Suggested Article
A Suggested Article is a Salesforce Knowledge article that Service Cloud surfaces to an agent automatically while they work a case, instead of making the agent search for it.
Definition
A Suggested Article is a Salesforce Knowledge article that Service Cloud surfaces to an agent automatically while they work a case, instead of making the agent search for it. Salesforce compares the text on the case against the text in published Knowledge articles and shows the closest matches inside the Knowledge component on the case record.
There are two engines behind the feature. The classic version uses keyword search and turns on automatically when you enable Lightning Knowledge. The smarter version, Einstein Article Recommendations (now also branded as Agentforce Article Recommendations for Cases), trains a model on which articles agents actually attached to similar cases in the past.
How Salesforce picks the right article for a case
Keyword suggestions versus Einstein recommendations
Salesforce ships two ways to suggest articles, and they work very differently. Keyword-based Suggested Articles turn on the moment you enable Lightning Knowledge. When a case is saved, the search engine scans the case fields you selected and looks for published articles that share those words. It is fast and needs no training, but it cannot learn. It ranks purely on text overlap, so a case full of vague language gets weak matches. Einstein Article Recommendations takes the opposite approach. It builds a predictive model from your own history, reading the language, key words, phrases, and text field values in a case and comparing them against articles. The official docs are blunt about the gap: keyword suggestions rely solely on keyword search and cannot refine their suggestions or incorporate data from past cases. Einstein can. It weighs how often agents attached a given article to similar cases and how often they dismissed it as unhelpful. That feedback loop is the main reason mature teams graduate from keyword matching to the model.
What data Einstein reads to rank articles
Einstein Article Recommendations draws on three sources, and knowing them shapes how you set the feature up. From the Case object it reads the Subject and Description fields plus any custom text fields you select. From the Knowledge side it reads the article Title, Summary, and the custom text fields you combine into an article body for analysis. The third source is the CaseArticle object, which records every time an article was attached to a case and supplies the historical signal the model learns from. The model does not treat every field equally. During setup you rank case fields by importance so Einstein knows what to read first. Salesforce advises choosing data-rich fields, meaning text fields that are rarely blank and usually contain several words that describe what the case or article is about. A custom picklist or a one-word status field adds little. A well-written Description or a detailed article Summary adds a lot. The model also detects language per case and matches articles in the same language.
Where suggestions appear and how agents use them
Both suggestion types surface in the same place: the Knowledge component inside the Lightning Service Console, docked on the case record page. As an agent opens a case, candidate articles appear automatically without any click. The agent can read an article in place, then attach it to the case or share it with the customer using the dropdown next to each result. If the automatic matches miss, the same component doubles as a search box so the agent can look further. Attaching an article is not just a convenience for the agent. Every attach writes a row to CaseArticle, which becomes training data for Einstein. So the act of resolving a case today improves the recommendations on tomorrow's cases. Removing an article works the same way through the component dropdown. This is why article hygiene matters: if agents attach the wrong article to close a case quickly, they teach the model a bad association that surfaces again later.
Languages and how the model stays current
Einstein Article Recommendations uses a single model to generate recommendations across seven languages: Dutch, English, French, German, Italian, Portuguese, and Spanish. Before you rely on a language, confirm it is also active in your Knowledge settings, otherwise there are no published articles in that language for the model to return. The model detects the language of each incoming case and keeps its suggestions in that language. The model is not static. Salesforce periodically retrains, or rebuilds, it to capture changes in your case and article data, and a retrain typically finishes within a day. A newly built model only replaces the live one if it produces better recommendations, so a bad training cycle will not degrade what agents see. The model draws on roughly the previous two years of case history, which means a brand new org with little data gets weaker results at first and improves as case volume and attach history accumulate.
Who can see suggestions and what it costs
By default every Lightning Knowledge user sees Einstein Article Recommendations in the Knowledge component. A Lightning Knowledge user is a standard user who holds the Knowledge User permission and has read access to Knowledge articles in Lightning Knowledge. If you want only certain teams to see recommendations, you can restrict access during setup rather than leaving it open to all. Keyword Suggested Articles have a lighter footprint. They come with Lightning Knowledge and require the same Knowledge User access, but no model and no extra enablement beyond the Support Settings checkbox. Einstein Article Recommendations is an Einstein for Service feature and depends on your edition and add-on entitlements, so check that your org is licensed before you plan a rollout. The practical sequence for most teams is to start with keyword suggestions on day one, gather a few months of attach history, then enable and train Einstein once there is enough data for the model to learn from.
Measuring whether suggestions actually help
A suggestion that no agent opens is just noise on the page. The metric that matters is the attach rate: how often an agent attaches a suggested or recommended article rather than searching manually or closing the case with no article at all. A healthy program watches that rate climb as the model learns and as the article library grows to cover real case topics. Coverage is the other half. If half your inbound cases are about a topic with no published article, no engine can suggest one, and agents fall back to manual work or freehand answers. Treat gaps in the suggestions as a content backlog signal: the topics that never produce a good match are the articles your Knowledge team should write next. Over time the loop tightens. Better articles produce better suggestions, agents attach them more, and those attaches feed the model so the next round of suggestions is sharper still. This is the difference between Knowledge as a dusty archive and Knowledge as a live part of case handling.
Enable suggested articles on cases
Keyword Suggested Articles arrive automatically with Lightning Knowledge, but you still control which case fields drive the matching, and you turn the feature on for cases in Support Settings. These steps cover the keyword version that any Knowledge org can enable. Einstein Article Recommendations is a separate, model-based setup layered on top.
- Enable Lightning Knowledge
From Setup, open Knowledge Settings and enable Lightning Knowledge if it is not already on. This switch makes the keyword-based Suggested Articles feature available and exposes the Knowledge component for case pages.
- Turn on suggested articles for cases
From Setup, enter Support Settings in the Quick Find box, select Support Settings, click Edit, then select Enable suggested articles. This tells Salesforce to surface matching articles on cases as agents work them.
- Choose the case fields to match on
In Knowledge settings, select which case fields the search engine should read when finding articles. Pick data-rich text fields such as Subject and Description, since fields that are usually blank or hold one word produce poor matches.
- Add the Knowledge component to the case page
Using the Lightning App Builder, place the Knowledge component on your case record page in the Service Console so agents see suggestions in their workspace. Confirm agents have the Knowledge User permission and read access to articles.
The Support Settings checkbox that activates keyword article suggestions on cases.
The set of case fields the search engine scans for keywords; Subject and Description are the strongest choices.
An optional model-based upgrade that learns from past case-article attaches instead of plain keywords.
By default all Lightning Knowledge users see recommendations; you can restrict visibility to specific users during Einstein setup.
- Keyword suggestions cannot learn. If matches are weak, the fix is better case fields and better article text, not more configuration.
- Einstein needs history. A new org with little case-article attach data gets thin recommendations until volume builds, drawing on roughly the prior two years.
- Confirm each supported language is also active in Knowledge settings, or the model has no published articles to return in that language.
- Sloppy attaches train a bad model. Agents who attach the wrong article to close a case fast teach Einstein an association that resurfaces on future cases.
Trust & references
Cross-checked against the following references.
Straight from the source - Salesforce's reference material on Suggested Article.
Hands-on resources to go deeper on Suggested Article.
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
Q1. What is a Suggested Article?
Q2. What drives the suggestions?
Q3. Why are they valuable?
Discussion
Loading discussion…