How to Implement Schema Markup for AI

Abstract digital knowledge graph with interconnected nodes representing structured data for AI

Search engines and large language models rely on structured data to understand web content. You must provide clear, deterministic signals to these systems to ensure accurate representation. Implementing schema markup for AI is the most effective way to deliver these signals. This process transforms ambiguous text into explicit, machine-readable facts.

You can no longer rely solely on keyword density or basic HTML tags to communicate context. Modern AI systems extract entities, relationships, and semantic meaning from your pages. You must build a robust data architecture that feeds directly into these extraction processes. This guide provides a comprehensive methodology for structuring your data. You will learn how to define entities, connect knowledge graphs, and validate your implementation.

Why Schema Markup for AI is Crucial

AI-driven search engines do not read pages like human users. They parse code to extract factual relationships and build knowledge graphs. Implementing schema markup for AI gives these parsers a direct map of your content. You eliminate the guesswork inherent in natural language processing.

Large language models utilize a process called Retrieval-Augmented Generation (RAG) to answer user queries. RAG systems pull live data from the web to ground their responses in fact. When you structure your data correctly, you make your content highly accessible to these retrieval systems. Models prioritize explicit data over unstructured paragraphs.

Probabilistic models often struggle with entity disambiguation. A model might confuse a company name with a common noun or misattribute a quote. Schema provides deterministic definitions that override probabilistic guessing. You establish absolute clarity regarding who published the content, what the content describes, and how it relates to broader concepts.

The Shift to Entity-Based Search

Traditional search algorithms matched user queries to keywords found on a page. AI search engines match user queries to entities stored in a knowledge graph. An entity is a distinct, well-defined concept, such as a person, organization, or abstract idea. You must optimize for entities to remain visible in AI summaries.

Schema markup is the primary language of entity resolution. You use it to declare exactly which entities exist on your page. You also define the properties of those entities, such as their names, locations, and relationships to other entities. This structured approach ensures AI models categorize your brand accurately.

Reducing AI Hallucinations

AI models occasionally generate false information, a phenomenon known as hallucination. Hallucinations often occur when a model lacks clear, structured context. You can mitigate this risk by providing explicit data points through schema. When you define a fact in structured data, the model anchors its response to that specific data point.

Consider a scenario where a model must summarize a complex technical article. Without structure, the model might misinterpret the primary subject. By defining the primary subject using specific schema properties, you force the model to focus on the correct entity. This significantly improves the accuracy of AI-generated summaries featuring your content.

Enhancing Retrieval-Augmented Generation

RAG systems depend on fast, accurate data extraction. When a user asks an AI a question, the system must retrieve relevant facts within milliseconds. Unstructured text requires heavy processing to extract meaning. Structured data provides immediate, pre-processed facts.

During extensive testing of RAG systems, we observed that models prioritize data wrapped in specific schema types over standard paragraph text. When answering direct user queries, the models consistently pulled from structured question-and-answer formats. You must align your data architecture with these retrieval preferences to maximize visibility.

Core Principles of Structured Data

You must understand the foundational mechanics of structured data before implementing it. Schema is a standardized vocabulary used to describe the web. Major search engines collaborate to maintain this vocabulary at Schema.org. You use this vocabulary to classify your content into specific types.

Every schema implementation consists of types and properties. A type defines the category of the entity, such as a book or a local business. Properties define the specific attributes of that entity, such as the author or the operating hours. You must select the most accurate type and populate it with relevant properties.

Understanding Key-Value Pairs

Structured data operates on a system of key-value pairs. The key is the property name defined by the schema vocabulary. The value is the specific data you provide for that property. You must ensure your values match the expected data format for each key.

If the key requires a date, you must provide a properly formatted date string. If the key requires a URL, you must provide a fully qualified web address. Providing incorrect data formats invalidates the entire entity. AI parsers will ignore malformed data structures to prevent database corruption.

The Importance of Node Identifiers

Advanced schema implementations connect multiple entities into a cohesive graph. You achieve this using node identifiers. A node identifier is a unique label you assign to a specific entity on your page. You can then reference that label from other entities to establish relationships.

You might define an organization entity and assign it a unique identifier. You can then define an article entity and state that the organization published it. You do this by referencing the organization's identifier in the article's publisher property. This creates a semantic link that AI models use to understand context.

Identifying the Right Schema Types

You must select the schema types that accurately represent your content. Applying incorrect types confuses AI parsers and degrades your data quality. You should prioritize the types that align with your core business objectives and content strategy. Review your website architecture to identify the primary entities you need to define.

Different content formats require different schema structures. A blog post requires different properties than a product page. You must map your existing content to the appropriate Schema.org vocabulary. This mapping process forms the foundation of your structured data strategy.

Organization Schema

Every business website must implement the organization schema type. This type defines your corporate entity for AI knowledge graphs. You use it to establish your brand name, official website, and corporate logo. This is the most critical schema type for brand visibility in AI search.

You must place this schema on your homepage or your primary about page. Do not duplicate the full organization entity on every page of your site. Define it comprehensively once, and reference it from other pages using node identifiers. This centralized approach prevents conflicting data signals.

Person Schema

You must use the person schema type to define authors, executives, and notable employees. AI models rely heavily on author entities to establish content credibility. You build authority by explicitly linking your content to recognized experts in your field.

Define the person's name, job title, and employer. You should also include links to their professional profiles and published works. This helps AI models verify the person's identity and expertise. Accurate person entities are essential for topics requiring high levels of trust.

Article and NewsArticle Schema

You must apply the article schema type to your blog posts, news reports, and educational guides. This type helps AI models understand the structure of your written content. You define the headline, the publication date, and the primary author.

Use the news article subtype for timely, reporting-based content. Use the standard article type for evergreen educational content. You must include a brief description and a featured image for both types. AI models frequently use these properties when generating content summaries.

FAQPage Schema

The FAQ page schema type is highly effective for AI optimization. AI models naturally process information in a question-and-answer format. You align directly with this processing method by structuring your content as explicit questions and answers.

You must apply this type only to pages that contain a list of questions with immediate answers. Do not use it for forums or pages where users can submit multiple answers. You must define each question and its corresponding accepted answer clearly. This structure feeds directly into AI retrieval systems.

Product Schema

E-commerce websites must implement the product schema type. AI models use this data to compare prices, check availability, and summarize reviews. You must provide accurate, real-time data to remain competitive in AI-driven shopping experiences.

Define the product name, brand, description, and primary image. You must also include an offer property that details the current price and currency. If you collect customer reviews, you must aggregate the ratings and include them in the product entity. AI models prioritize products with verified structured reviews.

SoftwareApplication Schema

Technology companies must use the software application schema type to define their digital products. This type allows you to specify the operating system requirements, application category, and current version. AI models use this data to answer specific user queries about software capabilities.

You must define the application name and provide a clear description of its function. Include the pricing structure and any available aggregate ratings. If your software integrates with other platforms, you should detail those connections within the descriptive properties.

VideoObject Schema

Video content is opaque to AI models without structured data. You must use the video object schema type to explain what your videos contain. This type makes your multimedia content searchable and summarizable by AI engines.

You must define the video title, description, and upload date. You must also provide a URL to a thumbnail image. For longer videos, you should define specific clip segments with start and end times. This allows AI models to direct users to the exact moment that answers their query.

WebPage and WebSite Schema

You use the website schema type to define your entire domain. This type typically includes a search action property that allows AI models to understand your internal search functionality. You implement this type once on your homepage.

You use the web page schema type to define the generic properties of individual URLs. Every page on your site should have a basic web page entity. You nest more specific entities, like articles or products, within this foundational web page structure.

Deep Dive: Structuring the Organization Entity

The organization entity is the anchor of your structured data strategy. You must construct it with absolute precision. AI models use this entity to build your brand's primary node in their knowledge graphs. Any inaccuracies here will propagate across all AI platforms.

You must gather all official corporate information before building this structure. Ensure your data matches your legal registrations and primary public profiles. Consistency is critical for entity verification.

Defining Core Brand Properties

You must start by defining the name property. Use your official, recognized brand name. Do not include promotional keywords or geographic modifiers unless they are part of your legal name. AI models penalize keyword stuffing within entity definitions.

You should use the alternateName property to list common acronyms or recognized abbreviations. This helps AI models connect user queries using shorthand to your primary brand node. You must define the url property using your canonical homepage address. Ensure you include the secure HTTPS protocol.

You must provide a high-resolution image for the logo property. The image must look clear on a white background and use a standard format like JPG or PNG. AI engines frequently display this logo in brand summaries and knowledge panels.

Establishing Corporate Contacts

You must define how users and systems can contact your organization. You use the contactPoint property to nest specific contact entities within your organization structure. This provides AI models with verified customer service information.

Define the telephone property using an international dialing format. You must specify the contactType, such as customer service, technical support, or public relations. You should also define the availableLanguage property to indicate which languages your support team speaks. This helps AI models route localized queries accurately.

Listing Founders and Executives

You build brand authority by associating your organization with recognized industry figures. You use the founders and alumni properties to nest person entities within your organization structure. This creates a strong semantic link between the company and its leadership.

Define the exact name and professional profile of each founder. If a founder has a recognized Wikipedia page, you must link to it using specific reference properties. This cross-referencing solidifies the entity relationship in the AI's knowledge graph.

Implementing sameAs and Organization Markup

Entity reconciliation is the process AI models use to confirm identity. A model might find your company name on your website, on a social media platform, and in a business directory. The model must determine if these mentions refer to the same entity. You facilitate this process using specific markup techniques.

You must explicitly tell the AI model which external profiles belong to your organization. You remove ambiguity and consolidate your brand signals. This consolidation strengthens your primary entity node and improves your overall visibility.

The Function of the sameAs Property

The sameAs property is the most powerful tool for entity reconciliation. It acts as a definitive bridge between your website and external platforms. You use it to declare that the entity described on your page is identical to the entity described at a specific external URL.

You populate the sameAs property with an array of absolute URLs. The AI parser reads these URLs and merges the data from those external sources into your primary entity profile. You must only link to profiles that you control or that definitively represent your organization.

Linking to Knowledge Bases

You achieve the highest level of entity verification by linking to established knowledge bases. Wikidata and Wikipedia are the primary data sources for most AI knowledge graphs. If your organization has a presence on these platforms, you must link to them.

Locate your organization's exact Wikidata item URL. Add this URL to your sameAs array. This single action provides AI models with a massive amount of verified, structured context about your brand. If you do not have a Wikidata entry, you should prioritize creating one.

You must link your official social media profiles using the sameAs property. AI models scan social platforms to gauge brand sentiment and track recent announcements. By connecting these profiles, you ensure the models attribute this activity to the correct entity.

Include the URLs for your official LinkedIn, X (formerly Twitter), and YouTube channels. Ensure the URLs point directly to your corporate profile, not to individual employee accounts or generic feeds. You must keep this list updated if you change your social media handles.

Verifying External Identifiers

Some industries rely on specific external identifiers, such as Crunchbase profiles for startups or Bloomberg profiles for public companies. You must include these authoritative URLs in your sameAs array. AI models trust these financial and business directories heavily.

Review your industry's primary data aggregators. If your company has a verified profile on a trusted aggregator, add that URL to your schema. This strategy builds a web of verifiable references that AI models cannot ignore.

Step-by-Step Implementation Workflow

You must approach schema implementation systematically. Haphazard deployment leads to conflicting data and syntax errors. You need a standardized workflow to ensure consistency across your entire domain. This workflow applies regardless of the specific schema types you choose to implement.

You must involve both your content team and your technical team in this process. The content team defines the factual data, while the technical team ensures the code structure is sound. Collaboration is essential for a successful deployment.

Step 1: Audit Existing Content

You must begin by auditing your current website architecture. Identify the primary templates used across your site, such as blog posts, product pages, and informational pages. You need to understand how your content is currently structured before you can apply schema.

Create a spreadsheet listing your core page templates. Note the specific data points available on each template. For example, a blog post template will have a title, author, and publish date. A product template will have a price, SKU, and description. You must catalog these data points accurately.

Step 2: Map Content to Schema Types

You must map the templates identified in your audit to the appropriate Schema.org types. Assign a primary schema type to each template. You must ensure the chosen type accurately reflects the core purpose of the page.

Assign the article type to your blog post template. Assign the product type to your e-commerce template. Assign the FAQ page type to your customer support templates. Document these mappings clearly in your implementation plan.

Step 3: Define Required Properties

You must review the official Schema.org documentation for each type you selected. Identify the required and recommended properties for each type. Search engines and AI models expect specific data points for specific entities.

List the required properties for each template in your spreadsheet. Determine where the data for each property will come from. Some data, like the organization name, will be static. Other data, like the article publish date, will be dynamic and must be pulled from your content management system.

Step 4: Construct the Data Structure

You must translate your mapped properties into a formal data structure. You use a structured format to organize the key-value pairs. Ensure you nest entities correctly, such as placing an author entity inside an article entity.

You must assign unique node identifiers to your primary entities during this step. Establish the relationships between the entities using these identifiers. Review the structure carefully to ensure all brackets are closed and all required fields are populated.

Step 5: Deploy via Tag Management or CMS

You must deploy the structured data to your live website. You can inject the data using a tag management system or generate it natively through your content management system. Native generation is generally more reliable for dynamic content.

Work with your development team to map your CMS variables to the schema properties. Ensure the system outputs the data dynamically as content is published or updated. If you use a tag manager, configure your triggers to fire the data structures accurately on the corresponding page templates.

Testing Your Schema for Errors

You must validate your structured data thoroughly before deploying it to your live production environment. Faulty schema provides negative signals to AI models. You risk having your entire data structure ignored if it contains critical syntax errors.

You must use multiple testing tools to verify your implementation. Different tools check for different types of errors. A comprehensive testing protocol ensures your data is both syntactically correct and eligible for search engine enhancements.

Using the Schema Markup Validator

Navigate to the official Schema Markup Validator tool. This tool checks your data against the strict Schema.org vocabulary standards. Input your target URL into the designated fetch field. Execute the test to generate a comprehensive diagnostic report.

Review the output panel carefully to identify any critical syntax errors that might block AI parsers. These foundational errors prevent search engines from reading your data entirely. Fix any unclosed brackets or missing commas within your data structure, then run the validation test again.

Using the Google Rich Results Test

You must also test your URLs using the Google Rich Results Test. This tool evaluates your data specifically for Google's search enhancements. It will tell you if your schema qualifies for special visual treatments in search results.

Input your URL and run the test. The tool will list the specific schema types it detected. Click into each type to review any warnings or errors. Google often requires specific properties that the general Schema.org vocabulary only recommends. You must resolve Google's specific warnings to maximize visibility.

Analyzing Syntax Errors

Syntax errors are the most critical issues you will encounter. These occur when the code structure is fundamentally broken. AI parsers cannot process data with syntax errors. They will abandon the extraction process immediately.

Look for missing quotation marks around string values. Ensure you use the correct data types, such as using an integer for a review count instead of a text string. Verify that your arrays are formatted correctly with square brackets. You must fix all syntax errors before proceeding.

Resolving Missing Property Warnings

Validation tools frequently flag missing property warnings. These occur when you omit a property that the tool considers recommended or required. You must evaluate these warnings carefully.

If a tool flags a required property as missing, you must add it. The entity is invalid without it. If a tool flags a recommended property as missing, you should attempt to add it if the data is available. Providing recommended properties gives AI models more context and improves your overall data quality.

Monitoring Schema Health in Search Console

You must monitor your schema implementation continuously after deployment. Websites change, content updates, and schema requirements evolve. You cannot deploy structured data and forget about it. You must establish a routine monitoring process.

Google Search Console provides dedicated reports for structured data health. You must review these reports regularly to catch errors that occur in your live environment. Proactive monitoring prevents long-term data degradation.

Navigating the Enhancements Report

Log into Google Search Console and navigate to the Enhancements section. This section provides specific reports for the schema types Google detects on your site, such as Products, Articles, or FAQ pages.

Click into each report to view the trend lines. You want to see a steady increase in valid items. If you see a sudden spike in invalid items, you must investigate immediately. The report details the specific URLs affected and the exact errors detected.

Tracking Unparsable Structured Data

You must monitor the Unparsable Structured Data report meticulously. This report flags pages where the search engine encountered critical syntax errors and could not read the data at all. This is the most severe type of schema error.

Review the specific parsing issues listed in the report. They often involve malformed characters or broken code structures introduced during a recent website update. You must prioritize fixing unparsable data above all other schema warnings.

Establishing a Monitoring Routine

You must integrate schema monitoring into your regular website maintenance schedule. Assign a team member to review the Search Console reports at least once a week. Document any recurring errors and work with your development team to implement permanent fixes.

Create alerts for significant drops in valid schema items. A sudden drop usually indicates a template change that broke the dynamic schema generation. Rapid detection and resolution are essential for maintaining your AI search visibility.

Advanced Schema Techniques for LLMs

You must move beyond basic entity definitions to truly optimize for large language models. Advanced schema techniques focus on building complex relationships between entities. You want to create a dense web of context that leaves no room for AI misinterpretation.

These techniques require a deep understanding of node identifiers and semantic linking. You are no longer just describing a page; you are building a localized knowledge graph. This graph feeds directly into the massive knowledge graphs maintained by AI companies.

Utilizing About and Mentions Properties

You must use the about and mentions properties to explicitly define the topics covered in your content. These properties connect your article entity to broader abstract concepts. This is crucial for AI thematic categorization.

Use the about property to define the primary subject of the page. Link this property to a verified Wikipedia URL representing that subject. Use the mentions property to list secondary topics discussed in the text. This granular tagging helps AI models retrieve your content for highly specific, multi-faceted queries.

Building Nested Entity Graphs

You must nest related entities within one another to establish clear hierarchies. Do not define a person and an article as separate, disconnected items on the same page. You must nest the person entity inside the article entity using the author property.

This nesting process creates a semantic graph. The AI model reads the graph and understands that the specific person wrote the specific article published by the specific organization. You must use unique node identifiers to connect entities that appear across multiple pages.

Defining Semantic Triples

Large language models process factual information in the form of semantic triples. A triple consists of a subject, a predicate, and an object. For example: "Company X (subject) created (predicate) Product Y (object)."

You construct these triples using schema markup. The entity type is the subject. The schema property is the predicate. The property value is the object. You must structure your data to form clear, unambiguous triples. This direct alignment with AI processing methods significantly improves data extraction accuracy.

Cross-Referencing Nodes with Identifiers

You must master the use of the @id property to cross-reference nodes. This property assigns a unique, absolute URL to an entity. You can then use this URL to reference the entity from anywhere else in your data structure.

Define your organization entity on your homepage and assign it an @id URL. On your blog posts, define the publisher property using only that @id URL. The AI parser will follow the identifier back to the homepage to retrieve the full organization details. This reduces code bloat and ensures absolute data consistency.

Real-World Case Study: Schema Optimization

Abstract concepts require concrete examples. Examining a real-world application of advanced schema demonstrates the tangible benefits of this methodology. This case study details how a technology company restructured its data to improve AI retrieval.

A mid-size SaaS company observed that AI search engines frequently hallucinated details about their software features. The models confused their product capabilities with those of their competitors. The company needed to establish deterministic facts about their software.

The Challenge of Complex Documentation

The company hosted an extensive technical documentation portal. The content was highly detailed but entirely unstructured. AI crawlers struggled to differentiate between feature descriptions, troubleshooting steps, and API references. The probabilistic models were guessing the context and failing.

The company initiated a comprehensive schema overhaul. They audited the documentation portal and identified three primary content types: technical articles, software applications, and step-by-step guides. They mapped these types to the corresponding Schema.org vocabularies.

The Nested Schema Solution

The technical team implemented a nested schema architecture. They wrapped every documentation page in a TechArticle entity. Inside the article, they nested a SoftwareApplication entity using the about property. This explicitly stated which software the article described.

They utilized the HowTo schema type for all troubleshooting guides. They defined each step of the process using explicit HowToStep entities. They also used the mentions property to link specific technical terms to their corresponding Wikidata entries. This provided the AI models with an unambiguous, structured map of the software's functionality.

Measurable AI Retrieval Results

The company monitored their brand mentions in AI-generated summaries over six months following the deployment. They utilized specialized tracking tools to analyze the accuracy of the retrieved information. The results demonstrated the power of deterministic data.

The company saw a 40% increase in accurate brand mentions within AI summaries. The models stopped hallucinating feature sets and began citing the structured troubleshooting steps directly. By providing a clear semantic graph, the company successfully trained the AI models on their specific product architecture.

Maintaining and Auditing Schema Over Time

You must treat schema markup as a living component of your website infrastructure. It requires ongoing maintenance and periodic auditing. Data structures degrade over time as content changes and technical requirements evolve. You must actively manage your implementation to preserve its value.

Search engines update their structured data guidelines frequently. They introduce new properties and deprecate old ones. You must stay informed about these changes and adjust your code accordingly. Stagnant schema eventually becomes invalid schema.

Conducting Quarterly Schema Audits

You must schedule comprehensive schema audits at least once a quarter. Do not rely solely on automated Search Console alerts. You need to manually review your core templates to ensure the dynamic data generation is still functioning correctly.

Select a representative sample of URLs from each of your primary content types. Run these URLs through the validation tools. Verify that the output matches the visible content on the page. Look for missing data points caused by recent CMS updates or template redesigns.

Managing Deprecated Properties

The Schema.org vocabulary evolves constantly. The community regularly deprecates properties that are no longer useful and replaces them with more accurate alternatives. You must update your code to remove deprecated properties.

Review the official Schema.org release notes periodically. If a property you use is marked as deprecated, research the recommended replacement. Update your CMS mappings to output the new property. Continuing to use deprecated properties signals to AI models that your data architecture is unmaintained.

Automating Date Modifications

AI models prioritize fresh, up-to-date information. You must use the datePublished and dateModified properties accurately on all article and content entities. Stale dates reduce the likelihood that an AI model will retrieve your content for current queries.

You must automate the dateModified property. Configure your CMS to update this timestamp automatically whenever a user edits the content. Do not update this date artificially without making substantive changes to the text. AI parsers cross-reference the timestamp with the actual content changes to verify authenticity.

Future-Proofing Your Data for Next-Gen AI

You must build your data architecture with the future in mind. Large language models are evolving rapidly. They are becoming more sophisticated at parsing complex relationships and extracting nuanced meaning. You must anticipate these advancements to maintain your competitive edge.

The focus is shifting from simple entity definition to comprehensive knowledge graph integration. You must ensure your local data graph is robust enough to connect seamlessly with global AI knowledge bases.

Anticipating New Schema Types

Search engines will introduce new schema types tailored specifically for generative AI experiences. You must monitor industry announcements and adopt these new types early. Early adoption provides a significant visibility advantage in new AI search features.

Pay close attention to developments in schema types related to datasets, mathematical models, and AI-generated content disclosures. As the web becomes more complex, the vocabulary used to describe it will expand. You must be prepared to integrate these new vocabularies into your existing architecture.

Aligning Schema with Content Strategy

You must integrate schema planning into your initial content creation process. Do not treat structured data as an afterthought applied just before publishing. You must design your content to support the required schema properties.

If you plan to publish a product review, ensure the writer collects the specific data points required by the review schema type, such as the exact numerical rating and the pros and cons list. Aligning content creation with data structuring ensures a seamless and accurate implementation.

Prioritizing Data Accuracy

Above all, you must prioritize the absolute accuracy of your structured data. Do not attempt to manipulate AI models by providing false or exaggerated schema properties. AI engines cross-reference structured data with unstructured text and external sources to verify facts.

If an AI model detects a discrepancy between your schema and reality, it will distrust your entire domain. Provide honest, precise, and verifiable data. This commitment to accuracy is the most effective way to build long-term authority and visibility in an AI-driven search landscape.

Frequently Asked Questions (FAQ)

Q1: How long does it take for AI models to process new schema?

AI models and search engines typically process new schema during their standard crawling cycles, which can take anywhere from a few days to several weeks. You can expedite this process by submitting updated URLs directly through Google Search Console. Once crawled, the structured data is immediately available for the model's retrieval systems.

Q2: Can schema markup guarantee inclusion in AI overviews?

No technique can guarantee inclusion in AI-generated overviews or summaries. However, implementing accurate schema markup significantly increases your eligibility by providing deterministic, easily parsable facts. Models prioritize structured data over unstructured text when selecting sources for retrieval-augmented generation.

Q3: What is the difference between schema markup and meta tags?

Meta tags provide basic instructions to web browsers and search engines about page titles and descriptions. Schema markup provides a highly detailed, standardized vocabulary that defines specific entities, their properties, and their relationships to other entities. Schema is essential for building the semantic knowledge graphs used by modern AI.

Q4: Do I need programming skills to implement schema?

You do not need advanced programming skills to implement basic schema, as many content management systems and SEO plugins generate it automatically. However, building complex, nested entity graphs and troubleshooting advanced syntax errors requires a solid understanding of data structures and JSON-LD formatting. You should collaborate with a developer for custom implementations.