Researchers propose new method to make AI chatbots better at writing online review responses — reducing the manual work companies spend on customer replies

A research team published a new technique for fine-tuning large language models (AI systems that generate text) specifically for responding to customer reviews. The method addresses a real business problem: companies receive thousands of reviews daily but lack the staff to reply to most of them, leaving potential relationship-building and sales opportunities on the table.

The new approach fixes three common problems with current AI review-responders: hallucinations (the AI making up false facts), misalignment with what customers actually want to hear, and overly cautious responses that sound robotic. By teaching the AI to learn from human preferences in the review domain rather than treating all writing tasks the same way, responses become more natural and trustworthy.

For e-commerce companies, restaurant chains, SaaS platforms, and any business managing customer feedback at scale, this means potentially automating 80–90% of review responses instead of 20–30%, freeing up customer service teams to handle only complex complaints or escalations. Businesses that respond to reviews see 10–15% higher customer retention and conversion, so better automation here directly affects revenue.