Will AI replace poets? Why the voice is the product
A poem or a set of lyrics is bought partly for whose it is. An editor, a label or a client commissions a named writer because of a particular ear, a history and a point of view. A language model can produce rhyming lines in seconds. It cannot supply the person credited on the page, and in this trade the credit carries much of the value.
Two tasks show why. Writing original poetry means deciding what is worth saying about your own life or your own reading, then defending those choices to an editor. Writing lyrics to fit a melody means sitting with a composer or an artist, catching what they are reaching for, and rewriting on the spot when the chorus lands flat. Both tasks are judgment and relationship work as much as word work.
The market shape matters too. About 47,940 people hold this job in the US, and median pay is $76,910 (BLS, 2025). Projected employment change through 2035 is close to flat, at roughly -0.3% (BLS projections, 2025-2035). Much of the work is freelance and commissioned. So the pressure from AI tends to show up as fewer paid jobs for routine, functional verse and copy, rather than a whole occupation closing down.
What AI drafts, what it assists, and what stays with the writer
Start with the work AI can already carry. Generating draft lines on a brief and producing alternate rhyme, meter or title options are both jobs a model does without a writer touching the keyboard. Our coverage score explains how this is measured, and here is the share of task time in that group: 35%.
Then there is the larger band where a model is a tool in a writer’s hands. Researching a subject or a period before drafting, and reworking a stubborn draft by testing forms and cuts, both move faster with a model suggesting options the writer throws out or keeps. Share of task time where AI assists rather than decides: 43%.
The rest sits with the person. Co-writing in a session with an artist, reading and performing your own work for an audience, and negotiating how a piece may be used all depend on being there and being accountable for the result. Share of task time that still needs a human: 22%.
What the evidence actually shows
There is no direct, published test of AI against working poets, lyricists or creative writers in our evidence set. That is why the quality parity grade reads D, and why no parity number is given for this job. Plenty of people have opinions about machine-written poems. Opinions are not measurement.
A few things would settle it. Blind comparisons where editors and readers judge human and machine drafts on the same brief, without knowing the source. Acceptance rates for commissioned work at publishers, labels and magazines, tracked over several years. Repeat-commission rates, which tell you whether a client came back. Until that kind of work exists, this page scores the job on what the tasks look like and what tools can demonstrably do. Our full method sets out how each score is built.
When the timing could move
Most likely between 2036 and 2044 (8 in 10 of our scenarios). That window comes from our model of task shift, not from a prediction about any single poem or song. The replacement-year page explains what the range measures.
Two things could pull it earlier. First, demand for cheap functional verse and lyric filler, where a client only needs something serviceable, keeps moving to tools because the cost gap between generated drafts and commissioned work is wide. Second, no robots or hardware are needed here, so adoption only takes a subscription and a habit.
Two things hold it back. Credit and rights are the product: buyers want a named author who can sign, clear and stand behind the work, and publishing and music contracts are built around that. And audiences pay for a person. Live readings, co-writing sessions and artist collaborations are bought as much for presence as for text. You can see how this job compares with its neighbors on the side-by-side comparison tool.
How to stay needed
Lean into the tasks that sit in the human column. Co-write in the room with artists and composers, where the value is reading a collaborator in real time. Perform and read your own work, because bookings and festivals pay for the writer, not the file. Hold the line on rights and credit: know what you are licensing, and what you are not.
Two skills pay off. One is editorial judgment, meaning the ability to say why a line fails and what should replace it, on someone else’s deadline. The other is working fluently with these tools for research and revision while keeping the authorial decisions yours, which is also where the ethics sit: disclose what a client is buying.
What to do: keep a record of commissions, collaborators and performances, so your value is visible as a track record rather than a word count.
If you are weighing nearby paths, look at Writers and Authors, Editors and Music Directors and Composers. The wider picture for this line of work sits on the media and communication workers family page and the arts and entertainment sector page. For a different angle, our list of what the AI assistants say records how the models answer this question about jobs like this one.