a rendering like □ or ☒. The term "Tofu" is used because this square resembles a block of tofu (the Asian soy cheese).
U
UI
Acronym for "User Interface." It refers to all the graphical elements of a web page. The structure, typography, images, and colors are all elements of the UI.
UX
Acronym for "User eXperience." It refers to how a user interacts with a web page. A quality UX involves relevant choices in terms of ergonomics, accessibility, and adherence to conventions and best practices.
V
VDS
Acronym for "Virtual Dedicated Server", which translates to "serveur dédié virtuel" in French. It is a type of hosting between a VPS, which is performant and flexible, and a dedicated server, which is very performant but less flexible. The main difference between a classic VPS and a VDS is that the latter has rigorous isolation, whereas some VPS offers rely on dynamic partitioning, with resources that can fluctuate depending on the load (CPU, RAM, and IO).
Viewport
Area of the current page visible in a web browser. When loading, the viewport corresponds to the area above the fold.
View Transitions
Browser API that animates the transition from one page state to another, or from one page to the next, without a JavaScript library. When used well, it gives an impression of fluidity comparable to an application, at a very low performance cost.
VOC
Acronym for "Voice Of Customer", which translates to "voix du client" in French. It is a family of tools that aim to collect user feedback on a site and anticipate their potential frustrations. They generally materialize as a button displayed in a fixed position, providing access to a chat window. By extension, chatbots and conversational robots are often included.
VPS
Acronym for "Virtual Private Server", which translates to "serveur privé virtuel" in French. It is a type of hosting between shared hosting, which is not very performant, and a dedicated server, which is very performant but also very expensive. A VPS offers an interesting compromise between performance, cost, security, and scalability: since it is not a physical machine, you can perform instant backups or change the configuration on the fly.
W
W3C
The W3C, or "World Wide Web Consortium", is the international organization responsible for overseeing web technologies. It is behind standards like XML, CSS, and SVG, which it regularly evolves. The HTML standard, on the other hand, has been maintained since 2019 by the WHATWG, with whom it collaborates.
WAF
Acronym for "Web Application Firewall", or "pare-feu applicatif" in French. Placed in front of the web server, it analyzes incoming requests to block attacks (injections, malicious bots, DDoS, etc.) before they reach the site. It also protects performance: parasitic traffic consumes server resources at the expense of real visitors.
Waterfall
Translates to "waterfall chart": a graphical representation of web page loading, where each row represents an HTTP request and its precise timeline. It is the reference tool for understanding how the browser loads a page and identifying bottlenecks. It can be found in browser developer tools (Network tab) and in testing tools like WebPageTest.
WCAG
Acronym for "Web Content Accessibility Guidelines", these guidelines aim to define the accessibility criteria of a website for people with disabilities and seniors. They are defined by the W3C.
Webfont
Font downloaded by a web page to display all or part of its text. The best practice is to provide it in Woff2, the lightest format, which is now supported by all browsers. We also recommend implementing font-display: swap to allow text to be displayed before the webfont is downloaded. For more information, see our article on optimizing web fonts.
WebP
Image format created by Google, lighter than JPEG and PNG at comparable quality, with support for transparency and animation. Supported by all current browsers, it often serves as a fallback for AVIF.
Web performance
Technical discipline aimed at ensuring fast display and good fluidity of web pages in browsers, during their loading and throughout their use. For a complete definition, see our article "What is web performance?".
Web Worker
Script executed in the background, on a thread separate from the main thread. It allows heavy computations (data processing, analysis) to be offloaded without blocking the page's display or interactions.
Z
Zstandard
Modern compression algorithm used by web servers to reduce the size of text-encoded resources (HTML, CSS, JavaScript...), and thus their download time by browsers. Its compression ratio is between those of Gzip and Brotli, but it compresses much faster than the latter, consuming less CPU resources. It is thus particularly suitable for dynamic content, such as HTML pages themselves.
