Guide has begun to use semantic search as a way to generate the most accurate search results possible based on the intent and context of user search queries. Unlike search methods that find literal matches to keywords, semantic search captures the meaning of search queries, helping end users and agents search for and locate content without prior knowledge of the exact keywords to use.
What is semantic search?
Semantic search uses artificial intelligence (AI) to process and understand the full meaning and intent of language used in search queries, similar to the way a human would. Specifically, machine learning (ML) and natural language processing (NLP) technologies work together in semantic search to help the search engine understand the user’s intent when they submit a search query. Understanding what the user is actually looking for, regardless of the keywords they use, helps the search engine return and rank the most relevant results.
For example, with semantic search, end users and agents can ask natural language questions instead of worrying about which keywords to use in order to get the best results. New agents can ask “how do I start using Guide?” instead of trying to think of the best keywords to use or the sequence of keywords to use in their query. With semantic search, they can enter their question in the way that makes sense to them, and still see the most relevant articles at the top of the results.
With semantic search, Guide is taking another step towards providing agents and admins a high-deflection help center with an intuitive search experience.
How semantic search works
Guide search has historically relied on keyword matches between queries and content (for example, articles, community posts, or external records). However, it didn’t capture the “meaning,” or semantics, of a query. Semantic search, on the other hand, deploys natural language processing to understand the content of the search query and content. This helps the search engine identify complex patterns that are otherwise missed. In many cases, semantic search is able to capture similarity even when there is not a strong word overlap between the query and indexed content.
Relying on language models, semantic search can translate queries and help center articles into vectors (a numerical representation), and measure the distance between them. Articles closer in the vector space are considered to be more similar by the language model. The language model is trained on many text examples, and through these examples has learned how to accurately interpret the meaning of text accurately.
When semantic search is used, the search engine boosts the relevance of semantically better matches. This pushes the most relevant content, based on intent and context of the search query, to the top of the search results. Each time a search is performed, the search results are ranked not only based on keyword matches but also boosted by semantic matches.
How Guide uses semantic search to improve the search experience
Semantic search offers a more intuitive search experience that lets help center visitors search using their choice of terms. Since Guide has begun to incorporate semantic search, we’ve observed an increase in search quality metrics such as mean reciprocal rank (MRR) and click through rate (CTR). Specifically, we found that MRR improves 7% on average with this first iteration of semantic search for English Zendesk help centers. We’re also seeing more searches with clicks.
By improving the search relevance, and ranking the most relevant results on top, semantic search produces the following help center improvements:
- End users can find the information they are looking for more easily, increasing the deflection power of your help center.
- Agents can be more efficient as they can find answers more quickly.
The degree of impact for semantic search depends on the search behavior of the user. Semantic search has particularly positive benefits for longer search queries, but still improves relevance for all types of search.
Understanding the rollout plan
As we expand our AI model to cover more search channels, we will continue to roll out semantic search into additional areas.
Specifically, the semantic search rollout will be phased into content type, language, and search channel. See the following tables for the current and planned rollout of semantic search based on these parameters.
Rollout by content type
| Content type | Coverage status |
|---|---|
| Articles | Covered |
| Community posts | Covered |
| External content | Planned |
Rollout by help center search type
| Help center search type | Coverage status |
|---|---|
| Two-pane search | Covered |
| Unified search | Covered |
| Federated search | Planned |
Rollout by search channel
| Search channel | Coverage status |
|---|---|
| Help center search engine results page (SERP) | Covered |
| Instant search | Planned |
| Request from article suggestions | Planned |
| Unified search API | Planned |
| Articles search API | Will not be supported |
| Posts search API | Will not be supported |
| Web Widget help center search | Planned |
| Knowledge in Agent Workspace | Planned |
Rollout by language
| Language | Coverage status |
|---|---|
| Arabic | Planned |
| Bulgarian | Not planned |
| Catalan | Not planned |
| Croatian | Not planned |
| Czech | Planned |
| Danish | Planned |
| Dutch | Planned |
| English (all variants) | Covered |
| Finnish | Not planned |
| French (all variants) | Planned |
| German | Planned |
| Greek | Not planned |
| Hebrew | Not planned |
| Hindi | Not planned |
| Hungarian | Not planned |
| Icelandic | Not planned |
| Indonesian | Not planned |
| Italian | Planned |
| Japanese | Planned |
| Korean | Planned |
| Latvian | Not planned |
| Lithuanian | Not planned |
| Norwegian | Planned |
| Persian | Not planned |
| Polish | Planned |
| Portuguese (Brazil) | Planned |
| Romanian | Planned |
| Russian | Planned |
| Serbian | Not planned |
| Simplified Chinese | Planned |
| Slovak | Not planned |
| Slovenian | Not planned |
| Spanish | Planned |
| Swedish | Planned |
| Thai | Not planned |
| Traditional Chinese | Not planned |
| Turkish | Planned |
| Ukrainian | Not planned |
| Urdu | Not planned |
| Vietnamese | Not planned |
How to check if your help center is enabled for semantic search
We’ve begun our semantic search implementation for Guide customers. As the rollout continues, we will continue to expand the feature to all Guide customers. You can use this procedure to determine whether your help center is currently enabled for semantic search.
- Perform a search in your help center.
- On the search results page, open Inspect element in your browser.
For example, in Chrome, right-click the search results page, select Inspect from the menu, then click the Elements tab.
- Select the <head> tag, then examine the
guide_search/semantic_search_enabledelement.If the element is set to "false" (as shown in the image below), semantic search is not yet enabled in your help center. If the element is set to "true," then semantic search is enabled.
28 Comments
Hi Orsolya Forster
Correct me if I'm wrong, but custom help centres will need to be updated manually?
We have been waiting all this year for our help center to get the semantics search...how long will it take Zendesk to roll it out to all? Our existing search is not that great and we have been hoping this would help us, Can we get an update as to what's going on?
Thomas D'Hoe no, custom help centers don't require additional setup.
이지훈 language extensions are on the roadmap but I can't yet share a release time. We're working on the expansion constantly and aim to get the service out to all Guide customers as soon as possible.
Cheers!
At present, content inside an accordion is invisible to the search engine. Will this change?
How do we know if our Help Center has been updated with this new search experience?
Viachaslau Skorbezh Thank you for pointing this out. By mistake, we had forgotten to add the Help center languages that are supported via crowdsourcing. We will add Ukranian together with the other missing languages to the list.
Orsolya Forster Gorka Cardona-Lauridsen Why Ukrainian language is not mentioned in the list even with "Not planned" clause? Guide supports this language so please add such information
Hey!
About the metrics CTR and MRR, how will they be available for us to analyze?
Is there a dash in which we can check users' most frequent searches (terms) and the metrics (CTR mainly)?
Thanks!
Hi Zendesk team, is it already available for Community search as well?
Becky Evans no change in knowledge management is anticipated for semantic search, as the keyword search did not change, it got boosted with semantic matches.
John, thanks for the feedback!
Gwyn Mabo, we will try to share the next milestones as we see them more clearly.
Cheers!
Brett Bowser:
Hi Becky,
Thanks for your response.
This implementation is automatic and requires no additional work on your end. However, we are still expanding this feature and our dev team will definitely give us an update on this.
Hi Zendesk,
I have confirmed that our HC is not yet enabled for Semantic search by checking the <head> element. When can we expect this feature to be available for us?
-- UPDATE --
Actually I realized it was only enabled in LIVE theme but when previewing another theme in development, Semantic search is not enabled. It would be great to have the feature enabled consistently for testing during development.
Mike McGuffie Assuming that the text is written in the article editor and thus part of the article content it will keep being searchable.
Hi, this sounds like a great enhancement. Is there anything we can do to help train our own help center?
Josef Prandstetter instant search is not yet supported, neither is federated search. As we're evaluating these expansions down the line, I'll be able to share a timeline, but not at this moment. Our next milestone is to expand the service to all English help centers with the Copenhagen theme, as Brett wrote above, by early Q3.
We are working on a quick solution so customers can check if the service has been enabled for them already or not yet. Cheers!
Darenne No, I meant the feedback form linked by Orsolya Forster above and at the end of the other article https://support.zendesk.com/hc/en-us/articles/5641000272922-Announcing-semantic-search-in-Guide-
Hi Becky Evans
Thanks so much for patiently waiting on this matter! I just confirmed with our internal team that training AI is not yet on the roadmap. If you encounter any issues, we suggest that you file it as a request via the same link provided by Orsolya.
We're keen to learn more about your help center search experience. To share your feedback with us, use this Feedback form. Cheers!
Hi Becky Evans,
You mentioned that when there's a search that doesn't return the expected results, it asks you to submit a feedback form. Can you please clarify to us what you meant by the feedback form? In what instance are you seeing it too?
We have checked our Help Centre and semantic search is not yet enabled. Will we receive an update to let us know when it is? Are there any time scales you can supply for when this will be enabled for all accounts?
Hi Darenne, Thanks, and sorry I wasn't clear. I understand that it is automatic and see that it has already been rolled out in our help center. My question is about when we see searches that don't return expected results. I see that Zendesk is requesting us to submit a feedback form each time, but that's not very practical. It would be nice if there were things our Agents could do on our end to help train the AI in real time. Cheers!
Hello,
Semantic search is not enabled for me yet either. Are time scales available when this will be enabled?
I believe early Q3 is past us. Is there a clear timeline on when to expect this feature?
It's what I was expecting!
Are there any plans to apply it to the Korean help center? If so, when?
Please sign in to leave a comment.