Writing a good filter

A filter is your standing order: “tell me the moment a listing like this one is born on Vinted”. The radar watches roughly 150 listings being born every second and shows you only yours. Knowing how to write a filter means knowing how to describe what you are after without describing it too much, and paying as little time as possible to describe it. This page is not the guided tour of the screen — that one is here. This is the trade manual: what to fill in, in which order, what it costs, and how to know it works. The form says it on its very first line: Every criterion adds up. What is left blank is not narrowed. A field left empty is never an oversight — it is freedom you are giving the radar.

---

​

1. A filter must have an anchor

This is the rule that costs clients the most, and the only one the screen does not remind you of. The radar recognises a newborn listing by its title and its brand. If your filter gives no grip on either, it has nothing to hold on to. Three perfectly plausible, perfectly silent filters:

---

​

2. What each criterion costs

Not all criteria are paid for at the same price. Some are settled with the listing itself — instant, free. Others require fetching information that does not arrive with the listing: they delay display. Each of the three regimes carries a pill on the filter row, and a sentence in the form: What it is worth, measured. On 26/07 in production: a category cost +396 ms, a precise origin +408 ms, and both together +399 ms shared — that is, the price of one. On 14/08, origin had dropped back to 362 ms while category climbed to 15.8 s: that is open work on our side, and the consequence for you is immediate — today a category costs a lot and a country costs almost nothing.

---

​

3. A listing’s journey, from birth to your feed

This diagram answers almost every question we are asked. Each diamond is a place where your settings decide. Four practical consequences, and they are worth money:

• MIN PRICE works before everything else. It drops listings at no cost, so it gives speed back to all your other filters. It is the only free lever in the product.
• The brand decides, never the title. A brand filter only accepts the brand declared by the seller; a title that says “Prada” on a listing not declared as Prada will not pass. Symmetrically, a filter without a brand does not look at the brand at all: its keywords are its only criterion.
• A card can gain a second filter after the fact, a few hundred milliseconds later. That is not a display glitch: the second filter was waiting for its check.
• A card marked “origin presumed” did not get its check and is excluded from automatic buying. That is deliberate: we would rather show it to you without buying it than hide it.

---

​

4. Where to start: the three-branch question

Before you open the form, answer a single question — what is my anchor? Everything else follows.

---

​

5. The order that does not waste your work

The form is filled in from top to bottom, but the efficient order is not that one. This order avoids the two most expensive mistakes: retyping a field that got overwritten, and estimating ten times in a row.

---

​

6. Five complete recipes

---

​

7. Knowing a filter works: the five-minute check

​

The five signals, and what each one does not prove

---

​

8. When a filter surfaces nothing

Six mechanisms, all silent. Go through them in this order.

---

​

9. When a filter surfaces too much

Noise can be seen, so it can be fixed — but it is not free: a filter that is too wide slows all the others down. Measured: at 4.77 kept listings per second against a capacity of 4/s, a card’s display goes from 0.55 s to 15.4 s. Nothing is lost, nothing fails: everything arrives too late. And no error counter moves when that threshold is crossed. Tighten in this order — from the freest to the most expensive:

---

​

10. Looking after a set of filters

One good filter is not enough: it is the set that decides your speed. The header controls exist for that. The weekly review, in four moves:

1. Sort by Detections and look at the bottom of the list: those are your silent candidates.
2. Open each of them and look for its anchor (section 1).
3. Sort by Autobuy first: re-read the cap and the origin of every armed filter.
4. Export before deleting anything — a deletion from ⋯ is immediate and without confirmation.

---

​

Our tips

​

1. Set a MIN PRICE on every filter, even a low one

It is the only free lever in the product, and the most powerful. Measured on 675 real listings, the share of the flow left to examine for each floor: ⚠️ The radar keeps the lowest MIN PRICE across all your active filters. A single filter with no floor cancels the benefit for all the others. If you remember one tip from this page, this is the one.

​

2. One intent per filter

A filter trying to cover “Prada, Miu Miu, and also leather jackets from anyone” will be bad on both counts. Two well-anchored filters cost less than one hybrid, and there is no limit on the number of filters.

​

3. Stay on Euro area until you actually need to tell countries apart

It is the default, and the only free origin shortcut: the seller’s currency arrives with the listing. Keeps listings whose seller prices in euros. Slightly wider than what can really be bought from France: it never misses a listing, it lets a few more through. It does not tell countries apart. The euro area covers 18 countries: “France only” is a different criterion, and it has to be checked.

​

4. Never write a category on its own

With no brand and no keyword, it drops into the slowest regime: a silent first pass, then batches every 4 seconds. One brand or one word is enough to get out, and the screen offers it to you itself.

​

5. A filter with no brand deserves one single keyword

The rarest, the most discriminating one. The others go into EXCLUDED WORDS. Writing three inclusion words with no brand is the surest way to end up with a silent filter.

​

6. Use the warnings in the keywords field

They are written to save you hours: “{mot}” is written twice — with no effect. · “{mot}” is already contained in “{porteur}”: it narrows nothing further. · “{mot}” is very short…. A word that narrows nothing is a word that makes you believe you have tightened.

​

7. Aim at the parent category, never at the leaf

Sub-categories are included ({n} categories — sub-categories are included.), and a category that is too deep forces us to load 2.39 MB instead of answering in 240 ms. You gain no precision, you lose speed.

​

8. Read the estimate as a ceiling, never as a rate

The estimate ignores your keywords and your origin. Estimating “Gucci, bags, keyword tote, France only” actually measures “Gucci, bags” — potentially ten times more. The real rate of a keyword filter is necessarily lower than the figure shown.

​

9. Estimate only once per filter

The Estimate button draws on the same capacity as the radar. Hammering it while you type slows down your own detection — and everyone else’s. Once per filter is enough.

​

10. Re-read Watching Vinted {pays}. before every save

It takes two seconds, and it is the only line that tells you which shop is really being watched. Six of a client’s filters were structurally blind for want of that read.

​

11. To watch another shop, paste its link — it is free

If you already run a search by hand on Vinted, its address contains everything: brands, categories, prices, sizes, and even the materials and patterns no picker offers. Paste, Analyse, then refine. Sort, page and session parameters are deliberately ignored: copying them has no effect.

​

12. Duplicate by region rather than listing countries

A “Euro area” filter is fast; a “6 precise countries” filter is checked. If you really want to separate countries, make copies with Duplicate by region: you will keep one readable filter per country, and you will see which one produces.

​

13. Name your filters for the list, not for yourself

The name is never translated and is used for sorting. PRADA · bags · <400 reads at a glance in a list of forty rows; “my filter 3” does not. There is neither search nor pagination in the list: at two hundred filters, you scroll.

​

14. File things into groups from day one

A group is a dashboard sub-page and the unit that decides which account pays. Deleting a group does not delete its filters — they simply go back to having no group.

​

15. Before a drop, review the armed filters

Sort by Autobuy first and check three things on each row: the regime pill (a checked filter buys later), the cap shown (no cap or ≤ 400 €), and the group’s paying account. An item can only be bought for 120 seconds after it is detected: whatever is not ready before the drop will not be ready during it.

---

​

What surprises people

• A category includes all its sub-categories — on Vinted as with us. Ticking a child on top of the parent narrows nothing: the screen answers Already covered — {chemin}.
• A card can gain a second filter after the fact. The first one was confirmed immediately, the second was waiting for its check.
• A card may never appear. If the check has not concluded within 15 seconds, the card is shown anyway, marked origin presumed, and excluded from automatic buying.
• The default keyword mode is the widest one. contains keeps “handbag” when you write “bag”. That is deliberate: the widest mode is the one that cannot make you miss a listing.
• In contains only mode, the brand does not count as a foreign word. A title “Prada bag” passes a contains only: bag filter if Prada is the declared brand.
• By market and By country cost exactly the same. Do not look for a third price: Country and region are one and the same criterion here…
• On a filter that already carries a category, adding a country is free — and the screen then shows the same figure on both sides, marked {delai} shared.
• Changing a filter’s price no longer wipes its materials and patterns. That was a real defect: a simple price edit used to change the filter’s detection regime, without a word.
• The AUTOBUY badge disappears when you disable the filter, but the arming stays. Re-enabling it puts it back in a buying state with no other move.
• Deleting a filter from ⋯ is immediate and without confirmation. Only bulk deletion asks for confirmation, and there is no undo.
• Colour is no longer a criterion (removed on 31/07). An old filter that carried one lost it on reload — checked before doing it: 0 filters out of 42 in production carried one.
• You can tick a great many brands. Measured on 15/07: about 1,031 brands fit in a single filter; around 1,062, Vinted refuses. A filter with several dozen brands is perfectly normal.

---

​

On a phone

What changes, and what it implies for writing a filter:

• The header buttons and the sorts wrap onto several lines.
• A filter row keeps the checkbox, the switch, the name (truncated), the marketplace pill, the regime pill, the Autobuy switch and the ⋯ menu. The criteria summary, the in total counter and the freshness line do not fit — so the weekly review is not done on a phone.
• A navigation bar appears at the bottom: Feed · Filters · Orders · Tracked · More.

---

​

The limits you should know about

We would rather tell you than let you find out on a drop day.

• Nothing flags a filter with no anchor. The diagnosis in section 8 is, for now, a manual move.
• No filter is ever refused for being too wide. The estimate informs, it does not authorise: a filter that would surface thousands of listings per minute will be created without a word.
• The estimate of a non-French filter is worth nothing: it is always run against the French shop.
• Materials and patterns have no picker. They can only be set by pasting a Vinted link that contains them; the form keeps them and offers to Remove them.
• ≈ {n} listings/min — estimated earlier. will probably never show up: the estimate is not remembered from one session to the next.
• Category is expensive right now (15.8 s measured on 14/08, against 362 ms for origin). That is open work on our side; in the meantime, prefer a brand or a keyword as your anchor.

---