What actually happened when our blog went AI-first: A manager's account

In June 2025, HubSpot’s blog team transitioned to an AI-assisted writing model, balancing excitement about efficiency gains with concerns over impact on writers’ craft and ownership of their process. The team, composed of experienced writers who prioritize the art of writing over traffic metrics, faced the challenge of integrating AI without compromising quality standards they had spent years building.

Before AI, the writing process relied on individual workflows and editorial guardrails, allowing writers to develop unique insights and personality in their work. Drafting a standard blog post typically required a full day or more, with the majority of time spent on structuring ideas and clarifying explanations. The adoption of AI tools like Claude, ChatGPT, and NotebookLM was approached with flexibility, giving writers autonomy to experiment with integration into their processes.

Initially, AI appeared to accelerate drafting, but time-tracking data revealed a more complex outcome. While the quantity of posts remained stable, their length and depth increased, absorbing the time saved in drafting with additional research, sourcing expert quotes, and deeper topic exploration. Writers reported pressure to both write faster and produce more detailed content, prompting a recalibration of expectations.

AI expedited general-topic research but struggled with accuracy in areas requiring up-to-date information, such as competitor pricing or feature updates. Writers identified inaccuracies in AI-generated data, highlighting limitations in tools trained on outdated information or structured web pages. For competitor-focused content, manual research remained essential, as AI tools often produced authoritative-sounding but incorrect details.