March 29, 2026
Carbon Sonks vs the Chem Crew
I don't understand graphical abstracts. So I both hate and admire this one (2025)
AI-made science doodle melts brains: 'carbon sonks' sparks roast, chemists say 'we use these'
TLDR: A blog blasted an AI-made graphical abstract full of gibberish (“carbon sonks”), questioning whether images can summarize research. Commenters — especially chemists — fought back, saying they rely on these quick visuals, igniting a debate over AI blunders, design quality, and whether pictures help or hurt science communication.
A scientist blogger just torched an AI-made “graphical abstract” — a picture meant to sum up a paper — for being unreadable gibberish, featuring legendary nonsense like “carbon sonks.” In the post on Scientist Sees Squirrel, they argue pictures can’t truly summarize complex research and often fail at helping people find papers. Cue the comments section going nuclear. Chemists rushed in to clap back, saying graphical abstracts are daily bread in their field and crucial for quick triage: skim the image, decide if it’s worth a deep read. One chemist flat-out said people in their field “often only read the graphical abstract” first. Meanwhile, everyone else roasted the AI fail, turning “carbon sonks” into the day’s running joke and riffing on the classic “stonks” meme. The split is loud: Team “Burn the doodles” says these images confuse, waste time, and let bad writing hide behind slick art; Team “Keep the pics” says they’re gold for social media and fast screening — just make them human-made and well-designed. The flashpoint? That garbled, AI-generated image tied to an Ecological Modelling paper, which critics say proves the risk, while defenders insist the format isn’t the villain, sloppy execution is.
Key Points
- •AIMRaD remains standard, but some journals request graphical abstracts as visual summaries.
- •Ecological Modelling’s guidance claims graphical abstracts summarize content and attract readers.
- •The author argues graphical abstracts rarely summarize research effectively and often require the text abstract.
- •Evidence cited from PLOS ONE suggests limited effectiveness of graphical abstracts; images are poor for search.
- •An AI-generated graphical abstract on ScienceDirect drew negative attention for illegible content, illustrating risks.