Help center performs a full-text search of your knowledge base articles, community posts, and if Federated search is enabled and if configured, external content such as blogs or websites.
When a user enters a search query in the help center, the search algorithms get to work, looking for indicators of the most relevant results and ranking them. The relevant snippet from the content of your knowledge base article, community post, or external content is created and the search results and relevant search filters are displayed on the search results page.
Generative search is available with all Zendesk Suite and Guide plans, and provides AI-generated answers to users' search queries in your help center based on your content. When generative search is activated, users can view generated answers without clicking through search results or scanning related articles. However, users can still easily click into the articles for more information.
Note: Help center search is one way to search for content in your help center. For information about other search methods, see Understanding help center search methods.
The articles contains the following sections:
- Which content is included and excluded in search results
- Understanding the relevance score in search results
- Understanding boosts in search results
- Other relevance features
- Improving the search experience for end users
Which content is included and excluded in search results
When you search the help center, you are searching all knowledge base articles (first 10,000 characters of each article) in your native help center. Your search can also include the following:
- Articles and community posts from other help centers in your application, if multiple help centers are enabled and search has been configured to include results from those help centers. See Enabling search across multiple help centers.
- Content from external sources, if federated search is enabled and search has been configured to include results from external content. See About Zendesk Federated Search.
Articles and community posts
When an article, post, or external content is returned, the search engine attempts to find a snippet from the document body that matches the search. If there is no match in the document body or comments, an extract from the start of the document body is returned. If there is a match, the search engine divides the article or post into sentences and ranks each sentence based on the number of matches. The score is then normalized by the fragment length to ensure that fragments are not too small.
The default snippet size for a search result is 120 characters, although results can vary slightly because the snippet engine will always try to return a fragment that includes a complete sentence.
These items might also be included in the search:
- Restricted content - Only users with permission to access restricted content will see in search results.
- New content - When you add or update content it usually takes only a few minutes before the content is indexed and can be searched.
- Comments - Article and post comments are included in Help Center search results. Comments will show up in the search results as long as the search result snippet is matched in the comment. If there are multiple comment matches within one community post, the algorithm will pick the most relevant comment snippet.
- Hyperlinks - URLs within the document body and linked text are included in Help Center search results.
These items are not included in the search:
- Attachments - Content within article attachments is not included in Help Center search.
- My Activities - Search in My Activities in Help Center is limited to tickets and, specifically, tickets you have access to. It does not include articles.
External Content
If external content is available, the title of the external content is displayed along with a link to open the content in a new browser tab and a snippet from the document body that matches the search. If there is no match in the document body, an extract from the start of the document body is returned.
External content source types and filters are defined during search crawler setup or during federated search API configuration. See About Zendesk Federated Search.
Understanding the relevance score in search results
The ranked search results are based on relevance scores and are displayed to the user in descending order of their scores.
Relevance scores are indicated by a weighted average per field score. A field is a part of a record, representing an item of data. Some examples are:
- Matches in an article or post title field score higher than matches in other fields.
- Matches in article labels score higher than matches in the body field.
These are the current field weights:
|
Field |
Weight for |
Weight for |
Weight for external content |
|
Title |
3 |
3 |
3 |
|
Details (Body of a community post) |
N/A |
1 |
N/A |
|
Body |
1 |
N/A |
1 |
|
Labels |
2.8 |
N/A |
N/A |
|
Comment |
1 |
1 |
N/A |
|
Section title |
1.5 |
N/A |
N/A |
Relevance scores are also impacted by a text analysis process that considers the following factors:
- Exact match - Results that exactly match a word in the search string. This scores higher than a stemmed match.
- Stemmed match - Results where a word matches after stemming. For example, the plural form of a word generally matches the singular form.
- Term frequency - Number of matches returned in a single field. Higher term frequency increases the score.
- Field length - Matches in shorter fields score higher than results in longer fields. For example, if you have a single word search, that matches a one-word title, that will score higher than a hit in a long article title with many words.
- Proximity boost - The score is boosted when all the search terms are close together in the same field. For example if all the search terms are included in an article title this puts them in close proximity and gives the result higher relevance.
- Phrase boost - In multiple term queries, exact word order is preferred. For example, when searching for “car park”, results containing “car park” are ranked higher than results containing “park car.”
- Query length - For one and two word queries, the algorithm returns only documents that match all the search words. For longer queries, 40% of the query terms must be present in a document for it to become a search result.
- Overall quantity and quality of relevant results.
- Semantic search -Guide has begun to use semantic search as a way to improve the ranking and generate the most accurate search results possible based on the intent and context of user search queries. Semantic search is being rolled out in phases to all content types, languages, and search channels. See About semantic search and how it works.
Understanding boosts in search results
In addition to text analysis we give extra weight to certain features of articles and posts. These include:
- Article votes - End users can rate articles as “helpful” or “unhelpful” so that over time an article may develop a score like “10 of 50 users found this article helpful.” We give articles with a higher percentage of positive votes a boost so that they show up a higher in results than they otherwise would. The more overall votes an article has weighs in too; for example, an article with a rating of 10 out of 50 gets more weight than one with 10 out of 100.
- Community post votes (requires Guide Professional or Enterprise) - End users can rate community posts as “helpful” or “unhelpful,” just as they can for articles. The percentage of positive votes functions as a boost and makes a certain post rank higher than it otherwise would.
- Labels (requires Guide Professional or Enterprise) - Labels are elements you can use to influence the relevance score of your articles in search results. Consider using labels carefully to balance your Knowledge base search results.
Other relevance features
Fuzzy search
Fuzzy search is available in certain languages and is a process where an article or post is deemed to be relevant to a search query even when there is not an exact match to the search terms in any of its fields. We use this technique to protect users from spelling mistakes.
Unlike stemming, which removes suffixes and prefixes to get to the root of a search term, fuzzy search uses edit distance to identify search results that contain terms close to the query terms. For example, if you search for “user segmemt” the search engine will also return results containing “user segment”.
The current rule for finding approximate matches is:
- Terms up to maximum two characters must match exactly
- Terms containing three to five characters are allowed one typo
- Terms longer than five characters are allowed two typos
Fuzzy search is not available in Japanese, Korean, and Chinese help center languages.
Optimized language support
For content written in certain languages, we apply specific optimizations.
Stemming is language specific. In English, the search engine knows that if you search for the term “films” you also want results that contain the singular form “film.” Similar rules apply to all languages.
Stop words are another language-specific factor. Stop words are the most common words in a language that are usually excluded from the search query to avoid returning too many results. For example, in English, “the” is a stop word.
The Help Center search is aware of the stemming rules and the stop words for a number of languages that together make up up to 99% of all searches performed by end users.
We are optimizing searches in the following languages:
Arabic,
All other languages benefit from basic search support.
Improving the experience for end users
There are a number of ways that you can improve a user's search experience.
Consider changing the color of search results highlighting in your custom theme. Use CSS to change the appearance of your search results keyword highlighting.
You can use the search analytics dashboard to review Help Center search terms from the last 30 days. For each search term you can see the number of searches for that term, number and type of search results returned (if any), click-through, and the next action taken.
Note: This requires Guide Professional or Enterprise.
Search analytics gives you insight into what your customers are looking for and where they are failing to find answers. To make end users more successful you can analyze search data, then take actions to improve search results and your knowledge base content. See Analyzing help center search results with Explore.
To access the Search dashboard in Explore
- In the Zendesk product tray, click the Explore icon (
). - From the list of dashboards, select the Zendesk Guide dashboard.
- Click the Search tab.
Providing tips for your end users to find content more easily
There are a number of operands you can recommend to help end users locate content in search.
-
Find multiple words: Use double quotes (") around each word to find content that contains all those words.
For example,"article" "title" "section" "author"retrieves content that contains all four words, in any order. Make sure you put spaces between the search words, otherwise the search handles the text as one string.
You'll get hits if there is a stemmed version of a word (e.g. articles). You won't get hits where content contains only the words title and section, for example.
If you use single quotes (') around a word, the single quotes are ignored. If you search for'article' 'title' 'section' 'author', you'll see hits for all content that contains any of the words title or article or section or author (exactly as if you had searched without the single quotes). -
Find a phrase: Use double quotes (") around a phrase to find content that contains all the words in that phrase.
For example,"article title"retrieves all content that contains the words article and title, in that order. You'll also get hits if there is a stemmed version of the word (e.g. articles). You won't get hits where content contains only the word title, for example.
If you use single quotes (') around a phrase, the single quotes are ignored. -
Exclude results containing certain words: Use the minus operator (-) in front of the search term to find content that does not include that word or phrase.
For example, reporting bugs -support returns content containing the words reporting and bugs, but excludes those that contain the word support from the result set.
Note: Do not repeat the same word after a minus operator (-). For example, the search"cannot send -cannot set"repeats the word “cannot” and therefore won’t return any results. Instead, search for"cannot send -set"so that the search returns results excluding the articles that contain the phrase “cannot set”. -
Combine operands for advanced search: you can combine the operands above to find a very specific set of results.
For example, "reporting bugs" -support returns hits for content that contains both the words reporting and bugs, but does not contain the word support.