Prepare your Salesforce catalog for AI search

AI chat referrals can expose outdated product names, prices, and eligibility rules. Use this checklist to align Salesforce catalog data with public product content.

AI chat referrals make catalog accuracy an operating issue. If a buyer finds an outdated product description, price, or eligibility rule before reaching your site, sales and service inherit the correction.

Salesforce published findings from its fourth State of Commerce report on July 28. Salesforce reports that agentic search activity, used as the first step in product discovery, grew 200% year over year. The report combines surveys of 3,450 commerce professionals and 4,690 consumers with behavior from more than 1.5 billion global shoppers.

The report does not prove that every company will see the same referral pattern. It does establish a practical reason to review the product facts that appear across your public site, sales materials, and Salesforce records.

A buyer may form an opinion about an offering before opening a case or speaking with sales. That makes product names, plan descriptions, pricing rules, and ownership more important.

What is Salesforce agentic search?

Salesforce uses “agentic search” to describe product discovery that starts in an AI chat, AI assistant, or another conversational interface.

It is not a Salesforce setting that an administrator turns on. It describes where a buyer begins their research.

Salesforce’s July 28 analysis reports that traffic referred from AI chats grew between 150% and 428% year over year in every quarter measured. Overall traffic grew at single-digit to low double-digit rates over the same period.

For a Salesforce team, the important question is not whether every AI-generated answer will be correct. It will not. The useful question is whether the public information an AI tool can find matches the facts your sales and service teams use.

A vague product page produces a vague answer. An active price for a retired package produces a sales and support problem.

The same Salesforce analysis reports that 86% of commerce leaders expect large language models to matter for product discovery within the next year. That is survey data, not a forecast for every buyer or industry. It is enough reason to check the catalog basics.

Does AI search require Agentforce or Commerce Cloud?

No. Salesforce’s report does not state that Agentforce or Commerce Cloud is required for AI-driven product discovery.

A company can receive AI-referred traffic whether its public catalog runs on Commerce Cloud, another commerce platform, a content management system, or a custom-built site. The work usually crosses marketing, commerce, operations, and the team responsible for product information.

This differs from deploying an agent on your own website or connecting a commerce channel to an AI provider. Those are controlled product decisions with defined data access and user experiences. AI search is broader. It concerns what external tools can locate and describe from public sources.

Salesforce administrators still have a role. Salesforce may hold the product data, customer eligibility rules, service history, and sales context needed after a buyer submits an inquiry.

Keep the boundary clear:

• Public product descriptions can support discovery.
• Customer-specific prices, contract terms, inventory commitments, and internal product notes should remain behind the appropriate access controls.
• A public page should not become a substitute for a customer’s quote, contract, or eligibility review.

How should Salesforce teams prepare product data for AI search?

Start by naming the source of truth for each product family. One owner must be accountable for approving product facts, even when marketing, sales operations, finance, and product teams contribute content.

If Sales Cloud is the source for product records, review the fields your team maintains on Product2. At minimum, check:

• Name distinguishes similar offerings without relying on internal shorthand.
• ProductCode matches the SKU or reference used by sales and operations.
• Description explains the offering in plain language.
• Family supports consistent grouping where your team uses it.
• IsActive reflects whether the product is currently sellable.
• Related PricebookEntry records have the correct UnitPrice, currency, and active status.

These fields do not replace a full commerce catalog. They do expose frequent failures: a product name written for an internal report, a description copied from an old sales deck, or an active PricebookEntry for a package that sales no longer offers.

Then compare Salesforce records against the public product page, any public feed, and sales enablement materials. The same offering should not appear under three names, two eligibility statements, and different price guidance.

Use real buyer questions for the review. Choose questions your sales or service team already receives:

• “Which plan includes [specific capability]?”
• “What is the difference between [product A] and [product B]?”
• “Is [product] available for a company of our size?”
• “What does [product] cost?”
• “Can I buy [product] without [related service]?”

The goal is not to make every answer public. The goal is to identify where a buyer could receive an incomplete or misleading answer before entering your sales process.

What product data should remain private?

Customer-specific commercial terms should remain private, including negotiated pricing, contract commitments, account-level eligibility, and service history.

This distinction matters most when a product catalog includes multiple price books, regional pricing, regulated offerings, or account-specific bundles. A public page can describe a starting price or a pricing model when that information is approved for publication. It should not expose a PricebookEntry that applies only to a customer segment, a contract, or an internal sales process.

Review related sources as well:

• Knowledge articles that describe discontinued products.
• PDF brochures stored in public resource libraries.
• Partner pages that use an old product name.
• Public case studies that describe a retired feature set.
• Sales email templates that send buyers to outdated URLs.

A correct Product2 record does not fix an incorrect public page. A correct public page does not fix a sales team quoting from an outdated price book. The review must cover both.

How should teams measure AI chat referrals?

Measure AI-referred traffic before changing the content plan or adding a new AI tool. Referral classification depends on your analytics configuration, so start by confirming what your web analytics platform can identify.

Salesforce reports that only 32% of organizations have fully defined AI success metrics and key performance indicators. A small baseline is more useful than a broad program with no owner.

Run a 30-day review for one product line:

1. Identify AI chat and assistant referrals where your analytics platform can classify them.
2. Track the landing page, product-detail views, form starts, and completed inquiries from those referrals.
3. Review sales and service conversations for incorrect assumptions about price, eligibility, features, or availability.
4. Record the page, document, or catalog entry that needs correction.
5. Review the findings monthly with the product-data owner and Salesforce administrator.

Keep the first review narrow. One product family is enough. The purpose is to create ownership and evidence before adding feeds, agents, or a larger content project.

Salesforce also reports that only 27% of organizations have fully unified customer data across sales, service, marketing, and commerce. That does not mean every mid-market team needs a large data program. It does mean AI-driven discovery can reveal inconsistencies that already exist between those teams.

This week, choose one active product family. Compare its Product2 and PricebookEntry records with its public page, then assign an owner for any mismatch.