The final product (so far)

A detailed definition of "good" was created to train HELM, a new generative AI designed to analyze vacation rental amenities, photos, and locations to generate high-quality descriptions. This framework ensures the AI produces content that is not only accurate but also aligned with brand expectations. The output is evaluated against key criteria, including:

  • Character count and sentence structure – Ensuring readability and conciseness.

  • Content goals that align with our branding – Maintaining consistency across platforms.

  • Tone that matches the site’s standards – Creating a cohesive user experience.

  • Accessibility and scalability – Making content inclusive and adaptable.

  • Overall evaluation metrics – Prioritizing trust, safety, cohesion, and clarity.

Why it works

This project established a Definition of Good to train HELM, a generative AI that creates vacation rental descriptions. By setting clear content standards, it ensured AI-generated text met business and user needs.

Key impacts:

  • Consistency & Branding – Aligned AI-generated descriptions with EG’s tone and style.

  • Quality Control – Defined criteria for sentence structure, clarity, and trust.

  • Scalability – Enabled high-quality content generation with minimal manual revisions.

  • Cross-Team Alignment – Provided ML scientists with a clear blueprint for training the model.

By shaping AI output from the start, this framework streamlined content creation while maintaining quality at scale.

What else?

This is an ongoing project with continued improvements to AI-generated content. We focus on refining outputs by identifying and addressing recurring issues:

  • Location accuracy – Ensuring correct neighborhood and area details.

  • Logical flow – Describing space layouts in a clear, intuitive way.

  • Natural language – Using human-like verbs and adjectives for engaging descriptions.

By iterating on these elements, we enhance the quality and reliability of AI-generated content.

This work is expanding globally to improve content quality for non-English partners.

Key goal: Demonstrate that creating content in local languages leads to better, more relevant experiences than relying on machine translation.

By prioritizing native-language generation, we enhance accuracy, cultural relevance, and overall user experience worldwide.

Behind the scenes

Daniele Cianfriglia

Content Strategist in Rome, Italy

Adam Johns

Senior Content Strategist in London, UK

Aaron Spencer

Content Strategist in Seattle, WA, USA

Jess Eggerth

Content Strategist in Seattle, WA, USA

This work requires cross-functional collaboration, and I feel incredibly lucky to be part of a team with so many talented AI thinkers. My role has involved working closely with machine learning engineers, content strategists, and developers to bring this project to life.