OpenAI Softens GPT-5 Tone to Be Warmer and Friendlier

OpenAI tweaks GPT-5 to feel warmer

OpenAI announced a subtle personality update to GPT-5 that makes the model “warmer and friendlier.” The change comes after a rocky rollout of GPT-5 and user feedback that some preferred GPT-4o’s responses. According to the company, the update adds small, genuine touches — phrases like “Good question” or “Great start” — without veering into flattery.

OpenAI emphasized these are subtle UX changes rather than capability upgrades. The company says internal testing showed no rise in sycophancy compared with the prior version, and executives used a recent dinner with journalists to steer attention to broader roadmap priorities beyond GPT-5.

Even so, the update signals how model personality affects perception. VP Nick Turley described the prior GPT-5 responses as “very to the point,” and OpenAI’s tweak aims to make interactions feel more human without undermining factual rigor or safety.

Why tone matters for products and policy

A model’s tone influences trust, usability, and perceived helpfulness. Warmth can improve user engagement and lower friction in customer support or educational tools, while excessive flattery or sycophancy can erode credibility and create safety concerns. For enterprises and regulators, the balance between friendliness and factual reliability is becoming an operational metric, not just a design preference.

Think of it like a receptionist who’s helpful but honest: a warm greeting makes the interaction smoother, but the receptionist must still route complex requests accurately and flag errors. The same trade-off applies to LLMs when deployed at scale.

Practical steps for teams responding to personality updates

• Run A/B tests that compare user satisfaction, task completion, and perceived helpfulness between tones.
• Track safety signals separately: hallucinations, inappropriate agreement, and overconfidence metrics.
• Give end users control: toggles for formality, verbosity, or directness help satisfy different audiences.
• Log interactions and surface tone-related anomalies for faster incident response and model iteration.

For regulators and public-sector teams, updates like this raise questions about transparency and change management: how are personality changes communicated to users, and what auditing is in place to ensure safety metrics remain steady?

How QuarkyByte approaches personality updates

When models change, organizations need more than a changelog. They need measurable rollout plans, UX experiments, and safety audits that translate research assertions into operational controls. QuarkyByte helps teams define the right metrics, run controlled experiments, and design governance that keeps products reliable while improving user experience.

As vendors iterate on personality and behavior, the smartest adopters will treat tone as a feature with SLAs and KPIs — not just as a cosmetic tweak. That shift turns announcements like OpenAI’s GPT-5 warmth update from PR into an operational challenge and an opportunity for product teams.

Bottom line: OpenAI’s change is small but telling. Companies should plan for ongoing personality adjustments, measure impact on trust and safety, and put controls in place so users get the warmth they want without losing accuracy or accountability.