Skip to main content
A Vestiaire filter is decided entirely at our end. Nothing you tick is ever sent to the marketplace to decide a detection: the radar asks Vestiaire for listings, and your criteria are applied here, on every one of them. Your criteria do leave for the marketplace in exactly one place — the preview counter at the top of the form, which counts and never detects. That single fact explains almost everything on this page — including the one thing that surprises people most, which is that a criterion is free in time and can still be expensive in coverage.
A grid of matt cubes crossed by a thin green beam lighting only one row

A filter is a sieve laid across the whole flow: everything passes through it, only one row comes out lit.

Where the form is. Vestiaire Collective screen → “Filters” tab → “Create the filter”. The guided tour of that screen is here; this page is the trade manual behind it.

The one rule that governs the whole form

Three sentences, and the rest follows from them.
  1. An empty field is not a constraint. It is freedom you are giving the radar, never an oversight.
  2. A field the listing does not carry never satisfies a constraint you set. It refuses the listing — it never lets it through by default. We would rather lose a detection than manufacture a false one.
  3. Between two different criteria it is an AND; inside a single criterion it is an OR. Three brands widen; a brand plus a colour narrows.
That third rule is the one people misread. Ticking Gucci, Prada and Chloé asks for any of the three. Ticking Gucci and the colour black and Italy asks for all three at once, and each one you add can only shrink what comes back. With nothing ticked at all, the form warns you before you save: “With no constraint, this filter will keep every listing.” It is a legitimate choice — see what a filter that is too wide gives you — but it should be a decision.

The seventeen things a filter can carry

The form edits seventeen fields, and this is the complete list. Nothing else exists — anything not in this table is either a behaviour switch (further down) or something the product does not offer yet.
Fields 16 and 17 are not criteria of their own: they govern both word lists at once. The form states it rather than hiding it — “Both settings apply to both lists, and words are searched in the title as well as in the description.”

Every criterion is free in time — some are paid for in coverage

On Vinted, a criterion can cost you milliseconds, and the form prices each one. Here, nothing costs time: your criteria never travel to the marketplace on the detection path, so no criterion can slow a detection down. What some of them cost instead is coverage — and that price is not written on any screen, so this is where we write it. The radar reads Vestiaire two ways at once, and the two do not carry the same information about a listing.

The item read — the fresh one

One listing at a time. It is the only read that gives freshness: on a measured sample it saw 77 out of 77 listings, and it serves the listing at the very moment it is validated — the shortest delay measured was around 0 seconds.

The batch read — the complete one

Four hundred listings in one go. It saw 54 out of 77, and the 23 it missed were precisely the freshest — created 149 to 229 seconds earlier. It lags 3 to 5 minutes behind, and exists for completeness, never for speed.
Here is what each read can decide about a listing. This table is the most useful thing on this page.
What that means for you, in three lines.
  • Constraining the seller’s country rules out the item read — so nothing will be kept during a listing’s first few minutes.
  • Constraining a size rules out the item read as well, and rules out bags, jewellery and watches, whose listings carry no size at all.
  • Constraining the condition, the material or the third category level rules out the batch read — you keep all the freshness, you lose the catch-up.
The “learned” line deserves a word, because it costs nothing and it works. Every listing read one at a time teaches us how its address maps onto the first two category levels, and the batch read then reuses that lesson for free — verified 6 times out of 6. When two addresses would teach contradictory lessons, we arbitrate nothing: that address stays ambiguous for good. It is the same principle as everywhere else here — losing a detection is cheaper than inventing one.
The radar counts, sweep after sweep, how many listings it had to refuse for a dimension the read did not carry. That count is kept and written down — it is simply not shown on your screen yet. It is listed among the limits at the bottom of this page.

Criterion by criterion

Field left empty: nothing at all is offered. The list says “Start typing a brand name.”, and the line under the field says “Type two letters to search the whole brand catalogue.”That silence is deliberate. The marketplace only ever hands back one hundred brands for your current scope, out of 17,530 — a list that would look complete while showing 0.6 % of the catalogue, and would make a brand that is merely absent from those hundred look like a brand that does not exist.Field filled in, from two characters: the search covers the whole directory. Brands the marketplace already counts for your scope come first, each with its count; the rest of the catalogue follows, without one. A missing count is not a count of zero — the directory does not count, and we do not invent the number. When the search finds more than it can show, it says so: {n} shown out of {total} found — narrow the search.A brand is the cheapest criterion in the form: decidable by both reads, and it narrows hard.
The universe (Women, Men, Children), then three levels of tree. The screen shows the full path, “Women › Clothing › Dresses”.They combine with AND, not with OR. Ticking “Women” and then “Shoes (Men)” does not give you both: it gives the intersection, which is nothing at all. We do not forbid it — the preview counter says so immediately, in figures, and that is a more honest answer than a rule we would have guessed on your behalf.The third level is one of three criteria the batch read cannot decide — the condition and the material are the other two. On 1,882 out of 1,882 listings read that way, the address stops one level above it, and nothing in the line lets us derive it. Constraining any of the three means asking to be told by the item read alone.The number of values offered moves with what you have already ticked: 14 at the first level, 99 to 112 at the second, up to 229 at the third with nothing else set.
Pick a category first. With no category ticked, this field offers nothing at all and says only that: “Pick a category first: the sizes offered depend on it.” The marketplace does hand back eighteen scales when nothing is ticked — eighteen intervals of bare numbers, several of them starting on the same value, which is precisely why we no longer put them in front of you.Once a category is set, sizes are offered scale by scale, and which scales appear depends on that category: under “Women › Clothing › Dresses”, exactly one comes back. The consequence is then written above the control, before you pick any size: “Constraining size rules out bags, jewellery and watches: the field is missing from 43 % of listings.”A saved filter that already carries sizes always shows them, whatever its category — a constraint you cannot see is a constraint you cannot remove.Scales have no name. They are shown by their two extremes, Scale {de} – {a}. The marketplace gives them no name, and inventing one (“Shoes EU”) would be writing down a piece of data we do not have.And a size is kept as the marketplace’s own value, never as its label — because one and the same size stands for “12 UK / 40 FR / 44 IT / 8 US / L International” all at once. A filter written on the label lost 40 % of the matching items in the most favourable case measured: 49 out of 81.
24 values, 22 of them with a swatch in the product.A “Blue” filter catches an item described as “Navy”, and a “Beige” filter catches a “Camel”. That is not a tolerance we added: the marketplace indexes it that way, and we read what it indexes.One value behaves differently, and it is worth knowing: “Other” weighs 17,362 listings, and it comes with no parent colour at all — on 60 listings out of 60 probed. Judging colour on the parents alone would have thrown away at least ten thousand listings, so we do not.
56 top-level values and 19 finer ones, and the two sets are independent of each other.The direction matters, and it was measured across 8 crossings with a control: ticking “Exotic leather” keeps an item described as “Python”; ticking “Python” does not keep an item described as “Exotic leather”. Tick the wider of the two when you hesitate.Material is decidable on the item read only: constraining it rules out the batch read.
5 values. Like the material, the condition is carried by the item read and not by the batch read, so constraining it costs you the catch-up and nothing else.The condition label stored on a detection is the one from the sweep, which runs in French for everybody. What travels beside it is the identifier, and that is what your screen translates — so an English interface shows you an English condition even though the sweep read it in French.
Three modes: “Any origin” (“The seller’s country is not constrained.”), “From Europe”, and “Chosen countries” (“Only listings sold from these countries will be kept.”). The marketplace offers 71 countries.“From Europe” is exactly 29 countries, and it is the marketplace’s own definition, not ours: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Liechtenstein, Lithuania, Luxembourg, Malta, Monaco, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden and the Isle of Man.Denmark and the Netherlands are not in it — and the Netherlands alone weighs 8,154 listings. Iceland, Liechtenstein, Monaco and the Isle of Man are. Three independent readings of the marketplace gave the same list, so we serve it as it stands rather than quietly correcting it into something you could not check.One honesty point: the form calls this criterion “Seller’s country”, and what is actually compared is the country the listing is in. On six items read through both doors, the two disagreed twice. Where you need the seller specifically, read the seller’s country on the detection line.
The marketplace publishes no list of models, and the field is missing from 79 % of listings. So there is nothing to pick from, and the form says exactly that: “The marketplace publishes no list of models: these are kept as they are, and can be removed.”A filter that already carries models keeps them, shows how many — {n} models carried by this filter — and lets you take them off. That is the whole of it.

Price: which number is being compared

Two fields, “Minimum price (€)” and “Maximum price (€)”, both in euros, both empty to begin with. Empty means no bound at all: with the ceiling left empty, no price is ever too high for the filter. What matters is which price is compared, and it is not the one you will pay.
A price bound therefore does not compare what you will actually pay. On the item above, a filter set to “under €270” keeps it, and postage is still to come on top — €14.90 to a home address, €9.90 to a pick-up point. That is a deliberate choice: the compared price is the one that exists at the moment of detection, while postage only exists once a delivery has been chosen. What an item really costs, every fee added, is set out in Buying on Vestiaire.
There is one Vestiaire to read, and the country is a criterion applied to the listing — not a shop to choose. Nothing here resembles the Vinted rule where the origin you pick decides which shop is watched and a mismatched filter goes structurally blind. The interface language is a setting, and it changes nothing about what the radar sees: the sweep is shared by every client and runs once, in French, and your screen translates what it shows you.

Keywords: the criterion nobody else applies for us

Matt plates arranged as a funnel, the green light at its neck

Every word narrows the neck a little further — and one single ruled-out word closes it.

Keywords are the one criterion that is never sent to the marketplace, not even for the preview counter. We check them ourselves, on every listing, in the title and the description.
Keywords are not counted in the preview, and the preview says so itself: “Keywords are not counted here: they are checked at detection time, against the title and the description.”The reason is worth a sentence. The marketplace’s own search is a relevance search, not a text match; mixing its answer with our rule would give you a believable number that belongs to neither. So the counter tells you what the rest of your criteria are worth, and your words are checked later, on each listing.
One inclusion word, several exclusions. That is the shape that works, here as on Vinted. Since we read the description too, an inclusion word is far wider than it looks — and an exclusion word far more useful.

The preview counter, and what it is really telling you

At the top of the form sits “On Vestiaire right now”, followed by the number of listings that would match at this instant. It re-reads on its own while you tick.
The marketplace applies your criteria in exactly one place: this counter. Detections are decided entirely at our end, on a different path.The consequence is worth knowing, because it makes the two easy to believe in agreement when they are not: a fault in the counter shows on the counter and not on your detections — and the other way round. The counter is an excellent way to catch a contradiction before saving. It is not proof that the filter will detect.
The number next to a value says what you would GAIN, not what you have. Each count is computed by relaxing that one value’s own constraint and applying all the others. With Gucci ticked, type “max” in the brand field: Max Mara comes back with a count, and that count is what you would get if you ticked Max Mara as well.

When two criteria contradict each other

That last row is not a gap we are hiding — it is the one contradiction case nobody has measured, and we would rather write it here than let you conclude something false from a zero.

The four switches that are not criteria

Beyond the seventeen fields, a filter carries four decisions. They change what the filter does, not what it keeps.
An active filter is what arms the radar. Creating an active filter arms the watch for your workspace, without you clicking anything. And the reverse: disarming by hand is immediate, but it survives neither a restart nor the creation of a filter, as long as one active filter remains. To stop detection, untick “Active filter” — that is the gesture that holds.One exception to know: re-ticking “Active filter” on a filter that already existed does not arm the radar there and then. If the header still reads “Radar disarmed”, click “Arm the radar”.
Autobuy never spends. It reserves an account, empties and fills that account’s basket, sets the delivery and takes the purchase up to the payment — and hands back. Nothing that runs without you can reach the step that actually charges a card: it is not a setting we could get wrong, it is a road that does not exist. What autobuy costs you is your basket, not your money: see Buying on Vestiaire.
The rest of a filter’s life is on the card in the Filters tab: its hits counter, last {quand} or “never”, Rename (an empty name is refused), and Delete, which asks first — “Delete “{nom}”? The radar will stop looking for these listings; detections already found stay available.”
The hit counter is the only thing that tells you whether a filter works. No description of criteria will ever say that they are too strict together. One hour after creating a filter: 0 hits · last never means widen it — drop the size first, then the colour, then one category level.

What a filter that is too wide gives you

Listings are born on Vestiaire at about 24 per minute during the day — measured over 7.8 continuous hours, between 13.5 and 46 per minute — and 7.2 per minute at night, between 22:30 and 00:30 UTC. That is a ratio of 3.3, and, in cadence, about two hundred times slower than Vinted. So a filter that keeps most of that flow does not break anything. It simply gives you more than you can read.
At ~24 births a minute, a filter keeping most of them produces on the order of 35,000 detection lines a day. That figure is a calculation from the measured birth rate, not a measurement of a real filter — we have never measured what a typical filter produces per day in production, and we will not publish a number we do not have.
Three things happen to a wide filter, and none of them is an error message:
  • The list caps its display. Past a certain number of visible lines a banner appears: {n} detections kept — the {plafond} most recent are shown. Nothing is lost; drawing hundreds of thumbnails every few seconds would make the screen unusable.
  • “Alert me” becomes unbearable long before the list does. Leave it off until the hit counter tells you the filter is calibrated.
  • Detections are wiped after 48 hours, on their own, with no warning beforehand. Whatever you want to keep, favourite it or open it the same day.
The first sweep after arming notifies nothing. It is an inventory: without that rule, arming would pour the couple of thousand listings already online onto you as if they had just been born. And a listing already filed is never filed again — the radar re-sees it on every sweep for as long as it stays in the window, and says nothing.

A Vinted filter, a Vestiaire filter

What transposes, and what simply does not exist here. Read it as pairs.

Our advice

1 · Start with a brand, or one word

Both are decidable by both reads, both narrow hard, and neither costs you any coverage. Everything else is refinement.

2 · Read the counter before you save

“On Vestiaire right now” is the only thing that catches a contradiction before it costs you a day of silence. At zero, loosen — most often it is two category levels.

3 · Never constrain a size on bags, jewellery or watches

The field is missing from 43 % of listings, and an absent field never satisfies a constraint. A “Gucci bags” filter with a size ticked finds next to nothing, and says nothing.

4 · Think twice before constraining the country

It rules out the fresh read: your listings can then only arrive through the batch read, which lags 3 to 5 minutes. On this marketplace, that is most of your advantage.

5 · Tick the wider material

“Exotic leather” keeps “Python”; “Python” does not keep “Exotic leather”. When you hesitate between a family and its member, take the family.

6 · One inclusion word, several exclusions

Every inclusion word must be present, and we read the description too. Three words is how a silent filter is written.

7 · Come back an hour later and read the hits

0 hits · last never after an hour means the criteria are too strict together. No amount of re-reading the form will tell you that.

8 · Leave “Alert me” off until the filter is calibrated

An alert per kept listing, on a flow of ~24 births a minute. Turn it on once you know the hit counter, never before.

9 · To stop a filter, untick “Active filter”

It is the gesture that holds — the one that survives a restart. Deleting keeps the detections but their line will read “deleted filter”.

10 · Take out what you want to keep the same day

Detections are wiped after 48 hours, silently. Favourite the find, or open it in a tab.

What surprises people

Not a contradiction: a refusal for want of information. If the read that met the listing does not carry the dimension you constrained, the listing is refused rather than let through. The alternative — letting it through — would fabricate detections that do not match.
This is the difference that surprises people most in practice. A word buried in a long description brings in a listing you were not expecting — and, symmetrically, an exclusion word buried in a description throws out a listing you did want.
Both defaults lean the same way: towards the widest reading, because the widest one cannot make you miss anything. And both apply to the list that includes as well as to the list that rules out.
Ticking “Women” and “Shoes (Men)” is an intersection, not a union. The preview counter answers immediately, in figures — which is why the sentence it shows at zero names this case first.
And it includes Iceland, Liechtenstein, Monaco and the Isle of Man. It is the marketplace’s own grouping, verified three independent ways. We serve it as it stands rather than correct it into something you could not check against the site.
Past that, the marketplace stops counting. We show the + rather than a bare figure, because writing “10,000” would pass a limit off as a measurement.
It is computed by relaxing that value’s own constraint while applying all the others. So the list keeps showing you brands you have not ticked, with the number of listings ticking them would add.
Because the marketplace files navy under blue itself. We did not add a tolerance — we read its index. The same goes for camel, which it files under both beige and brown.
The fallback name is saved as though you had typed it: created in English, it stays “Unnamed filter” even after you switch the interface to French — and one created in French stays “Filtre sans nom” in an English interface. Renaming it is the only way out.
You never have to click “Arm the radar” after creating your first filter. The reverse is the surprising half: as long as one active filter remains, disarming by hand does not survive a restart.

What filters do not do yet

We would rather write it here than let you find out halfway through.

Models cannot be picked

The marketplace publishes no list of models, and the field is missing from 79 % of listings. A filter that already carries models shows them and lets you remove them — that is all.

0.7 % of listings are out of reach

They are old items put back on sale, far below the window we sweep. No filter can reach them, whatever it is written with. The rest — 99.3 % — is covered.

Marketplace dimensions with no field here

Direct shipping and the countries it covers, the seller’s badge, discounts, the marketplace’s own editorial picks, stock, region, the movement of a watch, whether the shop is official, whether an item is already sold, and local delivery countries. That is not a closed list — the marketplace publishes more dimensions than the form carries, and we would rather name the ones we have seen than a count we cannot keep true. None of them can be filtered on here yet.

The refusal counters are not on screen

The radar counts how many listings each dimension made undecidable, and publishes it. Your screen does not show it yet — so “this criterion is costing me a read” is something this page tells you, and the product does not.

No groups in the Vestiaire form

A filter is filed from the list, not from the form, and drag-and-drop remains a Vinted feature.

Export and import stay Vinted

The file format saves and restores Vinted filters only. Vestiaire filters are not carried by it.

Bulk edit carries neither name nor autobuy

Deliberately: renaming twenty filters at once would make them indistinguishable, and arming twenty at once would fill twenty baskets.

Detections last 48 hours

The purge runs on its own and warns you of nothing. It is the deadline that governs how you work on this screen.
Two things we have not measured, and will not guess at: how many listings a typical filter really detects per day in production, and how the marketplace combines a top-level material with a finer one. Where a figure does not exist, this guide leaves the space empty rather than fill it.

Where to go next

The marketplace, and how it differs

Why there is no birth to watch for here, and what one shared sweep means.

The Vestiaire screen

Every control, every badge, every refusal — the guided tour.

Your Vestiaire accounts

What a filter needs before it can offer or prepare anything.

Making an offer

The floor, where it comes from, and why it never goes back down.

Buying on Vestiaire

What an item really costs once every fee is added.

Writing a good filter

The same trade, on Vinted — where criteria are paid for in milliseconds.