There’s a three-letter word floating around my world (and likely yours) that can make things very confusing. That word is NLP.
Depending on who you’re talking to, that acronym means wildly different things. It’s like saying “crypto” to a finance bro versus a digital artist. Same word, totally different context, right?
On one side, you have the world of artificial intelligence and machine learning. Here, NLP is Natural Language Processing: the engineering that lets a computer read this sentence, work out that I’m writing to you, and make a decent guess at whether the overall vibe is warm or cold. It’s the machinery behind the chatbots, translation tools and autocomplete we’re all glued to.
On the other side, you have the world of coaching, self-help and personal development. Here, NLP stands for Neuro-Linguistic Programming, an approach built on the idea that the words we use (to others, and to ourselves) shape how we think and act.
The tech world often rolls its eyes at the self-help crowd, and vice versa. But for me, as a strategist working at the intersection of AI and human-centric design, the tension between these two is the most fascinating part. Both are obsessed with language and what it does to the systems we live inside. They just differ wildly in how much of that they can prove.
✦ Neuro-Linguistic Programming: the self-help NLP
Let’s start with the one that feels more analog.
Neuro-Linguistic Programming dates to the 1970s, when Richard Bandler and John Grinder published The Structure of Magic (1975). Its pitch is that language, the nervous system and learned behavior are wired together, so change the words and you change the wiring. Think of it as the brain-as-operating-system theory: type the right commands, reprogram yourself.
Which brings us to the classic power shift you can make in a tough moment:
“I have to get through this project.” Sounds like a hostage situation, right?
“I get to finish this project.” Now that sounds empowering.
One swapped verb, and you’ve gone from captive to VIP. It’s the ultimate UX writing project for the mind: conscious, intentional, and deeply personal.
Here’s where the two NLPs split, though. There is real research on this kind of reframing, just not under the Neuro-Linguistic Programming banner. Wording can change how people feel about a task and even nudge motivation, but none of it is guaranteed.
Neuro-Linguistic Programming claims to totally rewire how you think, and that’s the part the evidence doesn’t back. It’s widely classed as pseudoscience, and a systematic review of its health effects found little evidence that it works.
Still, the instinct behind it isn’t wrong: how we talk to ourselves is worth paying attention to. So go ahead and try “get to” on your next dreaded task.
✦ Natural Language Processing: teaching machines to read the room
Now, the machines. Natural Language Processing is the discipline of getting a computer to work with human language: sort it, search it, translate it, summarize it, generate it. At its best, the result is a system that handles language in a way that feels effortless, human and clear.
What sets this NLP apart from the other one is that you can check its homework. Point a model at a pile of text, compare its answers with what a human would say, and you get a score.
A good example is a 2023 study in BMC Public Health. Researchers collected 180,128 tweets about COVID from four mid-sized American cities between January 2020 and June 2021, to understand vaccine hesitancy ahead of a vaccination campaign in Kansas City. Reading that many tweets by hand isn’t realistic, so two models did the first pass: one sorted the tweets into topics (vaccines, politics, mitigation measures, and local community issues), and one, built on a language model called BERT, tagged each tweet with an emotion.
That emotion classifier is where it gets fun. It had exactly four labels to choose from, joy, optimism, sadness and anger, and every tweet got one. Picture the control room from Inside Out with a different cast: Fear and Disgust are out, and Optimism got the part. To be clear, nobody ignored fear. The classifier simply had no separate fear label, so it couldn’t give “fear” as an answer. The human researcher who read 4,000 of the tweets, though, noticed plenty of it, especially in Long Beach.
The results even came with a plot arc: optimism rose when vaccines opened up in spring 2021, then gave way to anger within about a month in three of the four cities.
Two lessons hide in there. First, every model arrives with decisions someone made before a single tweet was read, like which four feelings were on the menu. Second, the authors treat the machines as a first pass, with a person reading to work out what the patterns meant. They’re also upfront about the wobbly parts: about three quarters of the tweets came from Long Beach alone, and Twitter users skew younger and more educated than the public overall. Good science tells you where it wobbles.
✦ NLP case study: Spotify’s Discover Weekly and its three friends
Most people assume Spotify’s recommendations are just math. Play counts in, playlists out. The reality is more social, and the best way I can explain it is with a group chat.
Going by public descriptions, picture Discover Weekly as three friends, each recommending music in a different way.
Friend one knows who you hang out with. “People with taste like yours also loved this.” That’s collaborative filtering: it looks at what listeners with similar habits play and save, and borrows from them. Netflix and Amazon lean on the same trick.
Friend two reads the group chat. This is the Natural Language Processing friend. It reads the language that orbits music: articles, blogs, and the way listeners name their playlists. Imagine tens of thousands of people filing a pop star under summer anthem, coquette energy or confidence boost. A text model can learn that those phrases travel with certain songs, and even a title like hot girl espresso run says something about when and why someone presses play. Researchers who dug through a million Spotify playlists found the signal gets stronger the more specific the title is: “Sabrina Carpenter essentials” tells a model more than “idk vibes” ever could.
Friend three has actually listened to the song. Audio models analyze the sound itself, things like tempo and timbre. Friend three earns their keep with a brand-new track, when the other two may have little to go on. A debut with no plays gives friend one almost nothing, and a song nobody has written about gives friend two almost nothing. How much data exists really depends on the song, since a new release from a well-known artist can inherit a fanbase and plenty of chatter on day one. Friend three doesn’t need anyone’s opinion either way. (That’s known as the cold start problem.)
So is Discover Weekly an NLP-powered vibe check? Partly. NLP is one friend in a three-person group chat, and a chatty one.
It’s also tempting to say this proves the coaching NLP right: the words we use reshape our behavior! Not quite. Here the words belong to the crowd, they’re read as data, and it’s the system, not the speaker, that acts on them. Language is still shaping the outcome, but through a different mechanism from the one the other NLP promises.
✦ NLP in AI strategy: why founders should care
Founders obsess over features and forget that the system has a voice. Nobody gets nostalgic about code that ran flawlessly behind the scenes. They remember the chatbot that kept them circling the help pages instead of actually helping, or the first screen after sign-up that felt cold and corporate when it should have felt like a welcome.
So the questions get oddly specific:
How should an AI use commas? (Or em-dashes; I still stand behind them!)
What’s the emotional weight of an error message? You want relatability and empathy, not a cryptic 404, 301 or 500.
How does a chatbot’s microcopy build trust and hold attention?
Choosing the voice is design work that sits on top of the model, not natural language processing itself. But the model has to be told how to sound, and the voice is the part users actually meet. In that sense, your language is your product.
Of course, charm only goes so far. Warm copy on a clunky product is just a nice blazer on a bad outfit. Does a friendlier tone actually build trust? Maybe! But treat it as a hunch and test it: do more people finish what they came to do, and do fewer of them end up emailing support?
✦ Both NLPs, one question
Here’s what I find most interesting. One NLP interprets language: it reads what people wrote, and you can check its reading against a human’s, score by score. The other claims to transform people through language, at a scale the studies so far haven’t backed up.
So the next time someone says NLP, ask which one they mean. Then ask the better question: how is language shaping the system I’m living in, or building, right now?