Zendesk Searches With No Results Report: Find Content Gaps

Last updated: September 06, 2026

Zendesk Searches With No Results Report

Most self-service metrics tell you that self-service is underperforming. This one tells you what to write.

A search that returns zero results is a customer typing, in their own words, a question your knowledge base cannot answer. There is no interpretation needed and no proxy involved. Each one is a documented gap and a likely ticket — and Zendesk records the exact phrase.

This guide shows how to report on zero-result searches in Zendesk Explore, how to separate missing content from unfindable content, and how to turn the list into a writing queue. Keep it beside the support metrics dashboard, the Zendesk Help Center Views Report, and the Zendesk Search-to-Ticket Ratio Report.

What this report should answer

  • What are customers searching for that we have not documented?
  • What share of all searches return nothing?
  • Is the gap growing, and is it concentrated in one product area, brand, or locale?
  • Which zero-result searches end in a ticket?
  • Are we missing content, or do we have the content under a name nobody searches for?

For the definition and formula, see search no-result rate. This guide covers building and acting on the report.

Three different search failures

Zero results is only one of three ways help center search fails, and they need different fixes. Separating them is the whole point of this report.

Failure What the data shows What it means The fix
No results Search returned zero articles The content does not exist Write the article
No clicks Results returned, none clicked Content exists but titles do not match the query Retitle, relabel
Click then ticket Article read, ticket created anyway Content exists but does not resolve Rewrite the article

Teams that report only the first number under-invest in the other two. In most help centers, “results but no clicks” is a larger population than “no results at all” — and it is cheaper to fix, because the content already exists.

Build all three panels. Then work the cheapest fix first.

How to build the report in Zendesk Explore

1. Start with the prebuilt Search dashboard

Zendesk ships a Search dashboard inside the Guide dashboard, and it already covers most of what small teams need:

  • Total searches and Searches with no results
  • Avg click-through rate
  • Tickets created from the help center
  • Searches with no results (top 5) — the actual phrases
  • Search effectiveness — searches with and without results over time, with click-through overlaid
  • Searches by results and clicks — the three-way split above, as a pie

It filters by Time, Brand, Search channel, User role, and Locale. Start here. Only build custom reports when you need a longer list than the top 5 or a segment the dashboard does not expose.

One quirk worth knowing: the dashboard excludes search events with a blank search query, which keeps content-tag data from polluting the results.

2. Build the custom version for a longer list

The top-5 chart is too short to work from. For a writing queue you want the top 50.

In the Guide - Search dataset:

  • Metric: Searches with no results
  • Rows: Search query
  • Filter: Search timestamp on a rolling window
  • Sort: descending, then increase the row limit

The underlying formula, if you want to understand or adapt it:

IF ([Search result type] = "No results") THEN [Search ID] ELSE NULL ENDIF

Add % No result rate as a second metric for the trend view:

COUNT(Searches with no results) / COUNT(Searches)

Export the query list. That export is your content backlog.

3. Add the no-click panel

Zero results is not the whole gap. Add a second report using Searches with no clicks, which counts searches where results appeared but the customer selected nothing:

IF MAX(Clicks by query) = 0 THEN COUNT(Searches) ELSE 0 ENDIF

Rows on Search query again. These queries are usually the fastest wins available: the article exists, so you only need to fix the title or add labels for the words customers actually use.

4. Group the raw phrases into themes

A raw query list is noisy. It will contain misspellings, single words, and near-duplicates. Rather than reading 200 rows, cluster them.

Create a standard calculated attribute to bucket a theme:

IF (CONTAINS(LOWERCASE([Search query]), "refund")) THEN "Billing - refunds"
ELIF (CONTAINS(LOWERCASE([Search query]), "invoice")) THEN "Billing - invoices"
ELIF (CONTAINS(LOWERCASE([Search query]), "password")) THEN "Account - access"
ELSE "Unclassified"
ENDIF

Use LOWERCASE() around the query so casing does not split your buckets, and keep the Unclassified catch-all so you can see what your rules missed. Build these iteratively: run the raw list, spot the recurring words, add a branch. See Zendesk Explore Calculated Metrics for the formula mechanics.

5. Segment where the gap lives

Break no-result searches down by:

  • Locale — a very common blind spot; English content is complete, translations are not
  • Brand — multi-brand help centers often share content that only fits one brand
  • User role — end users and signed-in customers search differently, and agent searches can skew the blend
  • Search channel — web widget, help center, and mobile SDK searches behave differently

Locale is worth checking first. A 6% no-result rate overall can hide a 40% rate in one language, and that segment’s customers have no self-service option at all.

6. Connect searches to tickets

The prebuilt dashboard’s Tickets created figure lets you drill into the search terms that preceded a ticket. That is your prioritisation signal: a gap that generates tickets is worth more than a gap with high search volume and no downstream cost.

Pair this with ticket deflection and the Zendesk Ticket Deflection Report to estimate what closing a gap is actually worth.

If you use AI quick answers

If generative search or quick answers are enabled, check the Quick answers dashboard too. It surfaces searches that failed to generate an answer, which is a different signal from zero article results: the article may exist but be too thin, too ambiguous, or too poorly structured for a generative answer to be produced from it.

A gap that fails both classic search and quick answers is a strong candidate for a rewrite rather than a new page. See Zendesk Suggested Article Acceptance Rate Report for the agent-facing equivalent.

How to interpret the patterns

No-result rate is flat but the top queries keep changing

Healthy. Customers are exploring new areas and you are closing gaps at roughly the pace they appear. Keep working the list.

No-result rate rises right after a product launch

Expected and urgent. Launch content lags launches. This is the highest-value window for writing, because demand is concentrated and the tickets have not been filed yet.

The same query tops the list for months

Someone has looked at this report and not acted, or the article was written and is not being found. Search for the term yourself in the help center. If an article exists, this is a labeling problem, not a content gap.

High no-result rate, low ticket volume

Customers are giving up rather than contacting you. This is worse than it looks — the demand became abandonment instead of a ticket, and it will not appear anywhere in your ticket metrics.

Low no-result rate but high search-to-ticket ratio

Content exists and is found, but does not resolve. Stop writing new articles and start rewriting existing ones. See Zendesk Help Center Article Views Report.

No-result rate spikes in one locale only

Translation coverage, not content coverage. The fix is a translation queue, not a writing queue.

Common mistakes

  • Reporting only the rate. The percentage is a health signal; the query list is the actionable output. Ship both.
  • Working from the top 5. It is a dashboard summary, not a backlog. Build the long list.
  • Ignoring searches with no clicks. Usually a bigger and cheaper opportunity than zero-result searches.
  • Treating every query as a content request. Nonsense strings, single characters, and bot traffic all appear. Set a volume floor before acting.
  • Skipping locale segmentation. The overall rate routinely hides a total gap in one language.
  • Writing an article per query. Cluster first — twelve phrasings of the same question need one good article with the right labels.
  • Forgetting the retention window. Guide search data is retained for up to 390 days, so year-over-year comparisons will eventually run out of history. Export what you need to keep.
  • Assuming agent searches are customer searches. Filter by user role before drawing conclusions about customer demand.

What to do when no-result rate rises

  1. Pull the full query list for the period, not the top 5.
  2. Apply a volume floor to remove one-off and nonsense queries.
  3. Cluster the remainder into themes.
  4. Search your own help center for each theme. If content exists, it is a labeling job — add the customer’s words as article labels.
  5. For genuine gaps, rank by ticket generation, not search volume.
  6. Check locale and brand splits before assuming the gap is global.
  7. Write or translate the top themes, then recheck the same queries in two weeks to confirm they now return results.
  8. Track self-service rate and search-to-ticket ratio afterwards to confirm the gap closure changed downstream demand.

Dashboard template

Panel 1 — No-result rate over time The health trend. One line, rolling window.

Panel 2 — Top zero-result queries (top 50) The writing queue. The most actionable panel you will build.

Panel 3 — Searches by results and clicks The three-way split: no results, no clicks, clicked.

Panel 4 — Top no-click queries The cheap fixes — content exists, wording does not match.

Panel 5 — No-result rate by locale and brand Catches the segment where self-service does not exist at all.

Panel 6 — Search terms that preceded a ticket Turns a content gap into a cost you can prioritise against.

FAQ

How is this different from the search-to-ticket ratio? Search-to-ticket measures whether searching prevented a ticket. This measures whether searching returned anything at all. A search can return excellent results and still end in a ticket; it can also return nothing and end in the customer giving up silently.

What is a good no-result rate? There is no universal benchmark, and chasing one is a distraction. A rate that is stable or falling while your product surface grows is a good sign. A rate rising faster than your content is being written is the problem worth naming.

Should I write an article for every zero-result query? No. Set a volume threshold and cluster related phrasings first. Most of the long tail is misspellings and one-off wording of questions you already answer.

Why do some queries return no results when the article exists? Usually vocabulary mismatch — customers use an old product name, a competitor’s term, or a misspelling. Adding those as article labels is the fix, and it is much faster than writing new content.

How far back does the data go? Guide search data is retained for up to 390 days. If you need longer history, export periodically.

Where should this sit in a support dashboard? Upstream of everything else — before ticket volume. It measures demand that has not become a ticket yet, which makes it the earliest signal available.

See which Zendesk help center searches find nothing - start free