There are very few mainstream sources one can trust when it comes to AI, and the NYT recently proved it’s not one such source. This is the article that irritated me, by Brian X. Chen, published on Aug. 13, 2026, and updated on Aug. 14, 2026: I Tested a Popular A.I. Slop Detector. It Felt Empowering.

It’s about Pangram, which was also covered by Niko Gallogly on July 29, 2026: Sick of A.I.-Generated Content? The ‘Slop Janitor’ Is Here to Help.

There is absolutely no reliable AI detector whatsoever, none, except for a cultivated brain. I know for a fact that there are countless false accusations of AI-generated essays and dissertations in academia, while numerous real frauds remain undetected.

But Brian X. Chen either was paid by Pangram or his brain melted into senseless slop. BTW, corruption or incompetence is rife. To quote from Pangram’s blurb: “An AI detector that actually works. Detect AI-generated content with 99.98% accuracy. Trusted by universities, schools, and enterprises worldwide. Proven the most reliable and accurate AI detector on the market by third party researchers, including the University of Maryland and the University of Chicago.” The university of my ass. (Incidentally, the correct form is “third-party researchers.”)

Fortunately, many readers proved to be smarter than Brian (kids, don’t be like Brian!). Selected comments:

1 ● MikeC, Adelaide

I’m a little disappointed that the word “slop” is being broadly painted across any AI-produced material. If someone uses AI to improve an article that is full of valid information, then are we that offended that it is not 100% human? I get that we don’t want info that is just thoughtlessly churned out filling the internet with garbage, but I have read many useful articles that I know are 100% AI but were useful in various instances.
Can we just reserve the word “slop” for anything that is careless and unnecessary, while acknowledging that AI can be a helpful tool if used responsibly?

2 ● MikeC, Adelaide

I’ve trained Claude in my tone and writing style. One thing I’m curious about is whether this detector can recognise where AI has written something that is backed by my own style influences.

3 ● Mr basilico, St ralph

These tools, including the one hyped here, are unreliable.

You know what is a better tool at identifying LLM products? You!

The models may get much better quickly, but right now they all have the same lifeless, overloaded quality. Every sentence is doing too many things. If you are exhausted after reading 2 paragraphs, it’s probably AI.

4 ● IL Ath, NYC

When everything is AI generated none of this will matter and everyone will be stupid because they will have stopped thinking.

5 ● Robin Garr, Louisville, KY

You’re abusing the useful term “slop.” Bad, easily recognizable AI output is slop by definition. Better AI – the kind that is harder to confirm – doesn’t earn that broad brush. Don’t be lazy. And don’t mock my em dashes. Or emoji. 😄

6 ● Nobody in Particular, Chicago

So they mess us up with unregulated AI that is actively spreading propaganda and misinformation, stealing the work of artists, deceiving people, and giving students new ways to plagiarize with impunity– then they sell special subscriptions to con artists, and sell cheating students extra programs aimed at masking the cheating from current methods of detection, then sell detectors to professors and teachers who need to stop cheating and to people who don’t want to be conned. It’s all a cash cow for them, playing all sides at once without regulation. Wake up, people– strict regulation is critical here.

7 ● CB, California

AI Slop. You, like most authors I hear complaining about AI slop use the term to really complain about AI generated writing. The reading public use the term to really complain about slop.
If an article were well written and interesting to me, I don’t care if it is AI assisted or generated or not. If it were poorly written and uninteresting I wouldn’t want to read it whether it was AI generated or human generated. That’s the difference you’re missing (or biased about).

8 ● Richard A Breuer, Pennsylvania

Wait till the AI detector starts making agreements with the AI slop creators to ‘pass’ on their work.

9 ● Terry, Columbus, OH

I look at this from two sides. I’m sure worthless articles are being generated by AI, but I’m also convinced that I’ve probably read factually based, well-presented AI articles. That said, I was reading garbage on LinkedIn in 2015 before AI. Good is good and bad is bad, regardless of human or machine.

10 ● JW, USA

I really hope that we can evolve the discussion of “AI Writing.” We are really talking about two different things: 1) when people outsource the thinking to AI and 2) when people use AI to help develop and refine their own original ideas.

Scenario 1 is bad! If you give an LLM limited context (i.e. a short 1-sentence prompt), it has very little information about your ideas or perspective, so it tends to fall back on the patterns and arguments it has seen in training. The result is often an amalgamation of things that have already been written on that topic. This type of writing is very noticeable once you know what to look for, it doesn’t add anything new to the discussion, and I doubt it’s going to have much value in the marketplace.

But it doesn’t have to be that way. It’s possible to provide enough of your own unique context to the model that it helps you develop and refine an idea without losing your voice. This is how I use AI for writing and I don’t actually save any time, but my final product is much better.

We should be encouraging this latter use of AI in writing. It can be immensely useful, especially for non-native English speakers and people with learning difficulties. Unfortunately, tools like Pangram and the watermarking feature in Claude outputs don’t differentiate between the two very well. As a result, I suspect the witch hunt against writers who use AI in any way is going to get much worse and we are all going to be worse off.

11 ● Vine Maple Farmer, Whatcom County

From this article and its comments, I don’t know if Pangram is a reliable AI detector, but as a software developer familiar with machine learning and the algorithms used for generative AI, I think an imperfect but fairly reliable detector is possible.

However, like Bret Stephens, I adjure writers to avoid generative AI. As habits go, genAI is the new three-pack-a-day for writers.

I discovered this several years ago when I started using Grammarly for copy editing my own writing. I soon found that my personal style was losing its character, not that my style was so great, but it was mine alone and was slipping away. So I quit.

Reading Benjamin Dreyer and other skilled writers regularly is much more satisfying. I like my productions better and better as I become more linguistically and stylistically sensitive.

Your brain is marvelously adaptive and powerful. Using AI to write is walking into an airborne pandemic without a mask or vaccine.

12 ● AD, WA

I tried Pangram with a page of writing I entirely wrote with no AI. Pangram claimed 75% AI generated. Nope, not sold on it at all.

13 ● Jen, California

Where AI detectors miss the mark is when it comes to collaboration, idea framing, and iterative work through the development of a project/idea. Also, who wrote the prompts, who directed the thinking, who shaped the idea?

There are miles of difference between:
1) “Write me an article on x.”
VS.
2) Iterative prompting with rejection, revision, and redirection.

Detection and authorship are not the same thing. That’s what’s missing.

I’m an editorial director and I edit others work, (sometimes heavily where needed) – does that mean they should not get credit? No.

Also, if I save someone’s life by Googling “How to do CPR,” does it mean Google saved their life or did we collaborate for a good outcome?

Seriously, that your article is taking a 99.9% accurate claim from a products website with a few obvious tests and ending there without deeper questions is the issue.

The question of collaboration is more nuanced.

14 ● Scott, New York City

Actually, my wife, a journalist and editor, has tested Pangram with known 100% human-written samples, and it often comes back with a result saying a good portion was written by AI.

It’s all got a long way to go.

15 ● JJ, NYC

Sad times for us humans who actually DO use em dashes. RIP.

16 ● AKJersey, New Jersey

Students and job applicants have forgotten how to write!
Pangram should be used on college essays and job applications.

17 ● AD, WA

It has been. My friend and former colleague who teaches college essay writing quit using Pangram because of too many false positives. I tested my own writing that had no AI-input whatsoever and Pangram claimed my writing was 75% AI. Its slop used to detect slop. My friend has gone back to blue books.

18 ● Fran, Oakland, CA

I ran some AI generated copy through Pangram that the model correctly identified as AI. Then I ran the same copy though ChatGpt, asking it to make the copy sound less like AI. When I ran the new AI version through Pangram, it rated the copy as 100% human generated.

19 ● My2Cents, NYC

Everyone will just move to a Chinese AI program instead. Using AI to detect AI is the craziest logic I’ve heard in a long time. Perhaps we should just consider the final version of something? AI is a tool like any other. I hate AI slop but it is very useful as a tool for storytelling.

20 ● MLG, California

Using a passage from Dickens was dumb. AI learns by harvesting information like that, particularly open source. A better example would have been to copy a page out of an old, obscure book that hasn’t been transcribed electronically and input that information into the program.
I am glad that the EU has forced AI companies to start watermarking their output. God knows, the USA would never do it, simply wring their hands and say there’s nothing we can do to stop it.

21 ● MR, US

I thought these AI detection “solutions” had been thoroughly debunked months ago? Call me skeptical…

22 ● Scott Summers, Los Angeles

I find it hilarious that anti-AI users are using AI to detect AI.

23 ● Karen C., Seattle

I tested Panagram with my human writing. I wrote 500 words and added two short sentences that were AI. When I ran it through Panagram, it flagged my document is 65 percent AI.
Clearly, more work ahead.

24 ● Oedipa Maas, San Francisco

NYT is part of the same attention economy that makes the AI slop farms profitable in the first place. An article about the dehumanizing effects of AI writing online next to ads for Google Ads every few paragraphs.

Institutions get to both diagnose the condition and be a vector for transmission.

25 ● RG, Massachusetts

I’ve got an a.i. slop detector in my head. Eyeballs and a brain. Duh.

26 ● blorx, hackneyed

As someone who is being pressured (yes, pressured) to use AI in my daily job as a writer and editor, I feel so torn. My company is actively soliciting examples of how we use AI to do more work faster. While the me who’s been doing this kind of work for more than 30 years is glad there’s a practically failsafe option for detecting AI-generated text, corporate me is worried that things I work on will be run through Panagram by others and then backlash will ensue. I feel like in my line of work, I can’t win. I primarily use AI for research and for structure and not for actual content generation. But those things aren’t very ‘sexy’ when it comes to trying to show the company bigwigs that what they’re paying for our AI licenses is worth the cost. Arrrggghhhh.

27 ● Bill, VT

AI is making the human brain irrelevant and obsolete. The content which we create with it is not a product of our thinking and feeling, much as we would like it to represent us.
Those of us who do our own critical thinking realize this. But the great mass of people who see generative AI as a chance to appear for thoughtful than they actually is what allows AI slop to dominate our online interactions.
As such, AI slop is a fraud against humanity.

28 ● Ncy, Denver, CO

There is something inherently hypocritical about using AI to detect AI so you can point fingers at people using AI. And Pengram has a serious false-positive error rate.

29 ● Mary, New York, New York

We’ve passed the tipping point with AI and now find it nearly impossible to discern real from computer generated. Some of us have felt this brewing for years when satire became harder to recognize with certain media and the absurdity of uneducated people unironically sharing downright wrong information.

But the rise of AI in the last few years is downright scary. We’ve allowed the tech bros to create a world where we can no longer trust our senses. We see and hear things that are not trustworthy and a lot of that information is processed by our bodies in really unhealthy ways, increasing our stress levels. This is in addition to destroying our physical environment to create this junk.

I miss the 80s. We had few computers and most of us didn’t spend half our day with our faces in a screen. The irony of typing this on a screen is not lost on me.

30 ● AD, WA

I checked out Pangram for detecting AI in writing by entering a page from the novel I’m writing. Pangram declared it was 75% AI generated. Considering I don’t use AI to generate any of my writing, how useful is Pangram at all? AI and slop are synonymous.

31 ● Tim Race, Connecticut

Great article, Brian! But would you please stop perpetuating the trope that em-dashes and bullet points are sure signs of AI slop? Your article uses em-dashes where appropriate.

Credible journalism, like the NY Times “Morning” newsletter and the family of Axios newsletters make effective use of bullets — which make text on a smartphone easier to read.

Many people who aren’t professional writers and editors now believe they can detect AI slop if they see an em-dash.

I would hope an influential writer like yourself wouldn’t fuel this false narrative.

32 ● David Thomason, Holt, Michigan

Interesting, but this is what I do: I work in Gemini with a minimally paid account (19.99/mo), and then check the output in an anonymous free account in Claude. Claude is more than willing to point out any deficiencies in Gemini. (BTW, ‘slop’ is the wrong word, stop using it). 🙂

33 ● CarsonDoc, Carson City

I agree. If it was slop, it would be obvious that the content was AI and nobody would need an app to assess it authenticity.

34 ● MM, Bethesda

Ok this is actually rubbish. I used Pangram last semester to detect my students’ AI work. To make sure I wasn’t falsely accusing anyone I first tested it with my own work written before the age of AI. In short, it was abysmal at detecting AI. And in fact it flagged one of my students’ work as not AI when I knew for a fact the kid was not capable of producing such an eerily polished paper. I was unable based on such results to credibly hold any student accountable. Later I found a piece in the WSJ by James Taranto which said that Pangram had serious errors and couldn’t for example detect between AI writing and the use of AI for editing.
An article like this is a terrible disservice to the public. It could lead to false accusations. There should have been at least some exploration of the hazards of using Pangram.

35 ● Sharon, New Jersey

Just one more subscription at $20 a month.

36 ● Susan, Maryland

This tool said my writing was AI written, but I have drafts of each piece that are evidence to the opposite. I guess my decades-developed writing skills are just like what AI was trained on. So now I get condemned as a fraud? I’ll do my best to write poorly so I’m not judged to be a fraud, I guess.

37 ● BKW, PA

One gap in this otherwise interesting discussion is the increasingly blurry line between “human” and “AI” writing. Many students, and plenty of professional writers, now routinely run prose through tools like Grammarly, which can substantially reshape syntax, diction, and style without generating the underlying ideas. In academic settings, at least, the important question may be less “Was this written by AI?” than “What kind and degree of machine assistance went into producing it?”

38 ● Linda, California

I use AI regularly, as an assistant. Ìt is only as good as the questions/prompts asked. It will produce an answer that I then read for accuracy, ask it to revise or rewrite if needed, and to create a clean docx. I have established canon rules, and I generally find it an efficient, useful tool.

The bigger issue applies to dishonest actors. When you have a President who controls media and lies as he breathes, and generates images of himself as Jesus, and has news outlets as an echo chamber, there is a line inevitably crossed.

I would prefer that the AI watermarks and detectors be used to identify falsehoods, lies and misleading information, rather than vilifying the use of AI as a research and business tool. Every company uses AI to scan resumes and make recommendations. Unfairly or not, the hypocrisy is glaring. Do what I say, but not what I do.

Pandora’s box is open and there is no going back. The real regulating work becomes who (or what), maintains the ethical line. Case in point, Doge used AI to gather (and steal) personal info and keyworded diversity and equality, and destroyed decades of important work in one fell swoop. That is the bigger, alarming issue. It can and is being used for nefarious purposes.

Copywrite protections on content in an open market is, sadly, becoming a thing of the past. AI is trainable – until it’s not. Pangram and other similar programs would be better used as a truth detector, and not another AI program to detect AI.

39 ● Lorenz, San Francisco

Once they publish a browser plug-in that highlights AI text on websites (like instagram or news sites) that company is going to be $$$$$.

40 ● Whitney Beach, Calif

Hmm, I use bullet points, emojis and em dashes. How can I be sure that I’m not a chatbot?

41 ● DylanTweets, Merrimack, NH

I’d suggest running the epigrams written by various octogenarian celebrities in the NY Times guest essay “Bob Dylan and Liza Minnelli Already Turned 80. They Have Thoughts for Trump.” through Pangram as I did. The results I got were, ah, surprising, to say the least. Pangram claimed that most of the essays except for Robert De Niro’s were either wholly or partially written by AI (two had too little a word count for Pangram to accurately measure). Personally I don’t buy Pangram’s results, but I’d e very interested in the Times’ response and what efforts it’s taking to ensure that published content is written by humans.

42 ● profwilliams, Montclair

I tested it out using old college essays from my wife, kid, friends, and past students (with permission), and it failed miserably.

Not sure what this writer is talking about. Honestly, after my experience, this seems like an AI generated Ad.

43 ● Dr. Schlitz, Mt. Moosilauke

I guess I’m not human by the author’s standards because I’ve been using bullet points and em dashes for my entire 30+ year career as a lawyer. You know, Microsoft Word automatically converts 2 hyphens into an em dash and 1 hyphen into an en dash. Word also has numerous built in “bullet” styles.

44 ● Buddhaa, Denver, CO

Why do you care if an article is written with an A.I.s help or not?. What’s more important is whether the facts mentioned are accurate. Does it contain inappropriate or misleading information or promote hatred and discrimination?.

A well researched and well written article is good regardless of whether the person had an AIs help or not.

Btw, a lot of people use em-dashes and bullet points to write, if you have spent any time in a corporate setting or read consultant documents you will know. The AI has learned to write like that from us!

45 ● Di, Pittsburgh

My problem with AI trackers for writing is that if you use spell check, it’s flagged as AI. If you copy and paste either a quote (and it is cited) or from another piece of your own work, it’s flagged as AI. I’m terrified for my son when he goes to high school and then college. He’s becoming an excellent writer like his father who received his PhD in history long before AI; however, some of these tools we now use to make writing easier, but are not AI, are still flagged as AI.

46 ● JM, Not DC

Surely I’m not the only one to notice the irony that the tech lord creators of the perfect disinformation machine have now deployed the perfect disinformation detector sold to their enslaved underlings for a mere $20 a month for the rest of our lives.

47 ● Marcel, The Hague, The Netherlands

Writings from me and many other autists, who write down their thoughts using bullet points and structured text blocks, are more and more wrongly identified as the work from AI’s/LLM’s.

Quite frustrating to explain and proof over and over again that I really wrote what I wrote. Lets hope thise “AI identifiers” will get more accurate towards the neurodivers way of formulating thoughts.

48 ● Theo, NYC

This seemed like a weird article praising Pangram – almost like an advertisement. No AI detection program is reliable and accurate – including Pangram. And in a nation without AI constraints and a madman determining AI policy direction, there won’t be any in the near future. We now live in a world of looted ideas packaged into slop.

49 ● Terri, Charlottesville, VA

The most important sentences are at the very end of the article: “‘Stop getting your news from social media,’ Dr. Farid said. ‘An assumption that most of what you are seeing is fake is probably pretty good right now.’”

50 ● Colm, Toronto

Panagram doesn’t flag AI “slop”, it flags AI content. All one needs to detect slop, whether AI or not, is your own judgement.

51 ● Bryan, Queens

Today in my campaign against AI, I used AI to test against AI. When the AI produced a result against AI that was comfortable with my personal narrative against AI, I believed the AI. Thankful I finally found a tool (AI) that helps me find things I don’t like (AI).

52 ● Upstate NY, Upstate NY

These tests all seem to use blocks of text, full sentences, etc. What AI is useful, and IMO legitimate for, is to collect and write summaries and similar research material, that the writer then uses to prepare drafts and then final text. By the end, the author wrote the text, but some words and phrases originally “thrown together” by AI will still be in the final copy. It’s like having an assistant gather and organize stuff for you, and then you use that to prepare an essay or article.

So is the final text AI produced? If Pangram says it’s 30% AI, but that’s only in scattered sentence fragments, is that really AI? Is it really human? I would really like to know where the line is drawn.

53 ● Joyce, Boston

Humans also produce plenty of slop. If AI produces a quality read that is accurate and factually correct, the only objection is whether it’s displacing humans. That’s a conversation worth having.

54 ● Kim Anderson, Atlanta

Exactly! I’d rather read well structured AI generated and human edited narrative than some of the poorly drafted meandering what’s-your-point drivel coming through all sorts of channels these days. But then there’s the discernment question; trusting anything published by a favorite source doesn’t make it true nor accurate. Use the (AI) tool; think for yourself.

55 ● MajorMajor, NY

Entirely depends on what the content is. If it’s about, say, a summary of last winter’s weather patterns in the northeast, no problem, who cares. But if content contains sentiment of any kind and is AI produced, that’s a deal breaker.

56 ● Sarah, NC

As a professional author I’ve tested Pangram multiple times. It has a massive error rate. Multiple times I’ve fed it an AI generated chapter with a few human written paragraphs. It flagged the human writing as AI and passed the rest of the AI generated text as human written. This is not an uncommon result.

It’s maddening to see Pangram regularly flag my own writing as AI, especially when articles like this praise its accuracy. Given the anti-AI sentiment in the literary world, it’s only a matter of time before I’m accused of publishing AI writing based on a faulty AI detector.

57 ● Ax, Westchester, NY

Same here, I tried Pangram several times and my (fully human-generated) content got flagged as AI over and over. I’ve been a professional for decades and this kind of overhyped tool can be devastating for a writer’s credibility.

I see they’ve integrated it into Substack, which seems like a lose-lose for anyone on that platform: you either get your content incorrectly flagged as AI, or you opt out and look like you’re hiding something.

It’s just another form of imperfect AI being lauded (by some) as miracle tech.

58 ● rmt, USA

A WSJ editor described these tools as “defamation machines”. We’re going to see a lot of journalism and academic careers destroyed by slop accusations, cheered on by people who believe that they are “anti-slop”.

59 ● Bram, NY

LinkedIn? In my experience pretty much every LinkedIn post is AI generated. Which would make sense: it is the perfect platform to induce superficiality and mediocrity. I wonder what the actual numbers are, but to me LinkedIn seems to be the embodiment of AI slop.

60 ● K. Kong, Connecticut

You can use AI to make basic corrections on a photo, color, light, clarity and so on. The results are quite good, fast, and no different than a photo editing tool but the image will be dinged as AI generated by these detection tools. It’s ridiculous. With writing, I now avoid dashes, bullet points and any sentence over 20 words. I feel that whenever I write, Big AI Brother is watching. I write to avoid AI dinging anything I produce. None of this liberating. AI detection is the modern Salem Witch Trial. No one wants that mark of shame. I’m sick of all of this. It’s too much.