A Salesforce chatbot in 2026 is one of three things: an Agentforce service agent that uses a large language model to reason and act on CRM data, an Einstein Bot that follows designed dialogs, or a third-party bot connected to Salesforce messaging. Start new work on Agentforce and keep fixed dialogs where every step must run the same way.
This article first appeared on our blog in May 2023, when the question was how ChatGPT would change Salesforce. It was rewritten in November 2026 for the products Salesforce sells today. If you are new to the platform, read what Agentforce is and how it works first, or see our Agentforce overview.
What is a Salesforce chatbot?
A chatbot is software that holds a conversation with a customer or employee by text or voice. A Salesforce chatbot is one that reads and updates Salesforce records while it talks: it can check an order, open or update a case, book an appointment, or pass the conversation to a person in Service Cloud with the history attached.
The reasons teams deploy one have not changed since 2023:
- Always on. Customers get answers outside business hours and across time zones.
- Scale. One bot handles many conversations at once, so volume spikes do not become queue spikes.
- Speed. Routine questions get an instant, consistent answer.
- Better use of people. Service agents spend their time on the cases that need judgment.
What has changed is how the bot understands people. Older bots matched a message to a trained intent and followed a scripted path. AI agents read the whole conversation, decide what to do next and call actions you allow.
What are your Salesforce chatbot options in 2026?
| Agentforce for Service | Einstein Bots | Third-party bot | |
|---|---|---|---|
| How it works | LLM-based agent that follows natural language instructions, prebuilt subagents and actions | Designed dialogs and manually trained intents | Vendor's own bot, connected to Salesforce messaging and data |
| Handles follow-ups | Yes, it follows the whole conversation | No context of the whole conversation | Depends on the vendor |
| Upkeep | Update a handful of subagents and instructions | High: keep dialogs and intents current | Managed in the vendor's tool |
| Acts on Salesforce data | Natively, through Flow, Apex and other actions | Natively, through dialog actions and Flow | Through APIs and the connector you build |
| Best for | Varied requests, multi-step tasks, new projects | Strictly deterministic, auditable flows | Teams already standardized on another bot platform |
| Pricing model | Per conversation, Flex Credits or per-user add-ons | Part of your Service Cloud entitlements; check your contract | Vendor pricing plus integration work |
The comparison of Agentforce and Einstein Bots in the first three rows comes from Salesforce's Trailhead module Get Started with Agentforce Service Agent (read November 11, 2026). For third-party bots, Salesforce's Bring Your Own Bot for Messaging is the documented route for connecting a third-party or custom bot to Enhanced Messaging channels.
How is Agentforce for Service different from Einstein Bots?
Trailhead describes three practical differences:
- Setup. Agentforce for Service sets up with prebuilt subagents and actions. Einstein Bots need natural language models trained by hand and dialogs designed step by step.
- Maintenance. An Agentforce agent needs updates to only a handful of out-of-the-box subagents and instructions. Einstein Bots need high maintenance to keep dialogs and intents current.
- Context. Agentforce has the context of the entire conversation. An Einstein Bot does not.
In practice this means fewer dead ends. A customer who asks "where is my order, and can I change the delivery address?" can get both answers from an agent that keeps the conversation in view, where an intent-based bot is more likely to handle the first request and miss the second. For a deeper look at the service side, see Agentforce for customer service.
When do Einstein Bots still make sense?
Trailhead notes that Einstein Bots may still be the better fit in regulated industries that need explainable processes with extra governance, and for deterministic, highly specific dialog flows. Examples include identity verification steps, a regulated disclosure that must be read word for word, or a payment flow with a fixed sequence.
Many service teams run both. A fixed dialog handles the steps that must never vary, and an Agentforce agent handles everything else. If you already have Einstein Bots, treat a move to Agentforce as a redesign: map each bot dialog to an agent subagent and its actions, then retire the dialogs you no longer need.
What happened to Salesforce legacy Chat?
Your chatbot is only as reliable as the channel it runs on. Salesforce's Trailhead module Enhanced Chat Migration and Benefits Overview (read November 11, 2026) says Salesforce is retiring its legacy Chat product on February 14, 2026, a date that has now passed. The replacement is Enhanced Chat, formerly called Messaging for In-App and Web, which adds asynchronous conversations that customers can pause and pick back up later.
If your website chat or bot still depends on legacy Chat, move that channel first. In Enhanced Chat, routing flows can send a conversation to an AI agent, a queue or a service rep based on conditions you set. Our Service Cloud guide covers the wider channel setup.
Can you use ChatGPT with a Salesforce chatbot?
Yes, and you no longer need a custom integration for it. In its October 14, 2025 announcement, Salesforce said OpenAI frontier models are available within Agentforce, and that customers can select OpenAI as the default model for the Atlas Reasoning Engine and Prompt Builder. The same announcement said Salesforce apps would come to ChatGPT through the Apps in ChatGPT program, with availability details to follow.
The difference from a do-it-yourself ChatGPT bot is control. Inside Agentforce, model calls go through the Einstein Trust Layer and the agent can only use the actions and records you grant it. A standalone bot calling a public model API leaves you to build grounding, permissions, logging and escalation yourself.
How much does a Salesforce chatbot cost?
Salesforce's Agentforce pricing page (read November 11, 2026, and marked as subject to change) lists these models:
| Model | Listed price | Fits |
|---|---|---|
| Conversations | $2 per conversation | Customer-facing chat and messaging agents |
| Flex Credits | $500 per 100,000 credits; a standard action uses 20 credits (about $0.10), an Agentforce Voice action 30 | Paying per action as usage grows |
| Agentforce add-ons (Sales, Service, Field Service and Operations) | $125 per user per month | Unmetered use by employees |
| Salesforce Foundations | $0 | Trying Agentforce before you commit |
The license is rarely the largest cost. Budget for cleaning knowledge articles, building the actions the agent calls, moving off legacy Chat if needed, and testing. Our Agentforce pricing guide explains the models in more detail, and the Agentforce cost calculator gives a first estimate for your volumes.
How do you launch a Salesforce chatbot that resolves cases?
- Pick three to five high-volume requests. Use case reasons and transcripts to find requests that are frequent, repeatable and low risk, such as order status, returns or password help.
- Fix the knowledge first. An agent answers from your knowledge articles and records. Retire outdated articles and fill gaps before launch.
- Get the channel ready. Run web and in-app chat on Enhanced Chat, not legacy Chat.
- Build the actions. Expose only the Flows and Apex actions the agent needs, with the permissions of a dedicated agent user.
- Design the handoff. Route escalations through Omni-Channel to the right queue, with the transcript and case details attached, and size the queue for it.
- Test with real conversations. Replay past transcripts, including angry customers and off-topic requests, and check every answer and action.
- Launch on one channel and measure. Track resolution rate, escalations, customer satisfaction and cost per conversation, then widen scope.
Agents are only as good as the data they read. If customer data sits in several systems, see how Data 360 grounds Agentforce in one profile.
Is a Salesforce AI chatbot secure?
Agentforce runs model calls through the Einstein Trust Layer, which Trailhead describes as providing secure data retrieval, dynamic grounding, zero data retention with model providers, toxic language detection and an audit trail. One caveat matters for chatbots: Trailhead states that data masking for LLMs is currently disabled for agents (read November 11, 2026).
So security still depends on your design: what records the agent user can see, which actions it can call, what it must never say, and when it must hand off. Review those rules with your security and compliance teams before go-live.
What are common Salesforce chatbot mistakes?
- Measuring deflection instead of resolution. A conversation that ends without an answer is not a success. Measure whether the customer's problem was solved.
- Launching on stale knowledge. The agent will repeat whatever your articles say, including what is out of date.
- No clear way to reach a person. Customers who cannot get to a human leave or complain. Make escalation easy and carry the context with it.
- Too many actions on day one. Start narrow, prove accuracy, then add scope.
- Leaving the channel on legacy Chat. Its retirement date was February 14, 2026. Move to Enhanced Chat before you build on top of it.
How can Techila help with a Salesforce chatbot?
Techila Global Services is a Salesforce Summit Partner headquartered in Frisco, Texas, with 15+ Agentforce programs delivered. Our architect-led Agentforce implementation team plans the use cases, moves channels to Enhanced Chat, builds the actions and handoff rules, and tests agents against real conversations before launch. If you are comparing partners, our guide on how to choose an Agentforce implementation partner lists the questions to ask. Talk to a Salesforce architect about your chatbot plans. We reply within 1 business day.
Frequently asked questions
Does Salesforce have a chatbot?
Yes. Salesforce offers two native options. Agentforce for Service is an AI agent that uses a large language model to understand requests, answer from your knowledge and take actions on Salesforce data. Einstein Bots are the older rule-based bots built from designed dialogs and trained intents. You can also connect a third-party bot to Salesforce messaging channels.
Is Agentforce replacing Einstein Bots?
Salesforce now leads with Agentforce for new service automation. We found no official retirement date for Einstein Bots as of November 2026, and Trailhead still says bots may suit regulated industries that need explainable processes, or strictly deterministic dialog flows. Plan new work on Agentforce and move bots to agents when the use case fits.
How much does a Salesforce AI chatbot cost?
As of November 11, 2026, Salesforce's Agentforce pricing page lists $2 per conversation for customer-facing agents, or Flex Credits at $500 per 100,000 credits, where a standard action uses 20 credits (about $0.10). Agentforce add-ons for Sales, Service, and Field Service and Operations are $125 per user per month. Prices change and vary by contract, so confirm with Salesforce.
Can I connect ChatGPT to Salesforce as a chatbot?
You do not need a separate ChatGPT integration to use OpenAI models. Salesforce announced on October 14, 2025 that OpenAI frontier models are available within Agentforce and can be selected as the default model for the Atlas Reasoning Engine and Prompt Builder. Calls then run through the Einstein Trust Layer and act on your Salesforce data and permissions.
Will an AI chatbot replace my service agents?
No. A well-designed Salesforce chatbot takes routine, repeatable requests and hands the rest to people with the conversation and case context attached, so customers do not repeat themselves. Service agents spend more time on complex cases, exceptions and customers at risk. Plan escalation rules and queue capacity before launch, not after.
What channels does a Salesforce chatbot work on?
Trailhead describes Agentforce for Service running across self-service portals and messaging channels, and Salesforce prices Agentforce Voice actions separately, so voice is an option too. Web and in-app chat should run on Enhanced Chat (formerly Messaging for In-App and Web), because Salesforce scheduled legacy Chat for retirement on February 14, 2026.
