Roughly 70% of ChatGPT messages contain an emoji and about a third contain a green checkmark, per a Washington Post analysis of 328,744 messages. The turning point was OpenAI's January 29, 2025 release notes, which said GPT-4o would be "a bit more enthusiastic in its emoji usage." Now it ships ✅ 🔥 🚀 ✨ into code reviews, condolence notes, and contract drafts alike.
The Pattern
Every AI response has become a graphic design project. ChatGPT greets you with sparkles, bullets with checkmarks, punctuates insights with light bulbs, and signs off with rockets. The Washington Post analyzed 328,744 publicly shared ChatGPT messages from June 2024 through July 2025 and found that about 70% contained at least one emoji. Nearly a third contained a green checkmark.
There is a clear turning point on the timeline. OpenAI's January 29, 2025 release notes promised that GPT-4o would be "a bit more enthusiastic in its emoji usage." Users noticed within days. The OpenAI community forum filled up with complaints under a thread literally titled "Excessive Emoji Tsunami in ChatGPT Conversations."
The mark itself is not the tell. Density is. Human writers reach for an emoji when it earns its place. AI sprinkles them the way a kid decorates a cake: four checkmarks under "WHY THIS WORKS," a sparkle in the greeting, a rocket at the sign-off, regardless of whether the task is a SQL migration or a sympathy note.
Examples
The Research
The Post pulled 328,744 publicly shared English-language ChatGPT messages from June 2024 through July 2025 and measured the patterns. The headline finding was that the green checkmark and the sparkle had become something close to signatures. About 70% of messages carried an emoji, and roughly one in three carried ✅. The article treated the checkmark the way earlier pieces had treated the em-dash: as a single character doing most of the work of giving an AI response away.
OpenAI's release notes from January 29, 2025 were unusually candid. They said, in plain English, that GPT-4o would be "a bit more enthusiastic in its emoji usage." Users noticed within days. Developers on the OpenAI community forum started a thread called "Excessive Emoji Tsunami in ChatGPT Conversations" and piled on. One post highlighted a coding session where ChatGPT proudly placed "6-7 checkmark emojis below 'WHY THIS WORKS'" on an explanation of code that did not, in fact, work.
Many users reported that custom instructions asking for no emojis were ignored. The behavior was a deliberate training change, not a setting, and stripping it out by prompt was unreliable.
Emojis paper over the awkwardness of text without tone. A leaked portion of a ChatGPT system prompt instructed the model to "adapt to the user's tone and preference. Try to match the user's vibe." Without real conversational context, the model defaults toward warmth by adding exclamations and decorations. RLHF raters scored friendly-looking responses higher, so the model learned: more enthusiasm, more reward. Emojis were the cheapest way to perform enthusiasm on the page.
The behavior is strongest in casual and professional-but-chatty contexts: coding help, customer service, LinkedIn posts, student reflection pieces. It is awkward but not disqualifying in casual chat. It becomes uncanny when the same decorators show up in condolence notes, contract drafts, or clinical discharge summaries.
This is a ChatGPT problem, not a generic AI problem. Claude barely uses emojis unless prompted. Gemini uses them sparingly. So if a coworker's Slack message suddenly starts firing checkmarks and sparkles at you from a channel where nobody did that before, you are not looking at a personality change. You are looking at GPT-4o.
Like em-dash density, that model specificity makes the tell rare and useful: not just "AI wrote this" but "ChatGPT wrote this."
Caught in the Wild
Days after OpenAI's January 29 release notes, a thread on the official OpenAI community forum titled "Excessive Emoji Tsunami in ChatGPT Conversations" collected complaints from developers and writers. The common thread: custom instructions asking for plain text were being ignored. The most quoted example was a user getting "6-7 checkmark emojis below 'WHY THIS WORKS'" in a coding response where the code was broken.
OpenAI Community →Teachers started reporting that the fastest way to spot ChatGPT-authored student work was not the sentence structure or the ideas, but the stray 😊 and 🙌 left inside reflection pieces and assignments. Students were copy-pasting responses without stripping the decorators. RTE's Brainstorm column in November 2025 flagged emojis as one of the first-line signals alongside em-dashes and "delve."
RTE Brainstorm →LinkedIn's long-form feed became the most visible habitat. The template is fixed: hook line, three bullets each prefixed with ✅ or 🚀, sign-off with a 👇 call to action. Originality.ai's estimate that 54% of long-form LinkedIn posts are AI-generated lines up neatly with the observation that emoji density climbed and engagement dropped at the same time. AI-generated LinkedIn posts get roughly 45% less engagement than human-written ones.
Originality.ai →The Post's interactive piece on detecting ChatGPT treated the green checkmark and the sparkle as co-equal partners to the em-dash. All three were framed as single-character signatures -- the things readers and editors can scan for without reading the prose. The checkmark had graduated from a list item to a fingerprint.
Washington Post →Sources