A style guide post pulled 1.3 million views on X last week. Not a celebrity scandal, not a market crash. Grammar advice. That is the state of the argument over artificial intelligence in October 2026: people will fight about verbs at scale.
Here is what the Associated Press Stylebook posted: “Artificial intelligence systems do not think, feel, want, or understand. Avoid language that gives them human characteristics.” Two sentences of routine guidance, the kind the account publishes weekly about state abbreviations and numerals. Those posts usually reach a few thousand people. This one pulled seven figures and started an argument that is still running days later, as reported in Micro Center News’ weekly AI roundup.
The fight itself is worth reading
The pushback did not come from random accounts. Alan Rozenshtein, a technical staff member at Anthropic, posted that he is “not sure a style book should be making such confident metaphysical pronouncements.” Employees at Google DeepMind and OpenAI objected along the same lines. Their position: nobody knows whether these systems think or feel, so a style guide should not declare the case closed.
The best counterpoint came from Futurism’s Frank Landymore, who noted that yes, philosophers debate whether models think or feel, but treating that debate as wide open “is probably lending the AI-is-sentient camp a little too much credence.” There is no scientific consensus that these systems have experiences. A style guide is not required to stay neutral on every open question in philosophy of mind before telling reporters how to write a sentence.
I land closer to Landymore, with a caveat. Whether large models have some inner life is a genuine open research question. But professional copy is not the place to litigate it. When your news article says ChatGPT “wants” something, you are being imprecise about mechanism, whether or not you mean to take a position on consciousness. Precision is the job.
Why a style ruling has real weight
AP Style is the default style authority for thousands of newsrooms, corporate comms teams, and government communicators. When AP updates a rule, style guides inherited by editors across entire industries inherit the update. Remember when AP accepted “they” as a singular pronoun in 2017, or lowercased “internet” in 2016? Those rulings quietly rewired style guides everywhere.
If this guidance holds, “ChatGPT thinks remote work is improving productivity” becomes a line your editor flags in red. Not because the sentence is philosophically wrong, but because it is now a style error in professional copy. That is a real downstream effect from one X post.
There is a marketing angle too, and it is the biggest offender of all. Product pages are full of lines like “our AI understands your customers” and “the platform learns what you want.” Under the new guidance, those are claims about mechanism presented as claims about minds. If you write copy for a living, this ruling lands on your desk, not just the newsroom’s.
The verb swap guide
The practical fix is simple. Describe what the model did, not what it felt while doing it. Output verbs beat mind verbs almost every time.
| Instead of | Write |
|---|---|
| thinks, believes | predicts, suggests, concludes (as in: the model’s output concluded) |
| feels, wants | is designed to, is optimized for, tends to |
| understands | processes, uses, takes into account |
| decides | selects, flags, routes |
| learns | was trained on, was fine-tuned |
| hallucinates | fabricates, produces inaccurate output |
A few of these deserve a note. “Hallucinate” is the interesting one. It is vivid, it is everywhere, and it borrows a word for a symptom of human illness to describe a statistical error. My honest take: in professional copy, “the model fabricated three court citations” is both more accurate and more alarming than “the model hallucinated.” The clinical euphemism softens a defect. The plain verb does not.
“Decides” is sneakier. “The agent decided to close the ticket” reads fine until you remember a decision implies deliberation. “The agent closed the ticket after the confidence check passed” tells the reader what actually happened inside the system.
Three rewrites you can copy
Before: “ChatGPT thinks the merger will close in Q4.” After: “Asked about the merger, ChatGPT’s answer leaned toward a Q4 close.” The reader now knows a question was asked and an output came back. No minds were borrowed.
Before: “The AI assistant wants you to confirm before it deletes anything.” After: “The assistant asks for confirmation before deleting files.” Same information, and now it is a description of an interface instead of a mood.
Before: “Claude believes this function is unsafe.” After: “Claude flagged the function as unsafe.” Flagged is what happened. Believed is a guess about the inside.
What to do now
One: search your current drafts for the mind verbs. Thinks, feels, wants, believes, understands, decides, learns. Any document, any editor, find-and-replace takes ten minutes.
Two: swap in the output verbs. Generated, produced, predicted, flagged, selected, was trained. If a sentence gets weaker after the swap, the sentence was leaning on the mind verb to sound smarter than its content. Rewrite the sentence.
Three: keep the casual usage where it belongs. Your car does not want to start and nobody files a complaint. Talking about your laptop being moody is how humans talk about machines and it is fine in conversation, in internal chat, on a podcast. The ruling is about published professional copy. Do not turn this into pedantry at dinner.
Four: add one line to your team style guide this week, before the argument resolves. Something like: “Describe model output with behavioral verbs (generated, flagged, predicted). Avoid verbs that imply mind (thinks, feels, wants).” If the metaphysical debate shifts in five years, you update one line. If it never resolves, your copy was precise the whole time.
The mind verbs were doing real work for us, and that is why they will be hard to give up. “The AI wants” is shorter than “the system’s reward structure selects for.” Borrowed sentience made the copy flow. The AP ruling forces the choice back onto us, and precise copy takes slightly more effort than sloppy copy always did. The argument on X will burn out. The verb swap is permanent.
Sources: Micro Center News (This Week in AI, Oct 2, 2026); Futurism. Original reporting on the AP Stylebook post and reactions via Micro Center News’ roundup.
