Un•AI•ify

In the last 3 days I've received 2 (unsolicited) email responses to content I wrote for a newsletter. I get excited about this kind of engagement because I usually learn something + it's good to be read. I responded to both right away and within a day, I got follow-up responses from both ... and the responses had several tells of AI writing (i.e. @UnAIify and then also marked "100% AI-written" by @Pangram). What the hell. Is there any chance this kind of behavior becomes unacceptable?

You might like what I built here and shipped about a year ago: unaiify.com (@UnAIify). Un•AI•ify was inspired by the (former) CMO of Jasper's blog back in May 2025. A clip saved for posterity:

Sisyphus circa 2026. (Made with Gemini, remaining imperfections accepted)

Communication is: - the words used - how the words are expressed - the meta communications, e.g.: -- that the communicator felt the communication was needed, at all -- in the case of AI, that the communicator: --- couldn't communicate w/o AI --- didn't think it was worth their time to write/edit their thoughts --- doesn't understand the subject well enough to spot the inevitable BS in AI writing and edit it out (etc.) Disqualifying. cc @UnAIify

If you spot AI-isms (especially INXIYs), you're likely reading someone's AI writing. But if you find yourself chatting with a person who speaks in half-sentences, that's likely a byproduct AI overuse: You're on the receiving end of how they tend to prompt AI. cc @UnAIify

To date, I have yet to see tuning eradicate "It's not X. It's Y" phrasing, this includes having Claude, Chat, etc., use a tool like unaiify.com on AI drafts and rewrite for any caught patterns before answering. Also this has been happening with people who should know better for a long time, CMOs, CEOs, global comms leaders at McKinsey. Proof people don't read closely, including the (supposed) original posters. Also proof that hyperinflation of (fiat) content will drive a flight to quality of some kind. What will matter most soon enough is having primary sources of data, including an audit trail back to ground truth. If it can be prompted by AI into existence, the value of that content is plummeting to zero. Instead the cost of that content, ie spending the time to validate it's good content, is a price not many will be willing to pay. Perhaps given how spuriously people read anything now, we are already there.

The cost of technology is intermediation: Communicate from anywhere via phone, video, email, chat, whatever. Lossy and (often) asynchronous. Control is gained. Optionality from spontaneity is lost. Every screen puts us at a distance, etc. Ours is the Era of Intermediation.

It's not the em dash — it's the "It's not X. It's Y" that's the real tell. The em dash is the cherry on top of the AI slop sundae.

Heraclitus: "There is harmony in the tension of opposites, as in the case of the bow and lyre." Every LLM: "It's not X. It's Y." The former perspective is holisti, the latter atomistic. All the money and investment of the last hundred years (and more) has been fixated on the reduction and exploitation of narrowly defined models of reality. This is what @dr_mcgilchrist, @rorysutherland, and many others have realized. Mankind has been optimizing itself into a corner and specialized itself to fragility. Now, LLMs regurgitate all that abstraction. They offer the illusion of insight, and lots of people are falling for the con. And I don't know how any of this broken approach to existence gets fixed. "A human being should be able to change a diaper, plan an invasion, butcher a hog, conn a ship, design a building, write a sonnet, balance accounts, build a wall, set a bone, comfort the dying, take orders, give orders, cooperate, act alone, solve equations, analyze a new problem, pitch manure, program a computer, cook a tasty meal, fight efficiently, die gallantly. Specialization is for insects." -Robert A. Heinlein

Have edited plenty of AI-generated business writing. It's straightforward, mostly an exercise in elimination and pushing for more details and facts. LLMs are best understood as sophisticated BS engines. Much of the answers they provide will be accurate, some will be bullsh*t, and all of it is expressed with confidence. Editing is about culling the rhetoric, grounding the hand waving with examples. Relatively straightforward. The problem I see is people who lack critical thinking being too susceptible to the BS, getting duped into thinking the AI writing is good, when it's merely rhetorically effective (on them, the editor). (See also the work I've done with @UnAIify)

A very, very convincing compulsive liar, making it so hard to tell what's true. Not surprisingly, this is a poorly appreciated problem. teapawt.net/i/status/2046355785475…

The Standard Bullsh*t Take - content on 𝕏, Linkedin, Facebook, YouTube, etc., that invokes "It's not X. It's Y" phrasing and/or other rhetoric suggesting AI provenance. "SBST" content, whatever its merit, is tainted by its intent to persuade. Proceed accordingly.

.@rorysutherland: "In engineering, you are peer-reviewed by reality." 👆 "Does it work?" vs. "Does it persuade?" "Does it work?" vs. "Is this bullsh*t?" Most of all: "What's the fallout if it doesn't work?" and "How do we know if it works?" (@nntaleb's skin in the game) The AI hype is driven most by its use for code. The code is written and tested. The code works, or it doesn't. This is not the case with so many other things that AI creates. Can you "test" the answer it spits out, the copy or the email draft it writes, the document it reviews and summarizes, the brief, and on and on. Do you have the experience to even know where it might be wrong? Non-code, non-mathematical AI creations are hard to test for all sorts of reasons. Communication is messy and nuanced like the people doing the communicating. Communication is also aimed at persuading the audience, to "win arguments" rather than "solve problems" as Rory Sutherland so wonderfully puts it. As people use AI to create difficult to test things, how might they be fooled into thinking the creations are good? Prompters, as revealed by their deference to the LLM to assist them, may not know enough about the subject matter they're getting help with. They very well don't know enough to identify where an AI-generated creation is wrong or incomplete. Nevermind how persuasive AI is with its knack for mirroring, pacing and leading, and sycophancy. AI is the palm-reader, the P.T. Barnum, the bullsh*tting con-man we want to believe. "Prompter's blindness" hits squarely on these two fronts: 1. On the subjects they know less about and 2. on the subjects of which they are readily influenced (i.e. that of which they're already convinced). Thus, outside of testable coding, AI creations have no first-line-of-defense, relatively low-cost means for filtering the bullsh*t from "what works," what's true or valid. Thus, AI creations get published, emailed along, go viral. The control against this is people who know better, and know more about the subjects in the prompter's blindspots. But almost necessarily, the know-how is limited to a smaller number of people, nevermind the impossible to fix asymmetry that comes with a combination of zero-cost scale on the Internet and instant-creation of AI. The amount of [stuff] being created that is half-baked, performative slop grows daily. Are we set on a path to drown in unvetted bullsh*t? "It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so." cc @UnAIify

Playing the detection game is stupid. Playing detective is smart. The difference is in having a set of heuristics that help you suss out what's worth attending to and what's worth dismissing. Get better at abductive reasoning. /s/ Builder of a detective tool: @UnAIify

Fiat content without market pressure to control supply ⤵️ Hyperinflation of content ⤵️ Flight to quality. blog.unaiify.com/p/the-hyper…