In June 2022 a small marketplace called PromptBase began selling sentences. Buyers paid 1.99 dollars for a string of words tested on DALL·E 2 or GPT·3, the seller kept most of the money, and PromptBase took a 20 percent cut, as TechCrunch reported at launch. Founder Ben Stokes told the site that the aim was to build tools to support prompt engineers, a job title that barely existed in public at the time.
The product was never really the sentence. It was relief. A model that could answer almost anything had left its users with a new and humbling problem, which was that they did not know what to ask. The person who claimed to know became valuable at once, not because the words were rare, but because the uncertainty around them was real.
The Bigger Story
Four years on, the stall has become a fairground. Reviews of PromptBase published in October 2026 put most single prompts between 3.99 dollars and 9.99 dollars, describe a subscription that unlocks a set number of prompts each month, and note whole packs and agent skills listed beside the single cards of the early days. PromptBase is the cleanest example, not the only one. Sellers also bundle prompts on Etsy and Gumroad, sorted by job and by wish: prompts for teachers, prompts for property agents, prompts for shops that need product copy, prompts for creators who need a month of posts and would rather not start from a blank box.
Read the listings as a customer would and a pattern shows. The language leans on secrets, on mastery, on results that arrive without the slow work of testing. A pack promises to make a model sound like an expert, to write the book, to plan the business, to produce the image that stops the scroll. The buyer fills in the rest. Their stalled project, their thin client list, their hope that the gap between them and the person ahead of them comes down to a phrase they have not yet bought.
That is where the tarot comparison earns its place, and it is a comparison about method, not about costumes. A tarot reading takes a set of general symbols and, in talk with the client, finds the one meaning that lands. A prompt pack takes a set of general instructions and, once the buyer adds their own facts, finds the output that lands. In both rooms the client does much of the fitting work, and in both rooms the fitting work is what makes the result feel personal.
The card was never the whole product. The feeling that the card knew you was the product.
The scale of the older trade shows why the pattern is so durable. Pew Research Center surveyed 9,593 adults in the United States in October 2024 and reported in May 2025 that 30 percent consult astrology, tarot cards, or a fortune teller at least once a year. Tarot alone drew 11 percent. The market research firm IBISWorld puts psychic services in the United States at about 2.3 billion dollars in 2025, spread across roughly 103,000 businesses. Nobody built that market on proof that the cards can see the future. They built it on attention, on framing, and on a client who arrives wanting a story they can use.
Context
Psychology gave this pattern a name long before software gave it a storefront. In the late 1940s the psychologist Bertram Forer gave his students a personality test and then handed each of them what he said was their own personal profile. The profiles were all the same text, lifted from a book of horoscopes. The students rated its accuracy at about 4.3 out of 5 on average. Paul Meehl later called the pattern the Barnum effect, after the showman P. T. Barnum, and the name stuck.
A Barnum statement works because it is broad, mostly kind, and full of small doors the reader can walk through. You have a need to be liked. At times you doubt a choice you made. Almost anyone can find a true memory behind lines like those, and the memory then does the persuading. The statement did not know you. You supplied the part that felt known.
Good prompts can work the same way, and this is the part the sellers rarely put on the listing. A broad prompt that tells a model to act as a skilled editor, to ask about the audience, to use a clear structure and plain words, will produce work that feels tuned to the buyer once the buyer pastes in their own draft, their own product, their own students, their own town. The prompt supplied the frame. The buyer supplied the substance. Selling the frame as if it supplied the substance is the trick, when there is a trick.
| Year | What happened |
|---|---|
| 1949 | Forer publishes the classroom test that shows students rating one shared vague profile as highly personal |
| 1956 | Meehl coins the name Barnum effect for the acceptance of vague statements as personal truth |
| 2022 | PromptBase launches in June, selling prompts for DALL·E 2 and GPT·3 at 1.99 dollars with a 20 percent cut |
| 2025 | Pew reports that 30 percent of adults in the United States consult astrology, tarot cards, or a fortune teller at least once a year |
| 2026 | Reviews of PromptBase report more than 330,000 listed prompts, most single prompts priced at a few dollars each |
There is a second fact that sits under the market and quietly limits every claim made on a listing. Prompts age. A prompt tuned hard to one version of a model can drift or break when that model changes, and marketplace reviews warn buyers to check which version a listing targets and when it was last updated. A tarot deck does not get patched overnight. A model does, and the spell that worked in March may read as noise in October.
Research keeps trimming the magic from the other side as well. A study led by researchers at UC Berkeley, reported by the university in 2026, tested common ways of asking models to reason across tasks and models and found that no single approach won everywhere. Simpler and more concise ways of asking were often as accurate as elaborate ones, and the best choice depended on the task and the model in use. That finding does not make skill vanish. It makes the idea of one secret wording, good for all work and all models, very hard to defend.
Different Perspectives
The case for the honest prompt seller is real, and it should be stated plainly. Testing takes time. A seller who runs a prompt against a model again and again, trims what fails, adds the examples that steady the output, and writes clear notes on where the prompt breaks has done work another person can fairly pay to skip. PromptBase adds one more layer that a random file sold from a social post does not. Reviews of the site describe listings being tested before they go live. Time saved is a dull miracle, but it is a real one.
The tarot side deserves the same fairness. Pew found that most people who consult astrology, tarot, or a fortune teller say they do it just for fun, and only a small share say they lean on it for major choices. Many readers present their work as reflection or entertainment, a way to look at a tangled problem from a fresh angle, and many clients leave with nothing more dangerous than a new way to frame a hard week. A reading can be a conversation tool without being a claim about the stars.
The buyer has a fair point too, and it is not that every purchase was foolish. Models are opaque to most users, the blank prompt box is a poor teacher, and the first hour with a new tool often decides whether a person keeps using it at all. A cheap prompt that gets a teacher, a shop owner, or a nurse past that first hour has value even if the same person could, in time, have written it alone. People pay for ramps as well as for rooms.
The critical case lands on the framing, not on every seller or every buyer. Listings that promise wealth, that call ordinary wording banned or secret, that sell a bundle of thousands of prompts no person could have tested one by one, are selling the Barnum effect with a checkout button. The output feels personal because the buyer made it personal, the failures get blamed on the buyer for using the words wrong, and the fix on offer is always the next pack. We have followed that loop before on this site, in Slop Is a Slur Now, where work pushed out without review ends up costing the person who receives it. A bought prompt that skips review belongs to the same family.
Why It Matters
The stakes look small because the prices look small. A few dollars for a prompt, a few dozen for a pack, a reading priced about the same as lunch. The habit being sold is larger than the receipt. If people learn that good output comes from finding the right incantation, they stop learning the skills that actually travel between models: stating the goal in plain terms, giving the facts the model needs, showing one good example, naming what to leave out, and checking what comes back. Those skills survive every update. An incantation survives until the next one.
That gap matters more as models move from answering to acting. A vague wish handed to a chatbot costs a weak paragraph. The same wish handed to an agent with tools, files, and a budget can cost real money and real mess, as the tools in our own coverage keep showing. The reader who was trained to hunt for magic words is the worst prepared reader for that world, because agents reward clear briefs and punish vibes. The tarot habit, carried into software, stops being charming when the software can book, buy, send, and delete.
There is also a quieter loss on the seller side. Real prompt craft exists, and it looks a lot like editing, testing, and technical writing. The spell framing buries that craft under claims it cannot carry, and it trains buyers to undervalue the honest version when they meet it. A tested workflow with notes, limits, and examples will always look less exciting than a secret that promises everything. Markets that reward the louder claim will keep producing the louder claim, and the careful sellers will keep having to shout over a fortune teller with a ring light.
Closing Reflection
None of this means the cards are empty, in either trade. A tarot spread can give a person a story sturdy enough to act on, and a bought prompt can give a person a start clean enough to finish something real. Value was never the question. Honesty about where the value comes from is the question, and in both trades the honest answer is the same. The meaning was not found inside the card. It was built, in the open, by the person who sat down with it.
So keep the prompt if it helps, and keep the reading if it helps. Just refuse the part of the sale that asks you to believe the words knew you in advance. A model does not need a priesthood, and a deck never needed one either. What both tools reward is the oldest skill in either room, which is a person willing to say what they actually want, in words plain enough to be checked.
Reporting draws on TechCrunch coverage of the PromptBase launch in July 2022 as reported at the time and in later reviews, Pew Research Center on astrology, tarot cards, and fortune tellers in May 2025, IBISWorld on psychic services in the United States in 2025, research on the Barnum effect from Forer and Meehl, marketplace reviews of PromptBase checked in October and November 2026, and UC Berkeley reporting on prompting research in 2026, checked November 2026.
Sources
- Pew Research Center, 3 in 10 Americans Consult Astrology, Tarot Cards or Fortune Tellers, May 2025: https://www.pewresearch.org/religion/2025/05/21/3-in-10-americans-consult-astrology-tarot-cards-or-fortune-tellers/
- IBISWorld, Psychic Services in the US, 2025: https://www.ibisworld.com/united-states/industry/psychic-services/4413/
- PetaPixel, You Can Now Buy and Sell AI Image Prompts on a Marketplace, September 2022: https://petapixel.com/2022/09/06/you-can-now-buy-and-sell-ai-image-prompts-on-a-marketplace/
- The Proof Desk, PromptBase Review 2026, checked November 2026: https://theproofdesk.com/promptbase-review
- UC Berkeley, How AI Prompting Shapes Accuracy, Efficiency and Cost, 2026: https://ieor.berkeley.edu/how-ai-prompting-shapes-accuracy-efficiency-and-cost/
- Barnum effect, background on Forer and Meehl: https://en.wikipedia.org/wiki/Barnum_effect
- Illustrations: Dirty Static
END OF TRANSMISSION
Stay curious. Question everything.
