Chatbase Review 2026

How we evaluated this

We evaluated Chatbase as a website-first AI agent builder, not as a full helpdesk or omnichannel CRM. Our scoring weights five buyer-critical dimensions: setup speed (how quickly a useful agent can go live), knowledge fidelity (whether answers stay close to the sources), channel breadth (where conversations can start and continue), action depth (what the agent can do beyond answering), and pricing clarity (how message credits, overages, and add-ons affect the real cost).

Key buyer tests: Can the agent answer questions from your actual website or documents without drifting? Can it hand off cleanly to a human when it is unsure? Does the credit model fit your monthly conversation volume? Can it perform the light actions you need, such as collecting a lead, checking an order, or booking a meeting?

Pricing calculation to run

Model message credits and add-ons before choosing a plan

Chatbase pricing looks simple until you factor in message credits, model choice, and the three public add-ons that often show up after the free trial. A plan that covers a low-traffic website can become expensive once WhatsApp or voice volume rises, auto-recharge kicks in, or branding removal is required for a client-facing site. Build a 12-month model by channel and by add-on before treating the sticker price as the real cost.

• Count web chat, WhatsApp, email, and voice conversations separately instead of using one blended number.
• Price the three public add-ons from the official rate card: auto-recharge credits at $40 per 1,000 message credits, extra agents at $300 per AI agent per year, and remove "Powered By Chatbase" at $1,188 per year.
• Note that Free users must subscribe to a base plan before any add-on, and billing intervals must match: monthly plans take monthly add-ons, yearly plans take yearly add-ons.
• Estimate reviewer time for the first 30 days because accuracy work (knowledge cleanup, grounding, escalation thresholds) is unpaid labor that determines whether credits buy resolved tickets or follow-up complaints.
• Run base, peak, and overage scenarios. Advanced models and action-heavy conversations consume credits faster than simple Q&A, and auto-recharge credits do not expire once purchased.

Worked example: Standard at $120/mo annual, plus branding removal (~$99/mo equivalent), plus one auto-recharge of 3,000 credits in a busy month ($120) is already a different budget than the plan card alone. If you also need a second agent for a second brand or language, add $300/year. Put those lines in the same spreadsheet as seat growth and peak-season volume.

Who Chatbase fits best

Channel coverage to verify

Chatbase supports more channels than a basic chatbot, but coverage and quality vary by plan. Ask which channels are included in your target plan and which require a separate provider or add-on. The list below is a starting point for a channel-by-channel demo checklist.

• Website chat widget and embed behavior.
• WhatsApp Business API availability and whether it uses a third-party provider with separate billing.
• Messenger and Instagram Direct Messages, including rate limits and handoff behavior.
• Slack and WordPress integrations for internal or site-embedded use.
• Voice agent support and concurrent call limits on Standard and higher plans.
• Email support and thread history handling.
• API access for custom channels, which is available on Standard and higher plans.

Workflow execution and handoff

Chatbase agents can do more than answer questions. On paid plans they can run AI actions and procedures for tasks such as looking up an order, collecting a lead, booking a meeting, or recommending a product. The buying question is whether those actions work reliably for your real policies and edge cases.

When a human needs to take over, Chatbase can route the conversation through a helpdesk view. Configure handoff triggers for low-confidence answers, billing or refund requests, and explicit customer asks. A clean handoff with full context keeps the agent useful rather than frustrating.

• Website support and FAQ deflection trained on help-center content.
• Lead capture with email or scheduling actions.
• Ecommerce use cases such as order status, product recommendations, and returns guidance.
• Voice and phone workflows on plans that include telephony.

Pricing checked July 29, 2026

Chatbase pricing: Free to $400 per month billed annually

Chatbase publishes four paid tiers plus a free plan. The annual rate card offers a 20% discount. Message credits are the main usage gate: Free includes 50, Hobby includes 500, Standard includes 4,000, and Pro includes 15,000 per month. Add-ons such as auto-recharge credits, extra agents, and white-label removal can change the real cost.

Source: Chatbase official pricing page. Confirm the live rate card, taxes, credit overages, channel-provider charges, and annual-commitment terms before signing.

Add-ons checked July 29, 2026

Chatbase add-ons that change the real bill

Plan stickers are not the full commercial surface. Chatbase publishes three self-serve add-ons on the pricing page. Free accounts must move to a paid base plan before any add-on. Billing interval must stay consistent: a monthly plan can only take monthly add-ons, and a yearly plan can only take yearly add-ons. Cancelling an add-on removes it immediately and credits unused time per Chatbase workspace settings documentation.

Sources: Chatbase pricing add-ons and workspace settings / add-ons docs. Confirm live prices, monthly vs yearly packaging, and Enterprise-only items such as custom domains before you sign. Third-party writeups sometimes list older monthly branding fees; prefer the official rate card.

Buyer framing: branding removal alone is roughly an 80% uplift on Standard's $120/mo annual rate if you amortize $1,188 across twelve months. Auto-recharge is the right tool for burst traffic, but it is also how a "predictable" plan becomes variable. Set a threshold and a monthly ceiling in internal policy so finance is not surprised by three top-ups in a launch week. Extra agents are cheap compared with spinning a second SaaS, but only if the second agent has a clear owner and its own knowledge hygiene. Otherwise you pay twice for two half-maintained bots.

AI capability, grounding, and model choice

Chatbase lets you choose from several large language model families. Free plans have limited model access, while paid plans provide access to advanced models from OpenAI, Anthropic, Google, DeepSeek, Meta, MoonshotAI, Mistral, and Z.ai. The model layer is only as useful as the knowledge and guardrails you provide. Clean source material and clear escalation thresholds matter more than picking the most expensive model.

Accuracy is the real product test. A fast wrong answer creates more work than a slow right one: the customer writes again, a human must correct the misinformation, and overall handle time rises. Chatbase's own accuracy guidance treats hallucination as real and damaging, but also as one of the more fixable failure modes when the agent is constrained to verified sources and told to escalate when it cannot answer. Other failure modes matter just as much: stale knowledge (true as of last training, false after a policy change), partial retrieval (the bot answers the wrong sub-question inside a long document), and no live system connection (it cannot answer "where is my order" without Shopify or similar actions).

• Knowledge retrieval from uploaded documents, website URLs, Notion pages, YouTube videos, and structured Q&A.
• No-code agent builder with instructions, branding, procedures, and guardrails that can force "answer only from sources" behavior.
• Confidence thresholds and escalation rules so low-confidence or out-of-scope questions reach a human instead of a fabricated reply.
• Widgets for interactive UI components the agent can display during conversations.
• Backstage debugging to inspect agent behavior and improve instructions after failed conversations.
• Testing playground to try models and configurations before going live.

Limitation buyers miss

Hallucination, stale answers, and what Chatbase can actually control

Every RAG-style support agent can invent a confident answer when the knowledge base is silent, contradictory, or out of date. Chatbase does not eliminate that risk by existing. It gives you controls (sources, instructions, guardrails, procedures, confidence-based handoff, and live actions) that reduce the risk when you operate them. If you skip the operating loop, you still ship a polished wrong answer.

Chatbase's public accuracy guidance separates four failure modes that buyers should test in a pilot, not argue about in a sales call:

• Hallucination: the model invents a policy or price that is not in your sources. The Air Canada bereavement-fare case is the industry warning: fabricated policy can become a legal and brand liability. Constraint instructions ("only answer from verified sources; escalate when unsure") plus clean retrieval cut most of this class of error.
• Stale knowledge: the bot is not inventing; it is quoting last quarter's return window or pricing tier. Accuracy decays the day after a product or policy change if nobody retrains or updates sources. Auto-retrain (plan-dependent) and a weekly source review are operational requirements, not nice-to-haves.
• Partial retrieval: the bot finds the returns document but misses the "opened items excluded" clause buried lower in the same page. Structure sources as clear Q&A, tables, and short headed sections so retrieval surfaces the right sub-answer.
• No live system context: documentation alone cannot answer order status, subscription state, or inventory. Connect Shopify, Stripe, or your helpdesk actions so the agent reads live data instead of guessing from a PDF.

How to measure accuracy in a Chatbase pilot (not vanity deflection):

• Repeat-contact rate within 48 hours on the same issue. Above roughly 15% usually means "resolved" was a label, not a fix.
• AI-handled CSAT vs human-handled CSAT. A large gap means the bot is hurting the relationship on the volume it keeps.
• Unknown and fallback logs. Review weekly. They are a map of missing knowledge, not a badge of failure.
• Abandon-to-contact-form rate. Customers who open chat then email support without finishing are silent accuracy failures.
• Handoff quality. Humans should receive transcript, customer identity, attempted answer, and reason for escalation without restarting the story.

Practical setup checklist inside Chatbase before you scale traffic:

1. Ground every response.Instruct the agent to answer only from connected sources, cite the policy or article when the answer is high stakes, and say it does not know when retrieval is empty.
2. Set confidence and sensitive-topic gates.Force handoff on billing disputes, legal claims, medical or safety content, VIP keywords, and low-confidence retrieval. Do not let the model freestyle refunds or policy exceptions.
3. Connect live actions where the question is live.Order lookup, subscription status, and appointment slots should hit Stripe, Shopify, Calendly, or your API rather than a static FAQ.
4. Run a scored sample set.Take 50 real tickets, score grounded correctness, not just "bot answered." Freeze the sample when comparing models or plans so you do not change the test and the tool at once.

Honest limit: no public Chatbase page publishes an independent, third-party hallucination rate for your industry. Vendor blog claims that grounding and confidence thresholds can cut hallucination rates substantially are directional, not a guarantee for your catalog. Your pilot numbers are the only numbers that count for a buy decision.

Feature areas to verify

Use this checklist during a free trial or demo. Do not accept a generic feature list; ask the vendor to show each one working with your data and policies.

• No-code agent builder with instructions, procedures, guardrails, and conditional logic.
• Knowledge-base training from documents, URLs, FAQs, Notion, and YouTube.
• AI actions and procedures for order lookup, scheduling, lead capture, and product recommendations.
• Chat, email, voice, and messaging channel deployment (plan-dependent).
• Helpdesk view for human escalation and ticket routing.
• Analytics for topics, sentiment, resolution rates, and channel activity.
• Grounding and hallucination controls: source-only instructions, confidence thresholds, escalation rules, and Backstage debugging of bad answers.
• Add-on surface: auto-recharge credits, extra agents, and branding removal (confirm live prices on the pricing page).
• Integrations with Stripe, Zendesk, Salesforce, Intercom, HubSpot, Shopify, Slack, Zapier, Twilio, and others.

Analytics and operating visibility

Chatbase provides conversation-level reporting, handoff tracking, and topic analytics. Advanced analytics appear on Pro and Enterprise. The operational value depends on whether your team can act on those reports. We recommend confirming whether the analytics surface clearly points to knowledge gaps, which channels drive the most volume, and which questions repeatedly fail.

Security, data handling, and compliance

Chatbase publishes SOC 2 Type II certification and GDPR compliance on its enterprise and security pages. Before connecting customer conversations, ask about data retention, model-training use, role-based permissions, and audit trails. Enterprise adds SSO, audit logs, custom security controls, and HIPAA eligibility.

• Request the latest security documentation and data-retention policy in writing.
• Confirm whether uploaded knowledge sources or customer conversations are used to train shared models.
• Map role-based permissions to your actual staff: agents, supervisors, admins, developers.
• Verify audit trails for actions such as refunds, account changes, or escalations.

What questions should you ask before buying Chatbase?

• Can you demo the agent using our website, documents, and real customer questions?
• Which channels are included in our target plan, and which require separate WhatsApp API or add-on fees?
• How many message credits will our expected monthly volume consume?
• What actions can the agent perform without writing custom code?
• How does a human agent receive context after a low-confidence or sensitive escalation?
• Can we export conversation data and agent performance metrics?
• How do auto-recharge credits work, what is the per-1,000 price, and what happens when the plan limit is reached?
• What is the process for updating knowledge when products or policies change, and who owns the weekly accuracy review?
• How do we configure source-only answering, confidence thresholds, and escalation so the agent does not hallucinate policies?
• What is the live price to remove "Powered By Chatbase" and to add a second agent for a second brand or language?
• Can we export unknown-query logs and AI-handled CSAT so we measure accuracy, not just deflection?

What red flags should you watch for with Chatbase?

• The vendor cannot demonstrate the agent answering from your actual website or documents.
• WhatsApp, voice, or advanced integrations are described as available but require higher tiers or separate billing.
• There is no visible approval gate or audit trail before the agent performs sensitive actions.
• The knowledge-base import fails on PDFs, tables, or multi-language content that your business relies on.
• Pricing conversations focus only on the subscription and omit credit burn and add-on fees (branding removal, extra agents, auto-recharge).
• The demo uses a pre-canned conversation that does not match your real customer inquiries.
• The vendor cannot show source-grounded answers, a forced "I don't know" path, and a low-confidence handoff on your own knowledge gaps.
• Knowledge is a dump of long PDFs with no owner for weekly updates, so stale answers are guaranteed after the first product change.

What are the best alternatives to Chatbase?

The right alternative depends on where your team already works and how deep the AI workflow needs to be. Below is a decision framework, not a ranked list.

• Intercom FinChoose Intercom Fin when your team already lives in Intercom and you want AI tightly coupled to your existing inbox and helpdesk articles.
• YourGPT AIChoose YourGPT AI when you need omnichannel support and sales automation with multi-step workflows across many channels.
• TidioChoose Tidio for a lighter live-chat and chatbot experience when budget and speed matter more than enterprise workflow depth.
• SiteGPTChoose SiteGPT for another website-trained chatbot option and compare fit against your content sources and deployment needs.
• GorgiasChoose Gorgias if your priority is Shopify-centric ecommerce support with macros, revenue-aware reporting, and order context.

Workflow test

What Chatbase needs to prove in a real support workflow

A polished chatbot demo is not enough. The buying question is whether Chatbase can answer from your real content and hand off cleanly when it should. Run this four-step test in a sandbox before committing.

1. Ground the answer.Use your help-center content and product pages; ask the agent to show its source. If it cannot explain where an answer came from, it will drift under pressure.
2. Take the action.Run an order lookup, scheduling request, or lead capture in a sandbox rather than stopping at a generated reply. The action should succeed or clearly fail with a reason.
3. Apply the guardrail.Force a low-confidence, billing, or refund case and confirm the escalation trigger and audit trail.
4. Hand off with context.Check that the human receives the transcript, customer details, and reason for escalation without making the customer repeat the story.

• Chatbase vs YourGPT for website AI agents
• Chatbase vs Tidio for small business chatbots
• Intercom vs Chatbase for customer support
• Best AI chatbots for business websites
• Customer support AI agents buyer guide
• Ecommerce AI agents for order and return support
• Visit Chatbase - Official product page.

Official sources to verify

Official Chatbase product page Source snapshot 2026-07-29 - chatbase.co Positioning, channels, and product suite checked against the vendor's own product surface. Official Chatbase pricing page Source snapshot 2026-07-29 - chatbase.co

Product walkthrough

See the Chatbase platform before you trust the claims

This official Chatbase walkthrough shows how to connect a Shopify store and add AI actions. Treat it as a product tour, not independent proof of resolution rates or implementation effort. After watching, ask the vendor to repeat the workflow with your own products, policies, and failure cases.

Should you choose Chatbase?

Chatbase is a practical choice for teams that want a website-first AI agent trained on their own content. It offers a free tier, affordable paid plans, and enough channel coverage to handle chat, messaging, email, and voice on higher tiers. It is not a replacement for a full helpdesk or enterprise CRM.

Best for small to midsize teams that need fast setup, branded chat support, and light actions. The main risks are message-credit limits, plan-specific feature gates, and the need for ongoing answer tuning. Verify your volume, channel mix, and required actions in a free trial or sandbox before committing.

Before you scale, run two non-negotiable checks. First, a hallucination and accuracy pilot on 50 real questions with source-only rules, forced unknowns, and handoff quality scoring. Second, a full commercial model that includes plan credits plus auto-recharge, extra agents, and branding removal if you need a clean widget. Chatbase is worth it when both checks pass. It is a poor fit when you need deep helpdesk governance more than a website-first agent, or when nobody will own the knowledge loop after launch.