Starbucks Korea and the cost of unchecked AI
- Sam Schofield
- Jun 8
- 2 min read
Have you read the story of Starbucks' recent PR disaster in South Korea, their third largest market globally?
The controversy was sparked after marketers blindly launched a campaign on the back of an AI tool’s suggested slogan, according to a report from The Guardian. A quick search will provide background detail - here I want to focus on the cause and outcome. Suffice to say the ill-fated campaign referenced one of the country's most sensitive and traumatic modern historical events.

The response was practically immediate, sparking protests, a sacked CEO, a billionaire owner apology, government ministry ties cut, double-digit drop in card payments, hundreds of millions in refund requests, and a social media storm.
Was overconfidence in AI the root cause?
The case study (from May 2026) illustrates the pitfalls of over-relying on AI as a time saver and/or replacement for real human intuition, experience, and skill.
While it is undoubtably a powerful tool, using AI technology to cut corners in an effort to save time and money leaves organisations exposed. Any output from gen-AI requires thorough review, fact-checking and soundboarding, relying on experienced professionals to prompt, check, and further develop any content, campaigns, code, and creative output. In essence, using it as a tool, not a replacement.
Where did it go wrong for Starbucks Korea?
In the case of Starbucks Korea, reports claim the campaign's slogan was generated by AI and was approved without proper content review, with some managers reportedly not even opening the supporting materials.
The East Asian coffee purveyor has severely tarnished its reputation in a country that clearly loved the brand (South Korea has over 2,000 Starbucks stores for a population of just over 50m, compared with around 1,500 stores for 70m people in the UK). And with a campaign that was ultimately pulled within hours following immediate and severe backlash.
Will it be the last PR crisis caused by overconfidence in AI-generated output?
Almost certainly not.




