The goal was to build a feature that makes use of AI technology that helps users create and iterate on social media content - images and text - in a way that was lightweight and wouldn't interrupt their usual workflows. It had to be non-intrusive, but readily available when the user needs it.

I ran 20+ user interviews with HubSpot customers who were active social publishers of content to social media, across multiple networks. The goal was to understand how we could compliment their existing social media content strategy with the help of AI. Users all said they needed something that could assist them in a way that felt seamless and didn't remove them from the social composing tool, but also took into consideration the fact that they, the user would need to make edits and "tweak" the final content to make it their own. The majority of users said they would most likely not publish the content generate by our LLM, but would use it as a "jumping off point" and would require their own human touch to finesse the final output for publishing.

After the initial Alpha and Beta releases, it was clear the concept was strong and of value to users, but the execution needed improvement. The quality of the text, and image-generation needed to be improved in order to make using this feature viable to our customers. Working with our engineers, I was able to help define the right prompts that were baked into each call made to ChatGPT from HubSpot to ensure the right kind of social content was generated, for each specific network, and things like hashtags, character counts and even "tone" were appropriate and what our users expected, and generated usable outputs.

By rolling out this feature to small groups of users and making incrementabl improvements to the outputs, this helped prove the value of the feature to our customers, and since it's rollout we've seen a dramatic month-over-month increase in the usage and publish rate of social posts that have used HubSpot's Social Content Creator.