Using this technical SEO and AI search glossary

Precise terminology makes technical work easier to evaluate. This glossary explains the concepts used throughout the guides, including what each term describes and the distinction that matters when interpreting a report or recommendation.

Definitions are grouped by subject. An indexing instruction is different from a crawl restriction. A successful request is different from a search appearance. An AI citation is different from a recommendation. Keeping those boundaries clear helps website owners ask useful questions and gives development teams a shared understanding of the intended result.

Start with the technical SEO guide or the AI search overview for a wider explanation of how the terms fit together. The specialist guides explain how the terms apply to actual investigations and implementation decisions.

Discovery and indexing

Crawling

A crawler requests a page or resource from a server. This collects material for a search engine to assess. It doesn't mean the page will appear in search.

Indexing

A search engine analyses a page and may store information about it in its searchable collection. A successful page request alone doesn't establish that it's indexed.

Indexability

The technical conditions under which a page may be considered for a search index. Passing local checks doesn't establish that a search engine has indexed the page or chosen to show it.

Robots.txt

A public file describing which paths supporting crawlers may request. It controls crawling rather than acting as a password or a reliable way to remove indexed pages.

Noindex

A page-level instruction asking a supporting search engine to exclude the page from its index. The crawler needs to access the instruction to process it.

Canonical URL

The preferred address for duplicate or very similar content. A canonical tag expresses a preference. A search engine can select another representative.

Duplicate content

Content available at several addresses with little meaningful difference. Identify which version should represent it before changing links, redirects or canonical tags.

Orphan page

A page without discovered internal links pointing to it in the checked dataset. Missing crawl coverage can create a false impression that a page has no links.

Crawl trap

A URL pattern that leads crawlers through excessive or effectively endless variants. Calendars, filters and repeated parameters can create these patterns.

Crawl budget

The crawling activity a search engine is willing and able to spend on a site. Large or frequently changing websites are more likely to need a detailed investigation.

XML sitemap

A machine-readable list of URLs intended for discovery. It communicates which pages exist but doesn't guarantee that a search engine will index them.

Sitemap index

An XML file listing sitemap files. It helps organise a larger URL inventory, but the referenced files and their listed pages still require independent validation.

URL Inspection

A Search Console tool for examining information about an individual URL. Stored indexing information and a live accessibility test answer different questions.

URL Inspection API

A Search Console API returning information about the version recorded in Google’s index systems. It doesn't perform the live page test available through the interactive inspection workflow.

Google Indexing API

An API with a documented scope covering JobPosting pages and BroadcastEvent embedded in VideoObject. It isn't a general bulk-indexing service for ordinary articles, and a notification acknowledgement doesn't prove index inclusion.

Soft 404

A page that appears to be missing or empty while returning a success response, or is otherwise classified by a search engine as an error-like page. Inspect the actual body and intended resource state.

HTTP, delivery and website changes

HTTP status code

A server’s description of a request’s outcome. For example, 200 indicates success, 404 means not found, and 503 indicates temporary unavailability.

301 redirect

An HTTP response indicating that a resource has permanently moved. For a website change, choose a relevant destination and verify the complete route rather than checking only that a redirect exists.

308 redirect

An HTTP permanent redirect that preserves the request method and body. It should be selected deliberately within the application’s routing requirements and tested with the requests the route needs to support.

404 Not Found

An HTTP response indicating that the requested resource wasn't found. It can be appropriate for genuinely unavailable content; not every 404 requires redirection to an unrelated page.

410 Gone

An HTTP response indicating that access to a resource is no longer available and is likely to remain unavailable. Use it according to the intended removal policy rather than as a promise of immediate search removal.

503 Service Unavailable

An HTTP response for temporary inability to handle a request. It should accurately reflect the service state. A permanent missing page and a temporary server problem require different treatment.

304 Not Modified

A response to an appropriate conditional request indicating that a previously received representation can be reused. The resource validator must reflect meaningful changes, otherwise a client may retain outdated information.

Redirect

A response sending a request to another address. When a page moves, choose a relevant replacement and verify that visitors reach it successfully.

Redirect chain

A sequence of redirects before the final destination. Check each step, then simplify avoidable hops where the intended mapping allows it.

Redirect loop

A redirect sequence that returns to an earlier address instead of reaching a usable final page. It prevents the request from completing normally.

HTTPS

HTTP protected by TLS encryption. A move to HTTPS needs certificate, routing and internal URL checks as well as a working secure homepage.

HTTP Strict Transport Security (HSTS)

A browser policy requiring secure connections for a host and, optionally, its subdomains. It has operational persistence and should follow reliable HTTPS deployment rather than being added as an isolated SEO setting.

Content delivery network (CDN)

A delivery layer that can cache and serve resources closer to visitors. Its routing, cache and security rules can affect the public response independently of the website application.

Cache key

The information a cache uses to distinguish stored responses. Incorrect handling of language, parameters or other meaningful variants can cause one visitor to receive a response intended for another state.

Cache invalidation

The process of removing or replacing stored responses when their content is no longer current. The publishing process should identify which resources and variants need to be refreshed.

Website migration

A significant change to a site’s URLs, platform or structure. Planning should preserve useful content, discovery routes and destinations for existing links.

Rollback plan

A documented way to recover a working state if a release causes a serious failure. Include dependencies such as routing, caches and generated files.

JavaScript, rendering and page structure

JavaScript

Code that can change content or add interactive behaviour in the browser. A page’s important information may depend on this code running successfully.

Rendering

Turning page code into the content and layout a browser displays. Comparing the original response with this output can reveal content added or removed by scripts.

Document Object Model (DOM)

The browser’s working representation of the page. JavaScript can change it, so it may differ from the HTML originally sent by the server.

Server-side rendering (SSR)

Generating the page’s HTML on the server before sending it to the visitor. Important content can then be available without waiting for browser scripts to create it.

Client-side rendering (CSR)

Using code in the browser to create or update page content. A successful browser session doesn't prove that every other client receives the same useful content.

Hydration

Connecting browser-side interaction to HTML already produced by the server. A failure can leave content visible while buttons or other controls don't respond.

Dynamic rendering

Serving a pre-rendered representation to selected clients while visitors receive the application. Maintaining equivalent, current versions introduces additional operational work.

Semantic HTML

HTML elements chosen to describe content meaning, such as headings, lists and tables. It helps make the structure understandable without relying only on visual appearance.

Content parity

Agreement between the meaningful facts in different representations of a page. An HTML version and a Markdown version shouldn't disagree about eligibility, prices or conditions.

Representation parity

Agreement in the important information delivered through different versions of a resource. An HTML page and an agent-oriented alternative shouldn't disagree about conditions, availability or the substance of the offer.

Performance and user experience

Core Web Vitals

Three measures of user experience: loading performance, responsiveness and layout stability. Each describes a different aspect of using a page.

Largest Contentful Paint (LCP)

A loading metric marking when the largest qualifying image or text block is displayed in the viewport. Investigate the actual element before choosing a loading fix.

Interaction to Next Paint (INP)

A responsiveness metric assessing the delay between interactions and their visible updates. Slow script work can contribute to a page feeling unresponsive.

Cumulative Layout Shift (CLS)

A measure of unexpected movement in the visible page layout. Reserving space for media can help prevent surrounding content from jumping.

Field data

Measurements collected from real visits. They describe the people, devices and conditions represented in that dataset, rather than a controlled test environment.

Lab data

Measurements collected in a controlled test. They help reproduce and diagnose behaviour, but don't represent every visitor’s experience.

75th percentile (p75)

The value at or below which 75% of the measured observations fall. Check the metric, time window and visitor group before interpreting it.

Viewport

The visible area of the page on a device. What is initially visible can change with screen size, layout, consent notices and browser controls.

Device pixel ratio (DPR)

The relationship between a screen’s physical pixels and its CSS layout pixels. It affects which image resolution is appropriate for a displayed size.

Long task

A period of main-thread work lasting more than 50 milliseconds. It can delay other work, so inspect what the browser was doing and which interaction was affected.

Critical rendering path

The sequence of work the browser needs to display the required content. Performance diagnosis identifies the dependency causing delay rather than assuming that every downloaded resource has equal importance.

Preload

A hint asking the browser to fetch a resource needed by the page early. It should match the resource the page actually uses; incorrect hints can trigger unnecessary or duplicate downloads.

Lazy loading

Deferring a resource until it's near being needed. It can reduce unnecessary initial work, but critical immediately visible content shouldn't be delayed by an unsuitable lazy-loading rule.

Fetch priority

A browser hint describing the relative importance of fetching a resource. It isn't a guarantee of timing, and promoting every image removes the useful distinction between critical and secondary resources.

Responsive image candidates

Alternative image files offered through markup such as srcset. The browser combines the candidate information with layout and device conditions when selecting an image, so the delivered result still needs checking.

Structured data and identity

Schema.org vocabulary

A shared set of labels describing things such as organisations, articles and products. The chosen types should reflect the information actually represented on the page.

JSON-LD

A format for expressing structured data in a script block. It labels facts and relationships without replacing the content visitors read.

Entity

A distinguishable person, organisation, product or other thing. Stable identifiers help separate two things with similar names and connect references to the same thing.

Shared @id

An identifier used to refer to the same entity across structured data. Reusing a stable identifier helps avoid describing one organisation as several unrelated organisations.

Knowledge graph

A representation of things and their relationships. Structured data can describe those relationships, but publishing it doesn't guarantee inclusion in a search engine’s graph.

Rich result

An enhanced search presentation supported by a search provider. Passing a relevant markup test establishes a technical condition, rather than guaranteeing that presentation.

Entity resolution

Identifying whether references describe the same person, organisation or other thing. Similar names alone are insufficient. Use relevant distinguishing facts and supported relationships.

International and multilingual SEO

Locale

A language and regional context, such as British English or French for Canada. A language alone doesn't define the market a business can serve.

Localisation

Adapting content and journeys to an audience’s market. This can include availability, units, currency and support information as well as translation.

Hreflang

Markup identifying equivalent pages for different languages or regions. It describes the intended alternatives and should point to relevant, published editions.

Reciprocal hreflang

A return relationship between equivalent editions. The pages should identify the same intended alternate set rather than disagreeing about which versions belong together.

x-default

An optional hreflang value for a fallback page that isn't aimed at one particular language or region, such as an appropriate language selector.

Country-code top-level domain (ccTLD)

A domain ending associated with a country or territory. Choosing one affects ownership and maintenance as well as the structure of an international website.

Language subdirectory

A path used for a language or market edition, such as /fr/. The folder name should be backed by accurate content and consistent links.

Language and market

Language describes how information is written. A market describes the audience and offer served. One language can cover several markets, and a market may need several language editions.

Auditing, analysis and automation

Technical SEO audit

An investigation of how pages are discovered, accessed and understood. A useful finding connects an observed problem with evidence, an expected change and a way to verify it.

Acceptance criteria

Specific conditions that define a successful change. They explain how the person making the fix and the person reviewing it will decide the work is complete.

Baseline

A recorded starting point for comparison. Keep its scope, collection date and settings so a later difference can be interpreted accurately.

Quality assurance (QA)

Checks that compare delivered behaviour with the agreed requirements. They should include the intended fix and nearby examples that must continue working.

Server logs

Records of requests received by a server. They can show requested addresses, timing and responses, but don't reveal every later use of the content.

Regression

Something that worked before a change and breaks afterwards. Regression checks help catch unintended effects outside the feature being changed.

Issue prioritisation

Deciding which findings deserve attention first. Consider affected journeys, the strength of the evidence, likely impact, effort and dependencies rather than counting tool warnings.

Application Programming Interface (API)

A defined way for software to request information or perform an operation. Document the access rules, response fields and limits before automating work.

Comma-separated values (CSV)

A text table used for exports and imports. Correct handling of separators, quotation marks and text encoding matters when combining data from different tools.

Join key

The field used to match records between datasets, often a URL or stable record ID. Inconsistent normalisation can make the same record look like two different pages.

API quota

A limit on requests or usage. An automated workflow needs to handle the limit without silently producing an incomplete report.

Data validation

Checks that confirm inputs meet the expected format and meaning. Valid syntax alone doesn't prove the data is complete, current or appropriate for the decision.

Technical SEO dashboard

A view of selected observations and trends. It should identify the data source, coverage and collection time so readers can assess what the figures establish.

Content management system (CMS)

Software used to publish and maintain a website. A shared template or field can affect many pages, making the scope of a change important.

Investigation scope

The agreed pages, systems and conditions included in a piece of work. Clear boundaries make it easier to interpret findings and identify what wasn't tested.

Comparison group

A defined group of pages, questions or visits used in a comparison. Keep the same members in the group so changes aren't caused merely by changing the group.

Data join

Combining records through a defined matching field. Validate the record level and matching rules so duplicates, missing keys or different URL versions don't inflate or remove observations.

AI search, retrieval and measurement

AI search

An experience that uses AI to answer or explore a question, sometimes with retrieved web sources. Different AI search tools use different retrieval, display and access rules.

Large language model (LLM)

A model trained to generate and process language. Its response may use learned patterns, supplied context or external tools. The presence of an LLM doesn't check whether a particular answer used live web information.

Retrieval

Looking up information from an available collection or source. A retrieval-based answer can use external material rather than relying only on information learned during model training.

Grounding

Connecting an answer to reference material used to support it. A linked reference still needs to be checked for whether it supports the specific claim.

Retrieval-augmented generation (RAG)

An approach that retrieves external information and supplies it to a model when generating a response. The quality of the answer depends on the retrieved material as well as how the system uses it.

Query fan-out

Generating related searches to investigate parts of a larger question. It helps explain why a detailed supporting guide can be relevant to a broad user request.

Generative engine optimisation (GEO)

A label often used for work intended to improve visibility in AI answers. Define the platform and measurable objective rather than treating the label as a separate technical standard.

Model training

The process used to develop a model’s learned behaviour from data. Training is distinct from retrieving a current web page to answer a particular question, so provider access policies may treat them separately.

Context window

The amount of information a model can consider in a particular processing context. A larger allowance doesn't mean every supplied statement is accurate or that the system will use every source.

Grounding query

A search or retrieval query used to obtain supporting information for an AI response. It isn't necessarily the same wording as the question a person entered.

Citation

A displayed reference to a source. A citation, an unlinked brand mention and a recommendation are different observations and should be recorded separately.

Brand mention

An occurrence of a business or brand name in an answer. A mention can be linked or unlinked, positive or negative, and shouldn't automatically be classified as a citation or recommendation.

AI recommendation

An answer that presents an option as suitable for a stated purpose. Assess the actual wording and conditions; merely naming a business in a list doesn't establish that it was recommended.

Prompt

The input supplied to an AI system. It can include a question, instructions and conversation context. A selected monitoring prompt is a test case, not a record of all questions real customers ask.

Prompt test set

A documented collection of questions used to sample an AI search tool’s responses. It describes the chosen test, rather than revealing all real customer prompts.

Multi-turn conversation

A conversation containing several exchanges. Later answers can depend on earlier messages, so a repeatable test should preserve the full sequence and identify the reply being measured.

Semantic search

Retrieval that considers meaning rather than depending only on an exact wording match. A useful comparison still needs a defined collection, query and measure of relevance.

Embedding

A numerical representation used to compare characteristics of data, often including similarity in meaning. An embedding can support retrieval but doesn't prove that two passages are factually equivalent.

Referral traffic

Visits attributed to a referring source. Some AI exposure produces no click or no usable referrer, so referral visits can't measure every appearance in an answer.

AI agent

A system that uses an AI model and tools to work towards a task across one or more steps. For a website investigation, those tools might search for sources, retrieve pages or interact with a browser. An agent is the working system around the model, rather than another name for the LLM itself.

AI clients and access management

Search crawler

An automated client collecting content for a search function. Its identity and controls should be checked against the particular provider’s documentation.

User-agent

A request header identifying the software making a request. The text can be copied, so the claimed name alone doesn't verify who sent it.

Web application firewall (WAF)

A security layer that filters requests before they reach the application. A bot can be permitted by robots.txt while still receiving a firewall block or challenge.

Browser challenge

A check that may require scripts, cookies or interaction before content is served. A client unable to complete it may receive the challenge page instead of the article.

Crawler access policy

A documented decision about which automated clients may access which content for which purpose. Search access and training preferences should be considered separately.

llms.txt

A proposed Markdown index providing guidance and links for agents using a site. Whether a particular client uses it needs evidence; publishing one doesn't guarantee AI search visibility.

Frequently asked questions

How is crawling different from indexing?

Crawling is requesting a resource. Indexing concerns the information a search engine may retain for retrieval. A page can be requested successfully without being included in the search index.

Is rendering the same as downloading HTML?

No. The HTML response is an input to the page. Rendering processes the document, styles and relevant code into the content and layout a browser can display.

Does valid structured data guarantee a rich result?

No. Syntax, accurate statements and supported feature requirements are separate checks. A valid implementation doesn't guarantee a particular search presentation.

What is the difference between an AI citation and a brand mention?

A citation displays a reference to a source. A brand mention names a business, with or without a source link. Neither automatically establishes an explicit recommendation.

What does a technical SEO acceptance criterion describe?

It defines an observable condition required to complete a change, such as an old URL reaching its approved replacement or a public template returning its complete content without noindex.

Should I read the glossary before every guide?

No. The guides introduce their subjects directly and link to this reference for unfamiliar terminology. Use the glossary when a term needs clarification or when similar terms describe different behaviours.

Technical references

Written by Richie Aubin

Published · Updated