Our approach to Artificial Intelligence

YouGov deploys Artificial Intelligence (AI) across its products, services and internal operations to enhance speed, productivity and analytical value, while maintaining robust security, privacy and governance controls.

At YouGov, AI accelerates analysis and content generation but does not replace human accountability and oversight. AI functionality is designed and deployed in accordance with YouGov’s Information Security and Data Protection frameworks and its evolving AI governance practices.

This document provides an overview of how YouGov deploys AI within its products, services and internal operations. The document is designed to support client due diligence and to address common questions relating to:

  • whether and how AI may interact with client data;
  • how AI models are accessed and governed;
  • how data confidentiality, privacy and security are protected; and
  • how YouGov applies human oversight, quality controls and responsible AI practices.

This document does not replace contractual terms.

For the purposes of this document, “Artificial Intelligence” or “AI” encompasses all AI-systems or services such as large language models (LLMs), machine learning algorithms, natural language processing (NLP) tools, or any other AI-based technology.

At a glance - what clients can rely on

  • No training on client data. YouGov does not permit client data, prompts or outputs to be used to train or improve underlying AI models.
  • Enterprise-only AI deployment. AI models are accessed only through AWS-managed services (Amazon Bedrock) or direct enterprise API agreements with model providers.
  • Confidentiality by default. Client prompts, inputs and outputs are treated as confidential and handled under YouGov’s contractual and security controls.
  • Human accountability and oversight. Human review and quality controls are applied at the system and workflow level, and to outputs where appropriate to the nature of the product or use case.
  • Transparency. AI-enabled client facing features are disclosed within YouGov products and platforms, and to panel members through their contract with YouGov, our Privacy Notice and, where appropriate, just-in-time notices.

How AI is used at YouGov

YouGov’s deployment of AI can be broadly grouped into the following categories, depending on the research use case and product context:

1. AI that does not process client datasets

This category applies where AI-enabled functionality operates on YouGov’s proprietary, syndicated datasets or internal systems operating only on YouGov-controlled data, rather than on client-provided data.

Typical examples include:

  • Syndicated platforms such as YouGov BrandIndex and YouGov Profiles, including AI-enabled features such as search, summarisation and persona-style exploration, which primarily operate over YouGov’s proprietary data.
  • AI-powered analytics and research support tools forming part of YouGov products where the underlying dataset is YouGov-owned and governed, for example, YouGov BrandIndex Voices.

How AI is used in this model:

  • AI acts as an interface and analysis layer to help retrieve, explore and synthesise insights from YouGov’s data.
  • Outputs are derived from aggregated, verified and structured YouGov datasets.
  • AI systems operate on YouGov data, not client datasets.

2. AI that receives client data in prompts

This category includes certain AI-enabled features, add-ons or functionality within YouGov’s subscription and platform products (for example, conversational interfaces within our platforms) that allow users to input natural language prompts which may contain client-specific or business information, even where the underlying data being queried remains YouGov-owned.

How AI is used in this model:

  • Client prompts are treated as instructions to guide retrieval, summarisation or exploration of insights from YouGov’s data or system logic.
  • Prompts may be used to retrieve or summarise YouGov data but are treated as confidential and are not used for underlying model training improvement.
  • The AI does not ingest or analyse client-owned datasets.

3. AI that processes client datasets

This model applies where YouGov provides its platform and AI-enabled functionality to support analysis of client-provided or client-controlled datasets, for example in platform-as-a-service (PaaS) deployments or custom research delivery via YouGov platforms.

How AI is used in this model:

  • AI-enabled features may process client datasets for the purpose of analysis, summarisation and insight generation.
  • Access by underlying AI models to client data, prompts and outputs is controlled through contractual agreements which prohibit client data, prompts or outputs to be used to train or improve underlying AI models.

Enterprise AI model deployment and provider controls

YouGov uses controlled, enterprise-grade AI model deployment. AI models are accessed only through:

  • Amazon Bedrock (AWS-managed AI services), or
  • Direct enterprise API agreements with model providers (such as OpenAI, Anthropic and Google Gemini).

Amazon Bedrock deployments

Where Amazon Bedrock is used, AI models operate within AWS’s managed service environment, with logical and contractual segregation between YouGov and other AWS customers. Client content and AI outputs are processed within AWS’s controlled service boundary and are not made available to, or used by, the underlying model providers for training or model improvement.

Enterprise API deployments

Where YouGov contracts directly with model providers, access is governed by enterprise API terms that:

  • prohibit the use of client content or AI outputs for underlying model training or improvement;
  • restrict processing of such data solely to the provision of the contracted services; and
  • impose confidentiality, security and audit obligations aligned with YouGov’s contractual commitments to clients.

YouGov does not use consumer-grade or public AI services for client delivery or production research workflows.

Data protection, confidentiality and security controls

YouGov applies data protection and security controls across all AI-enabled functionality, aligned with its contractual commitments and regulatory obligations.

Client data, prompts and outputs

  • Model access via secure, enterprise-grade deployment routes and controlled infrastructure.
  • No model provider receives or retains data for training.
  • Client prompts, inputs and related outputs are treated as confidential in accordance with YouGov’s contractual confidentiality obligations.

Model training and service improvement

  • YouGov does not permit client content or AI outputs to be used to train or fine-tune underlying models.
  • Client interaction logs are retained only as required for product functionality, safety monitoring, auditability and troubleshooting.
  • Aggregated and anonymised usage data may be used to improve product performance, reliability and safety controls, but not to train foundation models.

Personal data

  • Where applicable, YouGov’s obligations when processing personal data provided by our clients for research purposes, e.g. the provision of sample for use in a study, will be addressed in a data processing agreement.
  • YouGov’s processing of clients’ employee user data is described in our Privacy Notice.
  • YouGov puts in place appropriate controls in AI driven data workflows to align with data minimisation principles and avoid unnecessary surfacing of personal data from our proprietary data sets.
  • Where possible, we take steps to restrict data storage and processing to certain geographies. Where personal data is transferred to other territories, these transfers are safeguarded through ‘adequacy’ decisions and Standard Contractual Clauses.
  • Where AI services involve third-party model providers acting as processors or sub-processors, transparency is provided through YouGov’s standard data privacy documentation, including our Privacy Notices, list of core service providers, and data processing agreement, as applicable.

Core commitments

Regulatory awareness and compliance

Our AI-supported research is designed and delivered in line with the same rigorous research principles that underpin our long-standing methodologies grounded in impartiality, quality, integrity, reliability and comparability across all our products and services. YouGov monitors the development of applicable AI-related laws and regulatory frameworks, including the EU AI Act and other emerging standards. Our AI deployment is designed in line with the EU AI Act’s risk-based framework, with governance controls applied proportionately to the nature and context of each use case, rather than on a one-size-fits-all basis. Our approach will continue to evolve as regulatory requirements develop.

Responsible AI, bias mitigation and safety controls

Although YouGov does not train or fine-tune underlying foundation models, it applies layered risk-based controls across AI deployment.

  • Bias mitigation and fairness
    • A multi-model, agentic / deterministic AI framework helps to ensure outputs refer only to the relevant dataset.
    • Validation and representativeness checks are applied by YouGov research teams where appropriate to support fairness and quality of outputs.
  • Safety and reliability
    • YouGov’s AI functionality is not permitted to be used by clients for autonomous decision-making.
    • As part of YouGov’s product design process, measures are put in place to minimise hallucinations, and to enable citation of data source where appropriate.
    • Where data does not support an answer, the system is designed to surface that limitation.
    • Model behaviour is tested using internal benchmarks, industry evaluations and human review.
  • Human oversight and quality control
    • YouGov’s internal governance framework, which sets expectations for responsible use by employees, prohibit reliance on AI alone for decision-making.
    • AI-generated outputs are subject to quality control processes and appropriate human oversight.
    • Key requirements, risks and constraints for AI tools are identified and kept under review as part of product development and deployment.
  • Transparency and explainability
    • Explainability is provided at the system and data level rather than the proprietary internals of third-party models. This means YouGov can explain, at a feature and workflow level, what data sources are used, how the AI feature is designed to operate, and how outputs are generated and constrained, even where the underlying model architecture itself is not accessible or modifiable.
    • AI-enabled client features are disclosed within product interfaces, where appropriate to the nature of the feature and user context.
    • Where AI is used in data collection or interactions with panel members, transparency is provided through YouGov’s contractual terms with panel members, Privacy Notices and, where appropriate, just-in-time disclosures within the relevant interfaces.
  • Client validation and accountability
    • YouGov applies internal governance, quality control and human-in-the-loop processes to the design, testing and deployment of its AI-enabled features to support responsible system behaviour and risk management.
    • Clients are expected to apply appropriate validation and human review before relying on AI-generated outputs for their specific business, strategic or commercial use cases.

YouGov continues to develop and formalise its AI governance arrangements as AI capabilities evolve, applying controls proportionate to the nature and risk profile of each use case.

AI in YouGov’s internal operations and tooling

YouGov uses AI-enabled tools internally to support staff efficiency, consistency and quality in areas such as survey scripting, research automation, internal knowledge support and general productivity.

These tools are subject to YouGov’s governance framework on acceptable use, Information Security policies and Data Protection framework. Employees are not permitted to input client-confidential information into unauthorised systems, including public or consumer AI services.

Across all AI-enabled internal tooling:

  • AI is used as an assistive capability, not as an autonomous or decision-making system;
  • AI-generated outputs are subject to human oversight where appropriate; and
  • AI tooling is deployed within YouGov’s own secure environments, or under enterprise agreements with underlying model providers which prohibit use of data for model training or similar.

For further queries, please contact your principal YouGov account manager.