Preparing Salesforce documents for Agentforce starts with the files your team trusts each day. A current service guide can help an agent answer a policy question. An old draft or a scan with missing text can lead it toward a misleading answer. 

Before you connect documents to an AI agent, check what each source contains and who should use it. Then confirm that the agent can find the right content for the right Salesforce record. Good preparation gives the agent stronger evidence, while testing shows how well it uses that evidence. 

Start with the questions your agent should answer 

First, define a small set of questions that matter to your Salesforce users. A service team might want help finding the next step for an open case. A sales team might need answers about the terms in a customer agreement. 

Each question points to a different source and a different access need. A general support guide may suit many users, while a customer agreement needs a clear link to one account. Choosing the questions first helps you avoid adding files that contribute little useful information. 

Salesforce’s Agentforce Data Library guidance explains how libraries index source content for grounding. Grounding gives an agent relevant business information to use when it forms an answer. Your team still needs to choose the sources and configure how the agent retrieves them. 

Approve Salesforce documents for Agentforce 

Next, identify which documents your business accepts as current and correct. A file’s presence in Salesforce does not prove that the content owner has approved it. Teams often keep working drafts beside final copies, which creates doubt about which version should guide an answer. 

Assign an owner to each document group and give that person a clear review task. For a service policy, the owner should confirm the effective date and the rules that apply. Keep the review status easy to find through a field or a clear content process. 

Also, define clear scopes for general guidance and terms for individual customers. A standard policy may describe the usual response time, while an account agreement may set a different target. Define which source takes priority when both contain relevant information. 

Keep old versions for the purposes your business requires, but exclude them from current-answer sources unless the agent needs historical context. For historical questions, include dates that show when each version applied. Clear version rules help reviewers judge an answer against the right evidence. 

Link Salesforce documents for Agentforce to records 

A useful document needs enough context to explain where it belongs. Connect a customer agreement to its account and keep case evidence with the relevant case. A file title alone may give the agent too little detail to choose the right source. 

Use stable record identifiers wherever your retrieval design supports them. Two accounts can share similar names, and staff can rename files without changing the business relationship. Record IDs give your team a firmer basis for matching content to the correct customer. 

Then decide which details should guide retrieval. A case number may identify the right file set, while a product version may narrow a support search. Capture those details consistently and check that the retrieval process actually uses them. 

Salesforce’s example of an agent using a case and a PDF shows a configured workflow that supplies both record context and a specific file. Use that principle when you plan your own setup. Storing a document near a record does not, by itself, configure an Agentforce action to find and use it. 

Make Salesforce documents for Agentforce readable 

After you choose the right sources, inspect what your processing route can read. A PDF may look clear on screen yet contain only scanned images. Check the extracted content before you judge the file’s value for an agent. 

OCR can help capture text in scans, but the output still needs review. Pay close attention to dates and customer identifiers because small errors can change an answer. If a table contains terms or limits, check that the output keeps each value with its heading. 

Processing options also matter. Salesforce’s file upload guidance describes advanced processing through Intelligent Context for file-based data libraries. Test your actual documents with the configuration you plan to use, especially when they contain tables or visual content. 

Within the documents, give each section a clear heading and enough detail to stand on its own. Keep a rule close to its exceptions so a retrieved passage retains the conditions that affect the answer. Add written explanations for important diagrams, following Salesforce’s source content preparation guidance. 

Check access across the full retrieval path 

Next, identify whose permissions the agent uses. Salesforce’s execution context guidance explains that employee agents run as the logged-in user. Service agents commonly run as a dedicated agent user, although the channel and verification setup can change that context. 

Review access to the source documents as well as the actions that retrieve them. For flows, check the run context because system context can grant broader access. For a customer-facing agent, scope access to the verified customer’s records and test that scope. 

Also, check what happens when you copy content into an index or a data library. Do not assume the destination carries every access rule that governs the original file. Confirm how your chosen retrieval route enforces audience restrictions before you connect private customer documents. 

Test both permitted and blocked requests with the identities your agent will use. A user who can read one account’s agreement should not receive terms from another account. A successful access test must show the correct answer and the correct boundary. 

Test answers against source evidence 

Once your sources pass those checks, test the agent with questions your team actually asks. Include a question with a clear answer and another that needs a customer-specific condition. Then ask something that the approved sources do not cover. 

For each response, compare the answer with the source passage and the relevant Salesforce record. Check that the agent uses the correct version and applies the conditions that matter. When your setup supports source references, confirm that they lead reviewers to the content behind the answer. 

Agree on the response you expect when evidence falls short. The agent may need to ask for more detail or hand the request to a person. Your test should check that behavior as carefully as it checks a complete answer. 

Finally, trace errors to their cause before changing instructions. A wrong response may reflect a poor source match or a missing condition in the document. Fixing that cause gives your next test a clearer purpose. 

Keep document sources current 

Document preparation continues after launch. Assign someone to review source changes and confirm when the agent can retrieve the new content. Updating a file in its storage location does not prove that the agent’s search index now reflects the update. 

Define how your team will handle replacement files and changes to access. When you retire a policy, check that current answers no longer draw on it. When a user’s access changes, repeat the relevant permission tests. 

Track the document version alongside the questions you use for testing. That record helps your Salesforce team spot changes in answer quality after a source update. It also gives content owners a practical way to see which gaps need attention. 

Build a stronger document foundation with S-Drive 

S-Drive can support the document work that comes before an Agentforce rollout. Its Salesforce document management capabilities include folder structures and version control, with access based on user roles and permissions. These controls help teams organize the sources they plan to assess. 

For scans, S-Drive’s Document AI and OCR capabilities can extract text and map values into Salesforce fields. Review important values before using them in an agent workflow. Your Salesforce team must still configure and validate the retrieval path that makes the chosen content available to Agentforce. 

A relevant example is S-Drive’s Case File Management and Work Orders use case. Customers and partners can attach documents to the appropriate support case, giving service staff a clear place to manage case files. A team preparing those sources for an agent can first review the attachments and confirm their record links. Then it can decide which content the agent should retrieve. 

That preparation gives the team a defined source set to test against real case questions. The next step is to validate the configured agent’s answers and access within that scope. Contact us or visit our AppExchange page to learn more about what S-Drive can do for your Salesforce document workflows.