FAQs (AI Studio)

What LLMs does Zeta use?

We use OpenAI, Anthropic (Claude via AWS Bedrock), and proprietary in-house models. A multi-model approach lets us pick the best fit per task.

Where are Zeta's LLMs hosted?

AWS and Azure.

Is client data used to train any LLMs?

No. Prompts sent to cloud LLM providers are not retained, and no client data is used to train or fine-tune any third-party or proprietary models.

How does Zeta screen AI inputs/outputs for security?

Inputs and outputs are passed through Prompt Security (prompt.security) to screen for malicious content, PII, prompt injection, and abuse. They don't store or train on any data.

What's Zeta's incident response process for data breaches?

We maintain a formal Incident Response Plan with a Company Emergency Response Team (CERT) that includes our privacy team. Covers identification, containment, remediation, and recovery.

Where is client data stored?

All client input and output data reside in the Zeta Marketing Platform. Not stored by third-party LLM providers.

What's the data retention policy for AI-processed data?

Data is stored for the life of the contractual agreement, per each client's DPA.

How does Zeta validate/test its AI models?

Agent evaluation suite with structured Q&A benchmarks, security testing for prompt injection, runtime screening via Prompt Security, plus reliance on Frontier Lab audits.

How does Zeta address accuracy, bias, and hallucination?

Agent evaluation suite for accuracy/fairness, ongoing refinement against benchmarks, prompt audit capabilities in development, and frontier lab model audits.

How does Zeta handle AI model updates?

Clients control when to enable new model versions in ZMP. We test beforehand and provide rollback plans.

Does Zeta have a formal AI data policy?

Yes — formal policies covering privacy and protection of data within AI systems, reflected in client-specific data processing amendments.

Is any data pre-processing required?

No, AI features work natively in ZMP.

At a high level, in layperson’s terms, how does Zeta’s AI work? 

Zeta’s AI pulls from a centralized repository of models that we call our Intelligence Store. The models in the Intelligence Store work together to solve business problems and are embedded within our recommendations and tools. If you want to interact with our AI directly, you should chat with Athena, our chat-based interface within the Zeta Marketing Platform.  

What is generative AI, and what are Zeta’s generative AI capabilities?

Generative AI refers to a type of AI that enables machines to generate unique and creative content, ideas, and solutions. Zeta’s generative AI capabilities include Athena and email subject line generation: 

• Athena by Zetaᵀᴹ is Zeta’s superintelligent agent, combining conversational intelligence, adaptive Generative UI dashboards, and seamless integration with the Zeta Marketing Platform. Powered by Zeta Answers, it transforms data into insight and insight into action. Through its Agentic App Suite, Athena by Zeta also enables marketers to plan, simulate, execute, and optimize campaigns—all from a single, personalized interface.

Athena can help with a variety of situations and use cases, including: 

• Platform onboarding: Athena is there to guide you through every step of the Zeta onboarding process. 

• Product questions: Athena can provide instant responses to your Zeta product-related inquiries, ensuring you have the information you need at your fingertips. 

• Analytics questions: Athena goes beyond simple product queries to help you navigate complex data and provide valuable insights. With Athena, you can confidently make data-driven decisions. 

And there’s more to come! Soon, Athena will give you access to our vast Data Cloud, providing you with an extensive pool of information for enhanced decision-making. Additionally, Athena will seamlessly integrate with various platforms, allowing you to perform native workflows directly within the chat interface. 

In a nutshell, Athena simplifies your journey by providing comprehensive assistance, answering queries, simplifying analytics, generating reports, and offering access to a wealth of data—all through a convenient chat interface. Experience the power of Athena and unlock a new level of productivity and efficiency in your martech endeavors.  

• Zeta’s email subject line generation uses cutting-edge technology from OpenAI, first-party data such as campaign history and tone selection, as well as Zeta’s advanced subject line scoring model, to suggest subject lines that inspire and delight. With this generative AI feature, you can not only create subject lines but also select a tone for subject lines, see the actual predicted performance before clicking send, tweak copy based on those predictions, and run A/B tests to validate optimizations.  

Which AI technologies does Zeta use?  

At Zeta, we leverage a diverse range of AI technologies to drive the best possible business outcomes for your organization. Our approach is designed to provide flexibility and avoid restrictions within any single ecosystem.

Our AI technology stack consists of a powerful combination of in-house models alongside leading third-party models from providers such as OpenAI and Anthropic, accessed directly and via platforms like AWS Bedrock. This multi-model strategy allows us to select the best model for each use case rather than being locked into a single vendor.

We also deploy and scale these solutions on robust cloud infrastructure, including AWS and Microsoft Azure, ensuring high availability, security, and performance.

What sets us apart is our ability to seamlessly integrate these technologies and make them work together harmoniously. Our expertise lies in optimizing model performance and orchestration, ensuring they collaborate efficiently to deliver results in the most effective and user-friendly manner.

With Zeta, you can trust that we bring together the best AI technologies—both in-house and from trusted partners—to create a unified solution that delivers exceptional results while prioritizing simplicity and ease of use.

What data and content does Zeta’s AI utilize/reference when generating content? 

Zeta’s AI uses a combination of first-party data (e.g., aggregated campaign results, analytics reporting) and third-party data (e.g., ChatGPT outputs) depending on the task. For example, for analytics use cases, it uses first-party data, whereas for creating image-based targeting (i.e., using an image of the object to be purchased to generate a corresponding audience who would buy it), it utilizes third-party data. To answer Data Cloud-based insights, it uses both first- and third-party.   

Is there a way to tell when/where AI was used in the platform and what content was generated by Zeta’s AI? 

We are adding iconography to our platform to make it clear which features are powered by AI. In addition, we are developing a feature where brands have the option to apply a ‘Co-Authored by AI’ label to external messaging/campaigns (e.g., emails, ads, etc.) that use AI.  

What type of content does Zeta’s generative AI create?  

Zeta currently only generates copy; we do not generate image- or video-based creative. Additionally, we do not process or compare across client creatives.  

How does Zeta handle copyright for the content it generates?  

For copy generation, Zeta follows the user prompt ownership model. Under this approach, when a user provides a prompt to generate content, the user then has the copyright claim over the end content because it was their prompt that generated it.  

Does Zeta have AI terms of service that guarantee that the referenced source data/content is properly licensed?

Yes

What types of decisions does Zeta’s AI make on my behalf?  

At Zeta, our AI technology is designed to assist and guide you in making informed decisions. However, it’s important to note that Zeta’s AI does not make decisions on your behalf. Instead, it provides valuable insights, recommendations, and information that can help you make more informed choices.

Think of Zeta’s AI as a knowledge base advisor or trusted assistant. It analyzes data, identifies patterns, and generates insights based on its analysis. These insights are then presented to you, empowering you with a deeper understanding of your business, customers, and market dynamics.

With this newfound knowledge, you are better equipped to make decisions that align with your business goals. Zeta's AI acts as a tool to support your decision-making process by providing you with relevant information, highlighting key trends, and suggesting potential courses of action.

Importantly, clients remain fully in control of how AI is used in their environment, including when and how model versions are adopted. Zeta provides the ability to test, approve, and roll back model updates as needed, ensuring that changes align with your expectations and governance requirements.

Ultimately, the responsibility for making decisions rests with you, the user. Zeta's AI is designed to augment your decision-making capabilities, helping you make more informed choices that can lead to improved business outcomes.

Does Zeta’s AI suite include any tools that offer transparency and mitigate any potential bias? 

Zeta provides multiple capabilities to promote transparency and mitigate potential bias across its AI suite. For audience targeting, Zeta offers tools that give visibility into the models being used, allowing users to drill down into individual models, inspect the attributes driving outcomes, and adjust or remove specific attributes as needed.

Beyond targeting, Zeta’s agent evaluation suite enables ongoing monitoring and assessment of AI system performance, helping identify unintended behaviors, bias, or drift over time. In addition, prompt auditing capabilities provide visibility into how prompts are constructed and executed, ensuring they align with governance standards and do not introduce unintended bias.

Together, these tools give users greater control, explainability, and oversight across the AI lifecycle, helping ensure responsible and transparent AI usage.

Is data input into Zeta’s AI systems communicated to external AI applications?

No. Zeta’s retrieval augmentation system transforms client data (e.g., prompts, reference documents, etc.) input into Zeta’s AI systems into a set of embeddings (i.e., numerical representations of the text and documents). These embeddings are readable only by LLMs and include only the minimal features of the input data required to generate outputs.

All inputs pass through Prompt Security, Zeta’s screening layer, which ensures sensitive information is appropriately handled, redacted, or controlled before any model interaction. No client data is used to train any models—either Zeta’s in-house models or third-party models.

When third-party models (e.g., OpenAI, Anthropic via AWS Bedrock) are used, providers do not retain prompts or customer data, and all processing occurs under strict enterprise privacy and security agreements. This architecture ensures that client data remains secure and confidential and is not exposed, stored, or reused by external AI providers.

Overall, how does Zeta ensure the privacy and security of personal data?

• Zeta ensures the privacy and security of personal data by implementing robust measures to safeguard against unauthorized access, data breaches, and misuse of personal data. This includes encryption, access controls, and regular security audits to identify and address vulnerabilities.

• Zeta protects individuals’ privacy and fundamental rights by adhering to applicable privacy regulations, laws, and standards. Zeta protects the security of personal data in a layered approach utilizing multiple technical and administrative controls.

What is Zeta’s technical approach/process for protecting the security of personal data?

• Zeta leverages several tools and processes, such as encrypting personal data at rest and in transit with AES 256 and TLS 2.0+, respectively.

• Rapid7 IDR is used for real-time scanning across the entire architecture with 24x7 “eyes on glass”.

• Personal data collected in the course of doing business is logically segregated and protected by role-based access control (RBAC) with very limited access and on a need-to-know basis only.

• In addition, state-of-the-art networking equipment, such as firewalls, is utilized and includes modules such as DLP, etc.

What is Zeta’s administrative approach/process for protecting the security of personal data?

• The Zeta information security programs are evaluated by third parties regularly.

• These evaluations include penetration and WAVA tests, internal vulnerability scanning, SOC 2 and SOX audits, and third-party cyber audits. These audits include regular UAR reporting and up-to-date training and governance.

• All Zeta employees and technical operations are required to follow proper change management and separation-of-duties guidelines.

• Zeta has also created a robust set of policies that must be adhered to by employees and are measured and tested on a regular cadence.