New to Gladly (AI)? Start Here

Gladly AI helps your customers get fast, accurate answers automatically, and gives your team a clear, configurable way to shape how it sounds, what it can access, and when it should hand off to a person. This overview walks through everything from your first look at the interface to monitoring performance once you're live, in the order most teams tackle it.

How Gladly AI Works

Use the clickable graphic below to get familiar with the components of Gladly AI and how they work together:

Start with a Tour of Gladly AI

Use the clickable graphic below to walkthrough every aspect of Gladly AI:

Locate Gladly AI Sections in Gladly
To find these sections, open the collapsible main menu and select Gladly AI. You'll only see this section if your organization uses Gladly AI. The menu is organized into three groups: Analyze (AI Sessions, Reviews, Signals, Journeys), Manage (AI Agents, Workflows, Answers), and Deploy (Changelog, Entry Points). You'll also see Test (Simulations), which this article doesn't cover in depth.

AI Agents

Select AI Agents under Manage to see every Gladly AI Agent your organization has set up. Think of this page as mission control: it's where you pick which Agent to configure, and where you go to create a new one.

An agent is a distinct configuration of how Gladly should behave for a given use case. Most organizations start with one Agent, but you may want more than one. Create a new agent any time you need a unique set of behavioral guidance, a unique set of Guides, or both, for example to give a second brand its own voice, or to run a different experience on a different Channel. Create one from the Agent dropdown at the top of any Agent’s page by selecting Create Gladly AI Agent.

Each Agent has its own version history, its own set of Guides, and its own five behavioral tiles, covered next.

Tiles

Every Gladly Agent has five behavioral tiles at the top of its page. These shape Gladly's overall personality and decision-making, no matter which Guide ends up handling a given Conversation. Each one covers a different layer of behavior, from identity and tone to escalation and automation, and together they define how the Agent acts before you even get to its Guides.

Overview of Gladly AI features including basics, communication, help, topics, and triggers.

Configure the Basics tile

Basics is where you define the foundational identity of the Agent: its name (if you want it to introduce itself by name), your company name, and a company description that helps Gladly understand your business.

Form fields for defining Gladly AI's name, company details, and greeting options.

This tile also controls how Gladly greets customers. Choose a dynamic greeting (Gladly generates a greeting from guidance you provide, like "mention the brand name and sound cheerful but not overly casual") or a static, exact greeting. You can also add a signature (exact text appended to the end of every response) on a per-channel basis, using Add Channel signature.

At the bottom of Basics you'll find Early Access: a section for behavioral experiments Gladly is rolling out. These features come and go as they graduate to default behavior. Today this includes Self-Healing Responses, which lets Gladly rewrite a response that fails a quality check instead of immediately handing the conversation to a human agent. Toggle these on if you want to test upcoming behavior early; check the release notes for what's currently available here.

Set How to Speak to Customers

How to Speak to Customers holds tone and voice guidance that applies across the whole Agent, regardless of which Guide is handling the conversation. Use this tile for advice that should be a through line in every response, not guidance specific to one workflow.

Guidelines for effective customer communication and brand voice strategies displayed in a table.

Each piece of advice is a simple statement, like "Always include links to products you recommend" or "Do NOT make more than 3 recommendations." You can scope any piece of advice to specific Guides or specific channels, or leave it applying everywhere. If your guidance is guide-specific, it belongs inside that Guide instead. This tile is for the general rules that should hold true no matter what.

Set When to Get More Help

When to Get More Help defines your handoff scenarios: the situations where Gladly should hand a conversation to a human agent, such as a customer repeatedly asking for a person, legal concerns, or anything sensitive or complex enough to need live support.

Guidance for transferring customer support sessions to an agent based on specific scenarios.

Here you can also configure a handoff message: the text Gladly sends the customer when it transfers the conversation, such as "One moment please." This is especially useful in Chat, where it smooths the transition before a human agent joins. Handoff data, also configured here, is used to pass conversation context to external systems and isn't customer-facing.

Set up Automatic Topics

Automatic Topics is how Gladly labels conversations with a Contact reason even when no Guide explicitly assigns one. Add a topic with a name and description (for example, "Return process" or "Delivery fees") and Gladly will apply it automatically when relevant.

Overview of Automatic Topics feature for customer inquiries and topic management.

One rule of thumb: if a topic is already covered by a Guide, don't duplicate it here. For example, if you already have a Track Order Guide, you don't need an automatic topic for "track order" inquiries, since the Guide already handles that classification. Use Automatic Topics for contact reasons that don't have a dedicated Guide.

Connect Workflow Triggers

Workflow Triggers is where you connect Workflows (covered later in this article) to the rest of your Agent. It controls the order automations run in, based on session events, conditions, and outcomes. Come back to this tile once you've built a Workflow: it's the bridge between your Guides and your more deterministic, rules-based automations.

Workflow triggers for customer interactions, including messages and handoff processes.

Guides

Below the five tiles, you'll find your Agent’s Guides, organized into Active, Inactive, and Draft tabs. A Guide is a workflow, written largely in plain language, that tells Gladly how to handle a specific kind of customer interaction.

There are two broad kinds of Guides:

Questions and Recommendations Guides answer questions using information you already have: Public Answers, your website, or approved documents. They don't need access to any external system.
Action Guides handle tasks that require access to an external source of truth, like your order management system or a returns platform. Because they need real integrations, action guides can respond to the customer, take action in a third-party app (like Shopify), or take action in Gladly on the customer's behalf.

Guide for customer inquiries, including damaged items and Shopify tracking orders.

You can build from scratch with a Blank Guide, or start from a template:

Questions and Recommendations
Shopify Track Order
Shopify Cancel Order
Loop Returns Return Status
Shopify + Loop Returns Return Order

Gladly has built out a number of additional templates, mainly for organizations on Shopify; if you'd like to get started with one instead of building from scratch, reach out to your Gladly team to learn what's available for your organization.

Inside a Guide, you'll build one or more named sections, each with its own guidance (bullet-point instructions) and, optionally, a Workflow or an Action it can call, like looking up an order or checking product availability. Once you're happy with a Guide, mark it Active from the draft tab so it's part of your live Agent; draft Guides stay in Draft until you're ready.

Set up the Questions and Recommendations Guide

Every Agent comes with a Questions and Recommendations Guide out of the box, marked Fallback. This is the catch-all: any inquiry that isn't caught by a more specific Guide runs through Questions and Recommendations before handing off to a human, so Gladly always attempts to help before escalating. Because it's the fastest path to going live, get this Guide right first.

By default, this Guide's sections come with standard guidance already in place. You generally don't need to touch it unless you have a specific use case. What you do need to manage are its knowledge sources.

Overview of questions and recommendations for product policy and pricing guidance.

Manage knowledge sources

Each section that answers questions has a Manage button next to its Action, which opens the Knowledge Sources modal with three tabs:

Answers. Turn on Public Answers so Gladly can respond using them. Choose a language, and optionally restrict Gladly to a specific Audience. By default, Audience is set to None, meaning Gladly can draw from any Public Answer. If you want Gladly to use content that shouldn't be public-facing, create a dedicated Audience, something like "Gladly AI Answers," and add that content to it, then select it here. You can add more than one Audience.
Websites. Select from your available website URLs so Gladly can reference product pages and other content you don't want to duplicate as Public Answers. You can include or exclude specific URL paths to fine-tune what it references.
Documents. Gladly can also be set up to search a set of documents. This capability is still rolling out. Reach out to the Gladly team if you're interested in using it.

Build Action Guides for order and account tasks

Once your Questions and Recommendations Guide is solid, you can add Action Guides for anything that needs a live system, like checking an order's shipping status, canceling an order, or processing a return. Think of an Action Guide as the SOP you'd hand a human agent: it lays out, section by section, what to check, what action to call, and what to tell the customer depending on the result.

Each section's Action dropdown lets you search and attach the specific integration a step needs, such as a "Get Order" lookup or a Shopify action. A Guide can mix multiple sections with different actions, and can route to a Workflow instead of an Action where a more structured, deterministic flow makes more sense.

Instructions for tracking Shopify orders using customer email inquiries and responses.

Workflows

Workflows under Manage are for cases where you want a more deterministic, rules-based flow rather than open-ended guidance. Spam handling is the most common example. Start from one of Gladly's templates for the most frequent use cases, or build one from scratch.

Once a Workflow is built, connect it to your agent from the Workflow Triggers tile covered earlier. That's what makes it part of the agent's live decision-making rather than a standalone flow.

Customer service chat discussing a damaged item and requesting image evidence.

Versions and Entry Points

Every Agent shows a banner at the top of its page with its publish status, for example, "Not published. Published on 0 entry points. Connect one to start handling sessions." Nothing you build goes live until it's connected to an Entry Point.

Status indicator showing published entry point with management option available.

Select Connect entry points (or Entry Points under Deploy) to route real traffic to your agent. Each entry point (a phone number, chat domain, or similar) is mapped to one or more Agent versions, with traffic split between them if you want to A/B test. You can pin an entry point to a specific version or set it to Always latest, so it automatically picks up your newest published changes.

Entry points for routing traffic to different agent versions in Gladly platform.

Select Versions on an Agent to see its full history: every version, what changed, a plain-language summary, who made the change, and whether it's published. You can preview any past version or restore it if you need to roll back.

Document showing version history and changes made to Retalè Home settings.

AI Sessions

Once your Agent is live, AI Sessions under Analyze is where you review real interactions. Note the terminology: when Gladly is engaged with a customer, that's an AI session, distinct from the underlying Conversation and its timeline in Gladly Team. The Sessions page has two tabs: Gladly AI Sessions, for real customer interactions, and Test Sessions, for anything run through the Tester.

Overview of customer support sessions with messages and responses displayed in a table format.

Search and filter by status, outcome, channel, experiment, Guide, Answer, entry point, or Agent version to find the sessions you care about. Each row shows the customer, their first message, Gladly's first response, and the outcome: in progress, resolved by Gladly, or handed off.

Inspect Summary

Open any session to see the full conversation alongside an Inspect Summary panel. The summary shows the topic Gladly assigned, which Agent and Guide version handled it, and which section and knowledge sources it used.

Summary of a customer service conversation regarding order tracking and email requests.

Below that, you'll find a step-by-step trace of exactly how Gladly built its response: detection steps, the guide and section it navigated to, the action it ran, and the response it generated. This is followed by its Quality Checks, the guardrails that run behind the scenes on every response before it's sent, covering things like hallucination detection, invalid URLs, and whether the response properly directs the customer to self-service. Each check shows a pass/fail status and a rationale, so you can see exactly why Gladly made the decisions it did.

Reviews Log

From that same detail view, you can thumbs up or thumbs down any response and add a note on why. Every review you leave is logged on the Reviews page under Analyze, which acts as a searchable, filterable record of feedback, by session or by individual component, that you can share with teammates via a direct link, or use to jump straight back into the session.

Table displaying reviews with ratings, notes, and dates for various sessions.

Journeys

Journeys under Analyze is your day-to-day check-in on how Gladly is performing. At the top, you'll see customer inquiries, resolution rate, handoff rate, and a link into Answers Performance. Below that, a table breaks performance down by Guide, so you can see count, handoffs, resolutions, and resolution rate for each one. Filter by date range, Guide, or channel to narrow the view.

Gladly dashboard showing customer inquiries, resolutions, and agent performance metrics.

Answers Performance shows which specific Public Answers Gladly referenced when generating responses, along with the same count, handoff, and resolution metrics for each. Use Open Conversation Report from either view to dig into the underlying reporting in more detail.

Signals

Signals under Analyze let you measure the quality of your AI sessions using AI-graded criteria at scale, rather than reviewing every conversation by hand. Gladly runs some signals on your behalf automatically. You'll see these under Run by Gladly. You can also build your own under Your Signals, active or archived.

Each signal evaluates a sample of sessions against a specific question, for example whether a customer asked about product availability, overall customer sentiment, how cleanly a handoff happened, or how a return for a damaged item was handled. Every signal card shows the channels it covers, its top result, and how many sessions it's evaluated. Use New Signal to build one from scratch, or start from the Criteria Library of common criteria. Treat Signals as your recurring quality check-in, complementing the gut-check you get from Journeys.

Overview of customer sentiment metrics, showing positive interaction sentiment and session details.

Changelog

Changelog under Deploy is a chronological record of every change made across your Gladly AI setup: Agents, Guides, Workflows, and more. Each entry shows the date, what was changed (tagged by category, like Guide, Basics, or Workflow settings), a plain-language summary of the change, and who made it. Filter by date range, type, name, what changed, or who made the change to track down a specific update, or to audit what's changed recently across your whole setup.

Changelog displaying updates made to Gladly with details on changes and authors.

Reports

Journeys and Signals cover your day-to-day monitoring, but Gladly also has a broader library of out-of-the-box reports for deeper analysis, including session-level detail reports, guide and answer performance dashboards, and conversation time tracking. Explore these from the Reports section of Gladly Team when you're ready to go beyond the built-in Gladly AI views.