The interesting question about non-fungible tokens is not what they were used for during one loud period. It is what changes about a market when the units stop being interchangeable, because that single property invalidates almost everything a trader knows about how a price forms.
A fungible unit is interchangeable with any other unit of the same thing. One unit of a currency is the same as any other, which is why a single price can exist and why a single order book can serve everybody. Non-fungibility removes exactly that: each unit is distinguishable, and no two are substitutes.
The technical implementation is unremarkable. A contract tracks which identifier belongs to which address, exactly as a fungible contract tracks how much belongs to which address, and the difference is a data structure rather than a philosophy. Everything people find puzzling downstream traces back to that one change.
What is genuinely interesting is what it breaks. Order books, price discovery, liquidity, arbitrage and valuation all depend on units being substitutable, and when they are not, each of those has to be rebuilt in a different form or abandoned.
One unit of a currency is the same as any other, which is why a single price can exist. Remove that and the price has to be rebuilt differently.
An order book aggregates offers for the same thing at different prices. If every unit is different, offers cannot be aggregated, because a bid for one specific item is not a bid for another. What exists instead is a set of individual listings and individual offers, which is a classified advertisement structure rather than an exchange structure.
The consequence is that price discovery works completely differently. In a fungible market, the price is continuously visible because there is always a bid and an offer. In a non-fungible market, there is a price only when a specific item transacts, and between transactions the value of an item is an inference rather than an observation.
That is why floor prices exist as a concept. A floor is the lowest asking price in a collection, and it functions as a proxy for a market price precisely because no market price exists. It is a useful summary and a bad statistic, for reasons worth being explicit about.
A floor price is the lowest price at which somebody is asking, which makes it an offer and not a trade. Nobody has agreed to it. It can be moved by one seller listing one item cheaply, and it tells you nothing about what the item above it would fetch.
It is also structurally asymmetric in a way that flatters. It reflects the most motivated seller, which means it responds immediately to selling pressure and only slowly to buying interest, and multiplying it by the number of items to produce a collection valuation assumes every item could sell at the lowest asking price simultaneously, which is not a scenario that exists.
The honest alternatives are all about actual transactions: recent sale prices, the distribution of sales rather than the minimum, and the volume behind them. Those are harder to compute and less catchy, which is precisely why the weaker statistic became the standard one.
Liquidity means being able to sell close to the last price without waiting. Non-fungible markets are structurally illiquid, and not because of a lack of interest: they are illiquid because a buyer for one item is not necessarily a buyer for another, so matching requires finding the specific person who wants the specific thing.
This produces a distinctive pattern. Transactions cluster in bursts when attention arrives and stop almost entirely when it leaves, and the gap between an asking price and a completed sale can be enormous and can persist for months. Anybody used to a continuous market underestimates how different that is until they need to exit.
Several mechanisms exist to manufacture liquidity where it does not naturally exist, from pooled arrangements that accept any item from a collection to instant-offer systems that quote a low price for immediate settlement. Each is a way of converting a specific item into a fungible claim, and each charges for doing so, which is the honest price of the underlying illiquidity.
This is the part most often misunderstood, and it is worth being precise. A non-fungible token typically contains an identifier and a reference to where the associated content lives. It does not usually contain the content, because storing anything substantial on a chain is expensive and most chains are not designed for it.
So the token points somewhere. If it points at a conventional web address, the content depends on somebody continuing to host it, and there are many cases of the referenced content disappearing while the token remained perfectly intact. If it points at a content-addressed system, the reference is tied to the content itself, but somebody still has to keep a copy for it to be retrievable.
The practical check on any specific token is simply where the reference goes and who is responsible for what is at the other end. That is a five-minute investigation and it distinguishes a token whose content is durable from one whose content depends on an ongoing arrangement that may or may not continue.
The token usually holds a reference, not the content. Where that reference points, and who keeps what is there, is a five-minute check.
Holding a token is holding an entry in a contract saying an identifier belongs to your address. What that entitles you to, beyond the ability to transfer the entry, is determined entirely outside the chain by whatever terms the issuer set, and those terms vary from extensive to nonexistent.
The chain enforces the transfer, and it enforces nothing else. It cannot enforce a licence, a commercial right, a claim on revenue, or access to anything, unless a separate mechanism reads the chain and grants that access, which is a system somebody has to build and maintain.
This is not a criticism of the design, it is a description of what the design does. The confusion arises when the marketing around a token describes rights, and buyers assume the chain guarantees them because the chain guarantees the ownership entry. Those are different guarantees, and only the second is automatic.
Many collections were designed so that a percentage of each secondary sale returns to the creator, and for a period this was described as a solved problem. It was not solved, and how it came apart is a useful lesson about what a chain can and cannot enforce.
The payment was not enforced by the token contract in most implementations; it was honoured voluntarily by the venues where sales happened. When venues began competing on cost, some stopped honouring it, and creators discovered that a right they believed was enforced by the protocol was in fact a convention maintained by intermediaries.
The general lesson generalises well beyond this case. Whether something is enforced by the contract or by convention is a question with a definite answer, and the answer is frequently not what the surrounding description implies. It is worth asking about any mechanism described as automatic.
Setting aside collectibles entirely, a unique, transferable, publicly verifiable entry is a useful building block for several things that have nothing to do with art. Identity credentials, event admission, membership in something, records of provenance for physical goods, and the representation of a specific position in a system where positions are not interchangeable.
That last case is the one closest to trading. Some positions genuinely are not fungible, because they carry different parameters, and representing them as distinguishable tokens rather than as balances is the technically correct choice. In those uses the non-fungibility is a requirement rather than a feature.
What these uses share is that the uniqueness is doing real work rather than being the point. Where uniqueness is the entire proposition, the market inherits every problem described above; where it is a technical requirement of something else, it inherits far fewer.
Five questions, none of them about the picture. Where does the reference point and who keeps what is there. What does the contract actually let the holder do, as opposed to what the description says. What proportion of items ever transact, which is the real measure of whether a market exists.
Then: how concentrated is holding, because a small number of addresses holding most of a collection means the visible price reflects a very small free float. And what does the sale distribution look like, not the floor, because the floor is one seller's asking price and the distribution is what people actually paid.
All five are answerable from public data in under an hour, and all five are routinely skipped. This article takes no position on whether any of these are worth holding, which is not a question a mechanism description can answer, but the questions above are the ones that determine what you would be holding.
A contract entry recording that a specific identifier belongs to a specific address, where each identifier is distinguishable and no two are substitutes. The implementation is unremarkable; what matters is that non-substitutability invalidates order books, continuous price discovery and most valuation methods.
An order book aggregates offers for the same thing. If every unit differs, offers cannot be aggregated, because a bid for one item is not a bid for another. What exists instead is a set of individual listings, which is a classified advertisement structure rather than an exchange.
It is the lowest asking price, so it is an offer nobody has agreed to, movable by one seller listing cheaply. It responds immediately to selling pressure and slowly to buying interest, and multiplying it by item count assumes every item could sell at the minimum simultaneously.
Because a buyer for one item is not necessarily a buyer for another, so matching requires finding the specific person who wants the specific thing. Transactions cluster when attention arrives and stop when it leaves, and the gap between an asking price and a completed sale can persist for months.
Usually not. It typically contains an identifier and a reference to where the content lives, because storing anything substantial on a chain is expensive. There are many cases of referenced content disappearing while the token remained intact. Check where the reference points and who maintains it.
Only what the issuer's terms grant, which varies from extensive to nonexistent. The chain enforces the transfer of the entry and nothing else: it cannot enforce a licence, a revenue claim or access to anything unless a separate system reads the chain and grants it.
In most implementations the payment was not enforced by the contract but honoured voluntarily by the venues where sales happened. When venues competed on cost, some stopped honouring it. The lesson generalises: whether something is enforced by contract or by convention has a definite answer worth checking.
Five things, none about the picture: where the reference points and who maintains it, what the contract actually permits as opposed to the description, what proportion of items ever transact, how concentrated holding is, and the distribution of actual sale prices rather than the floor. None of them requires an opinion about whether the thing is worth owning; together they establish what the thing actually is, and that question has to be settled before the other one can even be asked sensibly.