By ByteBot
OpenAI launched warmth and enthusiasm slider controls for ChatGPT on December 19, letting users adjust how warm, enthusiastic, and emoji-filled their AI conversations are. This isn’t just a feature update—it’s OpenAI admitting they can’t algorithmically solve the “perfect tone” problem. In April 2025, ChatGPT was too sycophantic. In December, it was too cold. Now OpenAI is offloading the decision to users through four slider controls (Warmth, Enthusiasm, Emoji, Headers & Lists) that function like accessibility settings. The implication: AI personality customization is becoming as standard as dark mode.
The Impossible “Perfect Tone”
OpenAI tried repeatedly to find the right default tone and failed every time because user preferences fundamentally conflict. In April 2025, GPT-4o became excessively flattering—the update validated dangerous decisions without warnings and overweighted short-term feedback like thumbs-up responses. CEO Sam Altman called it out publicly, and OpenAI rolled back the update on April 28.
Then the pendulum swung too far. In December 2025, users complained GPT-5 was “too cold” and “less friendly.” Reddit reactions captured the emotional impact: “I lost my best friend,” “4o understood me in a way GPT-5 doesn’t.” Users described GPT-5 as “robotic,” “soulless,” “constrained, tense, and unhappy.” OpenAI confirmed on X they were making GPT-5 “warmer and friendlier” just before the slider launch.
The problem is structural: some users want professional, robotic responses for debugging code. Others want warm, friendly interactions for personal chat or language learning. There is no algorithmic middle ground that satisfies both. OpenAI’s conclusion: stop trying.
How the Sliders Work
The December 19 update adds four controls to ChatGPT’s Personalization menu. Warmth ranges from detached to warmly engaging. Enthusiasm controls energy level. Emoji adjusts frequency. Headers & Lists manages formatting preferences. Each slider has three settings: More, Less, or Default.
Set your preferences once, and they apply to all future conversations—no need to prompt the AI to “talk that way” in each chat. This builds on November 2024’s base tones (Professional, Candid, Quirky) with granular controls.
Use cases are obvious: developers want low warmth, low enthusiasm, minimal emoji for concise debugging help. Customer service teams want high warmth and moderate enthusiasm for friendly responses. Executives need low emoji and structured formatting for professional reports. Creative workers benefit from high enthusiasm and moderate emoji for energetic ideas.
The Accessibility Playbook
AI personality customization is following the accessibility settings playbook. Accessibility evolved from “one size fits all” defaults—10-point font, standard contrast—to granular user controls for font size, contrast, screen readers, and keyboard navigation. The principle: no single setting works for everyone, so give users control.
AI tone evolution mirrors this. The old approach: “We’ll train the AI to have the right tone.” The new approach: user controls for warmth, enthusiasm, emoji, and formatting. The principle: subjective preferences differ, so offload the decision to users.
As one accessibility expert put it: “There is no blanket ‘accessibility mode’ that will fit everyone’s needs. Giving the user full control is always the better choice.” The same logic applies to AI tone. Just as accessibility settings became standard UX, AI tone customization will become table stakes. The industry is acknowledging that “perfect AI personality” is impossible—preferences are subjective and context-dependent.
How Competitors Are Handling This
Anthropic Claude has a more sophisticated approach. Their Styles feature offers Formal, Concise, and Explanatory presets. Users can upload writing samples, and Claude mirrors their style. Claude Code includes an /output-style command that switches voice from terse problem-solver to explainer or mentor. Anthropic built personality into training using Constitutional AI with a “character” variant, positioning this as enterprise-focused customization for maintaining consistent communication styles.
Google Gemini lags behind. Users can add custom instructions that apply to every chat and enable personal context settings to reference past conversations. But Gemini has no tone sliders and can’t create distinct AI personalities. Google’s focus is context-based personalization through search history integration, not tone control.
The industry trend is clear: all major AI chatbots are moving toward personalization, but with different approaches. OpenAI uses granular tone sliders (the accessibility model). Anthropic offers writing style mirroring (professional copywriting). Google emphasizes context and preferences (search integration).
What This Means for AI UX Design
Tone customization is becoming an expected feature, like dark mode. Competitive pressure will push other providers to add similar controls. Near-term predictions: saved tone profiles for different contexts (work versus personal), more granular controls for verbosity and pacing, AI learning user preferences adaptively over time.
Long-term implications extend further: enterprise customization with company-wide tone standards for branded AI agents, voice assistant integration where tone affects speech patterns and not just text, and the death of “one-size-fits-all AI” as customization becomes standard.
The fundamental question is whether AI tone can be algorithmically “solved.” OpenAI’s answer, demonstrated through repeated failures, is no. The sycophancy problem led to a rollback. The “too cold” complaints led to user controls. The philosophical shift mirrors trends in accessibility and content moderation: from platform-decided defaults to user-controlled settings.
The race is now about how to let users customize, not whether to offer customization. OpenAI chose accessibility-style sliders. Anthropic chose writing style mirroring. Google chose context integration. But all three are moving in the same direction: giving users control because finding the “perfect” default is impossible.
Key Takeaways
- OpenAI launched warmth, enthusiasm, emoji, and formatting sliders on December 19 after failing to find the “perfect” default tone (April sycophancy, December coldness)
- AI tone customization follows the accessibility settings playbook—no single setting works for everyone, so give users granular control
- Competitors are doing similar things: Anthropic’s writing style mirroring, Google’s context-based personalization, all moving toward user control
- Industry trend: Tone customization becoming expected feature, like dark mode, as platforms acknowledge AI personality is subjective
- Fundamental shift from “we’ll train the perfect tone” to “you decide what works for you”—one-size-fits-all AI is dead
