Back to Explore
GuideVibeFix editorialUpdated Nov 20, 2026

No Results Is Not a Dead End: Zero-Result Search in Practice for Vibe Projects

Of 1,000 searches, 132 return zero results; 74 click a suggestion, 41 return to a results list, and 23 finish what they came to do. A zero-result page is not a dead end; it is the moment a user hands you their need directly. The product, people, and numbers in this walk-through are fictional.

A real photo: a woman sitting on a couch, frowning at the phone in her hands, frustrated at finding nothing.

No Results Is Not a Dead End: The Zero-Result Page Is Your Second Chance

Atang wants an out-of-print art book, so she types its title into the search box of Old Book Dock, a secondhand book app, and hits enter. The screen shows a single line: no results found. She stares for two seconds, goes back to the home page, flips through two pages of categories, finds nothing, and closes the app. The app did nothing wrong, and the book really is not in stock, but it just lost a user it could have kept. The product, people, and numbers in this article are fictional examples, used to explain zero-result search itself.

First, a boundary, so this does not collide with what the site already published. The site guide on site search covers the search engine itself: whether to pick Postgres full-text search or Typesense, how to build the index, how to handle tokenization. That guide spends one paragraph on the empty state of a zero-result page. This guide does not cover engine selection or indexing. It covers only what the screen in front of the user should do after a search returns zero results: what to say, what suggestions to offer, and how to lead people back to results. If you have not added search yet, read that guide first. If you already have search but your zero-result page is still one dry line of no results, this guide is for you.

A zero-result page is not a search failure. The moment a user types a query, their intent is the clearest anywhere on your site: they want something specific, and they are willing to tell you what it is in their own words. A bare no results throws that intent away. A good zero-result page does three things: admit nothing was found, help the user ask differently, and offer a way forward. None of the three requires changing your search engine, and all three can ship in half a day to a weekend.

Sort It First: Four Kinds of Zero Results, Four Different Causes

Before changing anything, work out which kind of zero result you are facing. The symptom is always zero items, but the cause differs, and so does the fix. Treating all four the same way guarantees the copy changes and the problem stays.

The first kind is genuinely absent. What the user wants is simply not in your catalog. Old Book Dock stocks thirty thousand books, and the out-of-print art book Atang wants is not among them. This zero result is not a bug; it is a fact. The fix is not in search but in the exit: say it is not here, and give the user a reason to stay, such as subscribing to the keyword to be notified when a copy arrives, or showing a few similar books.

The second kind is a typo. In Chinese search, the most common cause of failure is not missing content but wrong input: the input method suggests the wrong word, the pinyin is mistyped, or a homophone is picked by mistake. The user wants lamian (noodles) and types lahua; wants Minguo (the Republic era) and types Mingguo. The engine searches literally and returns zero. The fix is spell correction and rewrite suggestions, not an apology.

The third kind is over-filtering. The user did find results, then ticked three filters: only Beijing, only under 50 yuan, only 90% new or better. Three filters stacked turn 40 results into zero. The user thinks the item is out of stock, when in fact they closed the door themselves. The fix is to show which filter emptied the list, and let them loosen the filters one by one.

The fourth kind is invisible. The content exists, but the user has no permission to see it, or it is a draft, delisted, or visible only to its owner. After visibility rules filter the results, the count is zero. This is the most dangerous kind: the page says nothing was found, the user concludes the product is empty, and the truth is that permission rules hid the results. The fix is to separate absent from not allowed: for content the user cannot see, say log in to view or contact an administrator rather than saying it does not exist.

How do you know which kind dominates your product? Read your search logs. Record each search: the query, whether it returned results, and which filters were applied. After a week, count the four kinds, and the proportions will show themselves. Most small products find that the second and third kinds together are more than half. In other words, more than half of zero results are not missing content; the search experience itself pushes people away.

Diagnose First: Measure the Zero-Result Rate, Then Read the Queries

Whether to build a zero-result page, and how far to take it, should not be decided by feel. Measure one number first: the zero-result rate. The formula is simple: searches that returned zero results divided by total searches. The site search guide gives a rule of thumb: a zero-result rate above 15% means your index coverage or tokenization has a problem, so fix search itself before polishing the zero-result page. That 15% is a warning line, not a passing grade.

Take a concrete example. Old Book Dock has 1,000 searches in a week, of which 132 return zero results, a zero-result rate of 13.2%. Below the warning line, but of those 132, 58 users glance and leave without clicking anything. Those 58 are exactly what the zero-result page should win back. The funnel looks like this: of 1,000 searches, 132 return zero results; 74 of those click a suggestion or loosen a filter; 41 return to a list that has results; 23 finish what they came to do, such as placing an order or subscribing. 23 divided by 132 is about 17%: the recovery rate of the zero-result page. Record these four numbers before building the page and again after, and the numbers will tell you whether it worked.

Three fields in the log are enough: the query, the result count, and the filters. Do not start by building a complex analytics system. Add one line of logging in the search API and read it once a week, and you will get a free list of user needs: words people search often but that return zero results are things users want and you do not have. When a word is searched dozens of times and always returns zero, it is not a typo; your content genuinely lacks that area. That list is more honest than any user interview, because users are voting with their actions.

One more diagnostic angle is easy to miss: look at which words the zero results cluster on. If they cluster on three to five words, the problem is usually content or synonyms: users search ershou (secondhand) while your catalog only says xianzhi (idle items); users search huiben (picture books) while your catalog only labels tuhuashu. If zero results spread evenly across hundreds of words, the problem is usually tokenization or typo tolerance. The fixes are completely different: the first calls for a synonym table, the second for work on the query layer.

Layer One: Admit It, Repeat the Query, Do Not Blame the User

The first line of a zero-result page decides whether the user stays or leaves. The worst writing blames the user: your keywords are wrong, please try again. The user did not type it wrong; your product failed to find it. Good copy does three things: admit, repeat, and offer a next step.

Admitting means saying plainly that nothing was found, but not only that. Repeating means writing the user's query back to them exactly as typed, so they can confirm what they searched. Many zero-result pages do not even repeat the query, leaving users to wonder whether they mistyped or whether the search box ever received their input. No lamian found is far more useful than no relevant results, because the first proves the system heard them, while the second reads like a brush-off.

Here is a copy template you can use directly. Line one says no lamian found. Line two offers a guess at the cause, such as the book is not in stock right now, or try a different word. Line three offers an action: look at these similar books below, or subscribe to lamian and we will notify you when a new copy arrives. The whole zero-result page should be no more than three short passages. The user came to find something, not to read an essay.

Two lines should never appear. The first is please check your spelling. It only helps when the user really did misspell something, and you have no idea whether they did. The second is try different keywords, with no clickable words underneath. An empty suggestion is no suggestion: the user will not invent a new word just because you told them to. Give them specific words they can click.

Layer Two: Correction and Rewriting Help the User Ask Differently

Right after admitting the miss, help the user ask differently. This is the most valuable layer of the page, because it turns a failed search into a successful one.

The most common move is did you mean. The user searches lahua, and the system suggests lamian; the user searches Mingguo, and it suggests Minguo. There are two ways to build it: typo tolerance built into the search engine, such as Typesense, which works out of the box; or, in a Postgres setup, the pg_trgm extension for fuzzy matching as a fallback that runs only when the first search returns zero. The site search guide already describes this two-layer funnel: run full-text search first and return results directly when there are any; on zero results, run a trigram fuzzy query and show a did you mean prompt in the UI. Two layers keep both precision and recall. This guide will not repeat the implementation; one point matters here: the correction must be a clickable link, not a line of text. One click to search the new word saves the user from retyping, and every retype is another chance to lose them.

The second move is dropping one word. The user searched three words, jueban jiu huace (out-of-print old art book), and the system can suggest: try dropping jueban, or try dropping jiu. Many zero results come from users stacking too many qualifiers; removing one word brings results back. This works especially well for multi-word searches, and it is simple to build: split the query, drop one word at a time, run each variant, and suggest whichever variant returns results.

The third move is loosening a filter instead of changing the query. The user searched lamian and had 40 results, then ticked three filters and got zero; the system should tell them directly: dropping the 90% new or better filter brings back 18 results. That sentence is more useful than any generic advice, because it talks about the user's own actions, not generalities.

The order of the three moves matters too: correct first, then drop a word, then loosen filters. Correction stays closest to what the user meant and changes the least; dropping a word changes more; loosening a filter changes the most and goes last. Each step changes only a little, and each click brings the user closer to results instead of throwing them back to the start.

Layer Three: Offer an Exit, Do Not Leave Users Circling

If correction and rewriting still bring back nothing, the page has one last job: offer a way forward instead of leaving users to figure it out themselves.

The first exit is popular searches. Show five to eight of the most searched terms on the site right now, ideally related to what the user searched. A user who searched for an out-of-print art book can be shown jiuhuace (old art books), banhuaji (print collections), and sheyingji (photography books). Popular searches are not decoration; they are a next step that requires no thinking. Make sure they are genuinely popular, not terms the operations team wants to promote. Users can tell at a glance which words people actually search and which are ads.

The second exit is related recommendations. Take the user's query and recommend similar content. The out-of-print art book Atang wants is not here, but the site has other books by the same author, or photography books from the same period. Say why you recommend them: because you searched for out-of-print art books, here are these. A bare recommended for you tells the user nothing about the connection to their search, so they will not click.

The third exit is subscribing to the keyword. It is one of the highest-converting exits on this page. The user's intent here is extremely clear: they want something you do not have. Rather than a dry apology, offer a subscribe to this keyword button, and notify them when new content arrives. For content products, the subscription relationships this one small feature accumulates are worth more over time than search itself. Many newsletter products cold-start exactly this way: let users search first, let them subscribe when nothing is found, and notify them once the content exists. The site search guide already made this argument; here is the concrete version: put the subscribe button below the correction suggestions, not at the bottom of the page; after subscribing, confirm and say when the user will be notified; when the notification really arrives, the user will believe the button is not decoration.

The fourth exit is asking a human. For a small product, a help entry is cheaper than a recommendation algorithm: did not find the book you want? Tell us what you are looking for. Every message a user leaves tells you what to stock next. Someone has to actually read this inbox, at least once a week, and ideally reply with found it or not yet. A help entry nobody reads is worse than no entry at all.

You do not need all four exits. Products with little content can stop at popular searches plus keyword subscription; products with more content can add related recommendations. The help entry fits products still cold-starting, because at that stage every user message in their own words is valuable.

Filtered Zero Results Hurt Most: Loosen the Filters One Notch at a Time

Of the four kinds, filter-caused zero results are the easiest to overlook and the most painful. The user had seen results, then filtered them away step by step, while the page said nothing but zero.

The right approach is to show the filters on the zero-result page so users can see what they selected. Write one line: no results under the current filters, followed by the selected filters. Put a cross after each filter; one click removes that single filter, and the result count updates live. When the user removes 90% new or better and the count jumps from zero to 18, they understand immediately which filter caused it.

Going one step further, tell the user how many results each filter is blocking. Dropping only Beijing brings back 12 results; dropping under 50 yuan brings back 9. The numbers do not need to be exact; the order of magnitude is enough. Users can see at a glance which filter to loosen instead of clearing all three and starting over.

One more detail: when filters produce zero, do not hide the whole result list. If the user has seen results before, keep a way back to the previous filter state. A clear all filters button is good, but it should be a backup, not the only option. Offer loosening one filter at a time first, then clearing all. Do not reverse the order.

Finally, link filters to the query. If the query itself is rare, adding filters on top should trigger a warning when the user ticks the second filter: adding one more filter may leave no results. A warning before the fact is far cheaper than an explanation after it.

Three Tiers Compared: Half a Day, One Weekend, Two Weeks

A zero-result page does not need to be perfect in one pass. Build it in three tiers, each with a clear deliverable.

TierTimeBuildSkip
LightHalf a dayAdmit-and-repeat copy, popular searches, a clear-filters buttonCorrection, subscription, recommendation algorithms
StandardOne weekendEverything in Light, plus did-you-mean, loosening filters one by one, keyword subscriptionPersonalized recommendations, a search-term analytics dashboard
FullTwo weeksEverything in Standard, plus related recommendations, search-term log analysis, a weekly review of zero-result termsSwitching search engines

The Light tier fits products that just launched search: rewrite the copy, put popular searches on the page, make filters clearable. Half a day of work, and the page is no longer a dry no results. The Standard tier fits products that already have users: correction prompts and keyword subscription are the two most valuable items here; the first turns failed searches back into successful ones, the second turns a miss into a subscription relationship. The Full tier fits products where search is already a core entry point: at that stage it is worth setting aside time for search-term logs and reviewing the zero-result term list once a week, because the words people search often but that return nothing are exactly the content you should add next.

Which tier to pick depends on one number: the zero-result rate. Below 8%, the Light tier is enough; the problem is not the zero-result page. Between 8% and 15%, build the Standard tier. Above 15%, fix the index and tokenization first, following the site search guide, and come back to the zero-result page afterwards. Getting the order wrong means polishing a page for a broken search.

Launch Checklist and the Funnel Numbers

Before launch, go through this checklist item by item. Does the copy repeat the user's query? Can the correction prompt be clicked? Can filters be loosened one by one? Are the popular searches genuinely popular? Does the subscribe button give confirmation feedback? Is anyone reading the help inbox? Walk through every item with a real zero-result query, not just a design mockup.

After launch, watch four numbers every week, the same funnel as before: of 1,000 searches, 132 return zero results; 74 click a suggestion or loosen a filter; 41 return to a list with results; 23 finish what they came to do. The step to watch most closely is the second one: only 74 of 132 zero-result searches click a suggestion, and the other 58 leave directly. If that 58 does not come down, the copy or the ordering of suggestions on the first screen is missing the mark; fix the copy and the ordering before adding new features. Of the 41 who reach a results list, 18 do not finish, which means the recommended results are not accurate enough, or the results page itself has a problem; that is when to look at the results page, not keep changing the zero-result page.

Funnel chart: of 1,000 searches, 132 return zero results, 74 click a suggestion or relax a filter, 41 return to a results list, 23 complete the intended action.

Two more numbers are worth watching alongside. The first is subscription conversion: of the people who subscribe to a keyword on the zero-result page, how many actually come back when notified. The second is repeated zero results: the same user searching the same word three times in one week and getting zero every time. When that number climbs, the subscription or help entry is not catching them, and they can only keep coming back to try their luck. Watch these two numbers together with the funnel, and the zero-result page is genuinely running.

Anti-Patterns: The Five Most Common Ways to Get This Page Wrong

The first is saying only no results, with no next step. Users are not here to grade your search system; they are here to find something. One line of no results tells them they may leave.

The second is copying the 404 page copy. A 404 says the address is wrong; a zero result says the item was not found. The next steps are completely different. A 404 sends users home; a zero result should help them ask differently.

The third is filling popular searches with terms the operations team wants to push. A user searching for an out-of-print art book who is shown a new-book presale can tell immediately that it is an ad. Once popular searches lose credibility, users will never click them again.

The fourth is making the correction prompt plain text that cannot be clicked. The user sees did you mean lamian and still has to delete lahua and retype lamian themselves. One missing click is one more chance to lose them.

The fifth is treating the zero-result page as ad space, stuffing it with banners and recommendation slots. The user's intent here is clear: they want something you do not have. Pushing ads at that moment is like handing out business cards to someone in a hurry. Keyword subscription can stay, because it answers what the user actually asked for; promotions unrelated to the search should go.

Whether the page works is not a matter of asking users; the numbers will tell you. Watch the zero-result rate, the suggestion click rate, and the recovery rate once a week. Four weeks of movement in the right direction means the page is doing its job. No results is not a dead end; it is the moment a user hands you their need directly. Catch it.

Views 12Comments 0

Comments (0)

ME
0/1000
Loading comments...
Browse projectsPublish your project

Related articles

Real photo: a laptop, tablet, and phone on a wooden table showing the same green-leaf website, the phone view with navigation in one column, a potted plant nearby
Guide
Squashed on phones? Responsive layout and media queries, explained plainly

One page should stand up on a phone, a tablet, and a computer. This guide follows MDN's Responsive web design, Using media queries, and the viewport meta reference: fluid HTML, the viewport tag, fluid grids and max-width, images that never burst their container, media query syntax, breakpoints set by content, mobile first, responsive type, testing capability instead of width, and a pre-launch checklist.

Mobile AppsDesign ExperienceFrontend Engineering
Real photo: a bank of light gray metal filing cabinets with one drawer pulled open, revealing rows of hanging file folders inside
Guide
A real database inside the browser: IndexedDB local storage, explained plainly

localStorage only stores strings and has no indexes. IndexedDB is a real database in the browser: structured data, far larger capacity, indexes, and transactions. Following MDN's Using IndexedDB, IDBDatabase, and IDBObjectStore: the five-step routine, opening, upgrades, keys, indexes, transactions, add versus put, get, cursors, ranges, and multi-tab upgrades.

Tool TipsData WorkflowsFrontend Engineering
Real photo: hands typing on a MacBook Pro showing JavaScript code that changes HTML content, with a white mouse, papers, and binders on a white desk
Guide
Functions without the confusion: JavaScript functions, parameters, and arrow functions, explained plainly

You already call functions every day; this guide names what you have been doing. Following MDN's Functions, Arrow function expressions, and Rest parameters: declarations versus expressions, parameters and arguments, default values, rest parameters, spread syntax, return values, arrow function shorthand, the this trap, callbacks, and scope.

Tool TipsFrontend EngineeringAI Coding