The LCGC Blog: The Slop Era of Chromatography on Social Media

Across the wide social media landscape, the analytical chemistry community is most commonly found on LinkedIn posting recent research articles, discussing “hot topics” in the field, and sharing stories from recent meetings and conferences. Recently, I’ve been pleased to see some friendly discourse on research disagreements in the field, with respectful treatment of scientific work even if there is no consensus on the conclusions yet. At the same time that these positive aspects of social media usage have grown, the past few years have also seen a dramatic increase in the number of LinkedIn posts that include content generated by artificial intelligence (AI). Various members of the separation science community have discussed aspects of AI and ML (machine learning) as it relates to chromatography within the print and online formats of LCGC International: the impact in the classroom (both positive and negative), for use in method development (including retention time prediction) and troubleshooting, and as a way of aiding analytical chemists complete coding tasks for data analysis and simulation.

However, I have not yet seen anyone directly call out one of the most dangerous aspects of the crossover between separation science, social media, and generative AI: slop posts. I am sure most readers see at least one “slop-containing” post in their feed on a daily basis. They are easy to spot based on their use of cartoon-like figures designed to depict a specific message or purpose, bubbly letters with descriptive lists and emojis, videos with significantly distorted images, and/or pseudorealistic caricatures placed in absurd situations. Because social media algorithms guide creators and influencers to use multimedia content to drive interactions and impressions (comments, likes, re-posts, etc.), slop has proliferated in a way that surpasses most viral online trends. Although I believe there are many ways that AI/ML will benefit the separation science community in coming years, that does not mean we should not be cautious with the direct outputs of generative AI in the same way that we are critical of analytical data we collect.

Within our research community, I have observed an increase in a few types of posts on LinkedIn that cause me varying degrees of consternation:

The Summary Post
The “summary post” is the most common generative AI post that I see in my feed. This type of post is typically found when someone wants to update their connections on a new paper that was published, a conference presentation that went well, or provide some other newsworthy note. An easy way to spot this type of post is the excessive use of bulleted/numbered lists or several one-sentence paragraphs with excessive line breaks, emojis (especially as the bullets in the list), em dashes (“—”) to connect phrases in sentences, and a brief conclusion that links directly to the relevant material. This is the least “slop-like” of the posts described here, as it generally is a way for an author to quickly summarize a lengthier bit of material, but it still loses some of the personal touch once it has been generated by AI. I will note that this is the type of post that even many respected members of the research community regularly share, although usually with an understandable motivation to distribute the key points of their work to a broad audience rather than to just drive impressions. Thus, although this one bugs me personally, I can appreciate the reason for its existence and acknowledge that there are rarely blatant mistakes in these types of posts. The same can’t be said for the other two entries on this list…

The Diagram Post
The “diagram” type of content is most typically observed as a single graphical image or figure that may or may not contain a few descriptive sentences. The most egregious works have been graphical figures meant to describe how various components of a GC, HPLC, or MS work or demonstrate a specific analytical workflow.

The primary issue is that generative AI is still not very good at producing accurate images, which means that the description of underlying physical properties used to make measurements with instruments are wrong, components are placed in the wrong order or given the wrong label, and/or mathematical descriptions use made-up equations. Because the images routinely look like the graphics that are seen in textbooks, novices in the field can be easily led astray by mistake-laden figures. This is a true “quantity vs. quality” issue, as some content creators simply churn out AI slop images to saturate feeds rather than carefully considering the accuracy of their posts. However, these accounts (some of which are likely bots or AI agents) aren’t the worst threat, as verified leaders in the field that re-post such content without reviewing it themselves magnify the inaccurate content and lend it credibility, which can be worse due to the influence these renowned scientists have. Alternatively, I have been proud to see some colleagues directly post corrections on images that have mistakes, providing some context from experts in the field to people who come across such posts.

The Tutorial Post
The LinkedIn chromatography content that raises the most concern to me combines aspects of the previous two types of posts with the goal of teaching chromatography fundamentals that can be easily shared to multiple networks, thus creating an exponential amplification of impressions that aids the original creator. Again, bulleted lists full of emojis form the basis of the post, but here are usually concluded with a cartoon graphic showing some exaggerated version of a scientist performing some routine laboratory task while holding up a sign that summarizes their key points. The problem is that almost always, the information in the post is wrong, whether it involves incompatible stationary phase and mobile phase selection for a given separation, an approach to troubleshooting that would not solve a given problem, or a suggestion for method development that could actually risk damaging an instrument or column. Fortunately, I have not seen this type of content as widely shared by leaders in the field and instead have noticed many of them comment on inaccuracies much more frequently than is observed with “diagram posts”. The primary issue that leads to this type of incorrect information is that LLMs (large language models) are still not great at describing or summarizing high-level technical content. Although this will likely change based on the ever-evolving nature of generative AI, for now, it is best to take advice on experimental design and fundamental chromatographic theory with a grain of salt when you see it pop up on your social media feed from someone that you have no direct connections with.

Although I myself did not rely upon generative AI to share my thoughts on these topics, I did check with ChatGPT to see if there was overlap with my observations and opinions. I was not surprised that the output came in a form that mimicked some of the key signs of slop that I have described:

Author: “What are the primary issues with slop posts about chromatography on LinkedIn?”

ChatGPT
““Slop” posts about chromatography on LinkedIn tend to share a few recurring problems:

• Oversimplification → wrong conclusions: Complex ideas (e.g., retention mechanisms, selectivity, gradient effects) get reduced to catchy one-liners that ignore conditions, making them technically incorrect.
• Misuse of terminology: Terms from analytical chemistry—like resolution, efficiency, or polarity—are often used loosely or incorrectly, which confuses learners.
• Fake universality (“one rule fits all”): Posts present specific observations as universal laws, ignoring that chromatography behavior depends heavily on system, analyte, and method.
• Vendor or product bias disguised as education: Content is sometimes marketing in disguise, exaggerating performance claims or implying unrealistic outcomes.
• AI-generated or copy-paste errors: Low-effort content repeats common misconceptions or introduces subtle inaccuracies that non-experts won’t catch.
• Lack of peer review or correction: Unlike journals or lab environments, there’s little rigorous pushback, so incorrect explanations can spread widely if they’re confident and easy to digest.
• Visuals without context: Clean-looking chromatograms or diagrams are shared without experimental details, leading to misleading interpretations.

Overall, the issue isn’t just that the content is simplified—it’s that it’s simplified past the point of being reliable.”

I’m sure that with some prompt optimization that I could have eventually generated a more descriptive list similar to the one I wrote, but with the first pass that usually constitutes the types of posts that I have been discussing, the clear signs of AI like bullets and em dashes are evident. As analytical chemists, we know that we need to be critical of our measurements and ensure that proper techniques and methods are followed to ensure accurate results. In the same way, a little bit more care and effort in social media content generation may help prevent the spread of incorrect descriptions of the analytical chemistry theory, the proliferation of bad laboratory habits that can produce poor results (or worse, are unsafe), and the saturation of feeds with more and more slop.