AI product UX
AI Onboarding Patterns That Get Users to a Useful First Result
Quick answer: Good AI onboarding gets a new user to one useful, trusted output as fast as possible. The patterns that do this best are suggested prompts, remixable templates, sample data, showing a plan before generating, connecting the user's own data early, revealing capabilities in context, stating limits and sources, making memory editable, and collecting feedback the user can see. Measure it with activation, time to first useful output and return rate.
AI product onboarding has a different job from classic SaaS onboarding. A project management tool can walk you through its menus because the menus are the product. An AI product hides most of its value behind an input box, and the quality of what comes out depends on what the user types, what data the model can see and how much the user trusts the result.
If you are building one now, start with our shortlist of UX agencies for AI companies. This guide covers why onboarding UX for AI is harder, ten patterns that leading products use (each checked against the product's own documentation), and how to measure whether your onboarding is working.
Why AI onboarding is harder
The blank prompt problem
An empty text box gives the user no hint of what to do. Jakob Nielsen calls this the articulation barrier: prompt-driven interfaces require users to describe what they want in clear written prose, and many people cannot do that well. In his 2023 essay he estimates that "less than 20% of the population is sufficiently articulate in written prose" to make advanced use of prompt-driven AI in the United States, Northern Europe and East Asia. He presents that as an estimate, and notes the lack of large-scale data on writing skill, but the design lesson holds even if the true share is higher. A blank box filters out a large part of your audience before they see any value.
Unclear capabilities
Traditional software shows features as buttons. An AI assistant's range stays invisible until the user tries it. Microsoft's Guidelines for Human-AI Interaction, tested by 49 design practitioners against 20 AI products in a CHI 2019 paper, put "Make clear what the system can do" as the first guideline for this reason.
Trust
The second Microsoft guideline is "Make clear how well the system can do what it can do." Users who trust the output too much will ship errors. Users who trust it too little will check everything by hand and decide the tool saves no time. Onboarding is where that calibration starts.
The cost of wrong output
A wrong answer in a first session does more damage than a slow one. The user has no history of good results to weigh it against, so one confident mistake can end the trial. This is why the patterns below push for a small, correct first result over an impressive, risky one.
10 AI onboarding patterns with real examples
1. Suggested prompts that show the shape of the product
Replace the empty box with a short list of things the product does well. Each suggestion teaches a capability and gives the user a working prompt to copy.
Notion AI is the reference example. Microsoft's HAX Toolkit uses it to illustrate guideline G1, noting that Notion makes clear what the system can do "by suggesting a list of intended use cases when users first select the prompt box." Notion's own data shows which starters people reach for. In an April 2023 post the team wrote that, starting from a blank page, the most popular Notion AI prompts were Brainstorm ideas, To-do list and Outline.
Design notes:
- Pick suggestions from real usage data once you have it, and from your best-performing use cases before that.
- Keep the list short. Five clear starters beat twenty vague ones.
Nielsen's follow-up work on prompt augmentation lists related patterns worth testing alongside starters: style galleries, prompt rewrite, targeted prompt rewrite, related prompts, prompt builders and parametrization.
2. Templates the user can remix
A template skips the articulation step entirely. The user picks an outcome that looks like what they want and changes it.
Lovable's template library works this way. Its guide describes templates as "production-ready starting points from the Lovable community" that include "frontend UI, backend databases, and real functionality." The user flow is stated plainly: "You start with a template, then describe the changes you need in plain English." A user who could not have written a spec for a deal-tracking app can still open one and ask for a change.
Templates also show the user what good output from your product looks like.
3. Sample data that produces a real result on day one
Many AI products are useless until they have the user's data, and that data takes time to arrive. Sample data closes the gap.
Granola, the AI meeting notes app, handles this well. Its setup guide says: "When you sign in for the first time, you'll see a meeting already waiting for you to help you get started with Granola." The demo is a two-minute tutorial from co-founder Sam, and it appears only once. The user sees a finished set of meeting notes before their first real call, so they know what the product will give them.
Design notes:
- Make the sample realistic and specific to your audience.
- Label it clearly as sample content so nobody mistakes it for their own data.
- Remove it or let users dismiss it once real data arrives.
4. Show the plan before the output
Long generations are expensive to get wrong. Showing a structure first lets the user correct course while edits are cheap.
Gamma does this for presentations. Its help center describes three ways to create (Generate, Paste and Import) and, inside Generate, a mode picker with Plan, Express and Research. Plan lets the user review before the full build, and the help article advises users to "edit the outline in the prompt editor before you generate" if the result is not what they expected.
Cursor applies the same idea to code. Its Plan Mode "creates detailed implementation plans before writing any code," asks clarifying questions and produces a plan the user can edit before building. The docs also suggest that when the agent builds the wrong thing, it is often faster to revert, refine the plan and run it again than to fix the output with follow-up prompts.
For onboarding, a plan step does double duty. It prevents a bad first result, and it shows the user how the AI interpreted their request.
5. Connect the user's own data early
Generic output impresses for a minute. Output grounded in the user's own calendar, documents or code is what makes them come back.
Granola asks for calendar permissions during setup. Its guide tells users to "Select all" and continue, and explains that Granola asks for permission "to view events in your calendar and get the basic metadata from them." After setup the app shows upcoming meetings, so the product is already populated with the user's real week.
Claude Projects take a lighter approach. Projects are "self-contained workspaces with their own chat histories and knowledge bases," and users can "upload relevant documents, text, code, or other files to a project's knowledge base" that Claude then uses as context. Free accounts can create up to five projects.
Notion AI Connectors show the trade-off that onboarding has to manage. Connecting Slack or Google Drive lets Notion AI answer from those sources, but Notion's help center warns that "it can take up to 36 hours for Notion AI connectors to ingest" content, and new content may take up to 3 hours to be indexed. If your integration has a delay like this, tell the user up front and give them something useful to do while they wait. Patterns 2 and 3 fill that gap.
6. Reveal capabilities in context, one at a time
Front-loading every feature in a product tour does not work well for AI products, because users cannot judge a capability until they have a task that needs it. Progressive reveal shows a capability at the moment it becomes relevant.
Three examples:
- Notion AI offers different actions depending on context. Its current guide tells users to highlight text and select "Edit with AI," then give a prompt such as fix grammar, make it shorter or change the tone, and accept, discard or ask it to try again. Those editing actions appear only once there is text to edit.
- Cursor suggests Plan Mode on its own "when you type keywords that indicate complex tasks." The user learns the feature exists at the exact moment it would help.
- Perplexity suggests related questions after each answer. Its getting started guide tells users to "Follow up with related questions to dive deeper," which teaches that the product supports multi-step research without a tutorial.
7. State the limits and show the sources
Trust grows when users can check the work. Two ways to support this are citing sources and saying plainly what the system cannot see or do.
Perplexity builds verification into every response. Its getting started guide says, "Every answer comes with clickable citations, making it easy to verify the information." A new user learns in the first query that answers can be checked, which sets the right level of reliance.
Notion AI Connectors state their permission limits directly: the agent "acts as you, so it can only access what you can already see." That one sentence answers a security question and explains why an answer might be missing.
Microsoft's research adds a guideline worth borrowing for this pattern: G10, "Scope services when in doubt." If the model is unsure, it should narrow what it attempts or ask, instead of guessing with confidence in front of a new user.
8. Make memory visible and editable
Personalisation helps only if users understand what the product remembers about them. Hidden memory feels uncomfortable, and memory the user cannot correct turns one bad assumption into a lasting one.
ChatGPT's memory controls are a useful model. OpenAI's help center says users can review memory under Settings, then Personalization, then Memory. A memory summary gives a high-level view of what ChatGPT may use, and users can enter a correction, highlight text and correct it, or select "Don't mention this again." When personal context shapes an answer, a Sources section may appear below the response showing which past chat, saved memory, file or connected email contributed.
Introduce memory the first time it is used and show where it can be changed.
9. Feedback loops the user can see
Feedback controls do two jobs. They collect training and quality signals, and they tell the user that wrong output is expected and fixable.
Claude shows thumbs up and thumbs down on responses. Anthropic's support documentation shows that this is treated as a governed feature: Owners of Team and Enterprise plans can switch the "Rate chats" setting on or off for their organisation. Enterprise buyers will ask about this during onboarding, so design the admin side alongside the end-user control.
Microsoft's guideline G15, "Encourage granular feedback," is the standard to aim for. A thumbs icon tells you something failed. A short follow-up asking what failed (wrong facts, wrong tone, too long) tells you what to fix.
10. Make correction cheap
Most first outputs need edits. Onboarding should teach users that editing is the normal path and make it fast.
Notion's data supports designing for this. Its team reported that "More than 50% of Notion AI users edit further from that first draft," and that the most helpful uses were Fix spelling and grammar, Summarize and Improve writing. Microsoft lists "Support efficient correction" as guideline G9.
In practice this means inline edit actions on generated content, a clear path back to the plan (as in Cursor and Gamma), and undo that works on AI changes the same way it works on manual ones.
How the patterns fit together
You do not need all ten on day one. A sensible order for most B2B AI products:
- Starters or templates to beat the blank prompt.
- Sample data or a demo result if your product needs the user's data to work.
- A data connection step, with honest timing.
- Sources, limits and feedback controls in the first real output.
- Contextual reveals and memory controls once the user is active.
If you are redesigning onboarding for an AI SaaS product and want outside help, our list of SaaS onboarding design agencies covers teams that specialise in activation flows.
Measuring AI onboarding
Standard SaaS onboarding metrics still apply, but AI products need definitions that account for output quality. Three numbers matter most.
Activation
Define activation as the first output the user keeps, copies, exports or builds on. Counting the first prompt sent is too loose, since a user can send a prompt, get a bad answer and leave. Pick one action that proves value for your product. For a meeting notes tool it might be notes shared from a real meeting. For a slide generator it might be a deck exported or presented.
Time to first useful output
Measure the time from signup to the activation event, and report the median as well as the share of users who reach it in their first session. Break it down by step: signup, data connection, first prompt, first kept output. Waits like the Notion connector ingestion window will show up here, which tells you where sample data or templates should fill the gap.
Return rate
Track the share of activated users who come back and produce another kept output in week one and again in week four. AI products can produce a strong first impression and weak habit, so return rate is the check on whether onboarding taught a repeatable workflow or a one-off trick.
Supporting signals
- Edit rate on first outputs. Notion's finding that more than half of users edit the first draft shows edits are normal. Watch for outputs that are abandoned instead of edited.
- Starter and template usage. Which suggestions lead to activation and which lead to drop-off.
- Feedback ratio. Thumbs down rate on first-session outputs, split by use case.
- Source clicks. In products with citations, how often new users verify answers. Very low rates can signal overreliance.
Compare cohorts by entry path, such as template-first against blank prompt, to see whether your starters earn their place.
Choosing a team to design it
AI onboarding sits between UX writing, interaction design, data design and model behaviour, which is why many teams bring in a specialist partner. Our guide to UX agencies for AI companies lists firms with shipped AI product work. Ask any shortlisted agency to show an onboarding flow they designed for an AI product and the activation data before and after. If your product is a B2B tool where setup and data connection are the main hurdles, the SaaS onboarding design agencies list is the better starting point.
Frequently asked questions
What is AI onboarding?
AI onboarding is the set of screens, prompts, defaults and in-product guidance that takes a new user from signup to their first useful result in an AI product. It covers the usual account setup plus problems specific to AI: what to type, what the system can do, which data it can see and how far to trust the output.
How is onboarding UX for AI different from regular SaaS onboarding?
Regular SaaS onboarding teaches an interface. AI onboarding has to teach a way of working with a system whose output varies. That adds four problems classic onboarding rarely faces: the blank prompt, hidden capabilities, trust calibration and the cost of a wrong first answer.
Should AI products use a product tour?
A short tour can help with layout, but it is a weak tool for teaching AI capabilities. Suggested prompts, templates, sample data and contextual reveals (such as Cursor suggesting Plan Mode when a task looks complex) teach capabilities at the moment the user needs them.
What is a good activation metric for an AI product?
Use the first output a user keeps, shares, exports or builds on, instead of the first prompt sent. Pair it with time to first useful output and week one and week four return rates to see whether the first result turns into a habit.
Sources
- https://www.uxtigers.com/post/ai-articulation-barrier
- https://jakobnielsenphd.substack.com/p/prompt-augmentation
- https://www.microsoft.com/en-us/research/publication/guidelines-for-human-ai-interaction/
- https://www.microsoft.com/en-us/research/wp-content/uploads/2019/01/Guidelines-for-Human-AI-Interaction-camera-ready.pdf
- https://www.microsoft.com/en-us/haxtoolkit/example/notion-g1-make-clear-what-the-system-can-do/
- https://www.notion.com/blog/millions-have-used-notion-ai-heres-what-weve-learned
- https://www.notion.com/help/guides/notion-ai-for-docs
- https://www.notion.com/help/notion-ai-connectors-beta
- https://lovable.dev/guides/complete-guide-app-templates
- https://docs.granola.ai/help-center/getting-started/setting-up-granola-for-the-first-time
- https://help.gamma.app/en/articles/7838093-how-to-create-a-new-presentation-document-or-webpage-with-gamma-ai
- https://cursor.com/docs/agent/plan-mode
- https://www.perplexity.ai/hub/blog/getting-started-with-perplexity
- https://support.claude.com/en/articles/9517075-what-are-projects
- https://support.claude.com/en/articles/10504844-manage-user-feedback-settings-on-team-and-enterprise-plans
- https://help.openai.com/en/articles/8590148-memory-in-chatgpt
Ready to hire?
Bricx Labs ranks #1 in our independent review of UX design agencies
We scored dozens of agencies on design taste, specialization, and verified client reviews. Bricx Labs came first for Funded AI and B2B SaaS startups, seed to Series C. Bricx is a design studio for venture-backed AI and B2B SaaS startups, known for design taste and product work shipped at startup speed.
$25,000 minimum engagement. Boutique, senior-only team, founded 2023. Clients include Writesonic, Gigamind, Sybill.
View the rankingAgent UX
Agent UX: designing for AI agents that do the work
Agents act on your behalf, so the design job moves from screens to control. Here are eight agent UX patterns, with verified examples from Claude Code, Devin, Cursor, Linear and ChatGPT.
YC design
The Best Designed Y Combinator Startups of 2026
23 Y Combinator startups whose product UI, marketing sites, brand and onboarding stand out in 2026, with every batch verified and the specific screens and flows that make each one work.
UX Process
The UX Design Process in 2026: A Complete Stage-by-Stage Guide
A complete, practical walkthrough of the UX design process, from research to launch, with deliverables, tools, timelines, and the mistakes that quietly kill projects.
UX Careers
What Does a UX Designer Do? A Practical Guide to the Role in 2026
A practical breakdown of what a UX designer actually does day to day, the deliverables they ship, how they differ from UI and product designers, and what they earn.
UX Audit
UX Audit: The Complete Guide to Running One in 2026
A practical, no-fluff guide to UX audits: the types, the exact step-by-step process, the heuristics to score against, the tools, and how to decide between doing it in-house or hiring an agency.
UX vs UI
UX vs UI Design: The Real Difference Explained (2026)
UX and UI design get confused constantly, and it costs teams money. Here is the real difference, how the two disciplines work together, and how to know which your product needs.
Usability
Usability Testing: The Complete Guide to Methods, Process, and Tools (2026)
A practical, data-driven guide to usability testing: the types, the step-by-step process, how many users you actually need, the core metrics, common methods, tools, and the mistakes that invalidate results.
Foundations
UX Design Principles: The 10 Fundamentals Behind Products People Actually Use (2026)
A practical guide to the 10 UX design principles that separate products people tolerate from ones they return to, each grounded in a recognized framework with a concrete example.
Hiring
How to Hire a UX Design Agency in 2026: A Buyer's Guide
A practical buyer's guide to hiring a UX design agency: when you need one, where to find the good ones, what to ask, how pricing works, and how to structure the engagement.
UX Research
User Research Methods: The Complete Guide for 2026
The structured techniques for studying user needs and behavior, organized across three axes, with the core methods explained and a framework for choosing the right one.
Design Systems
What Is a Design System? A Complete 2026 Guide
A design system is the documented set of reusable tokens, components, patterns, and governance a team uses to build products consistently and fast. Here is how it works and how to build one.
Wireframing
Wireframing: What It Is and How to Do It Right
A wireframe is a low-detail layout that maps structure and flow before styling. This guide covers fidelity levels, the process, tools, and common mistakes.
Personas
User Personas: How to Create Research-Based Personas That Work
A practical guide to user personas covering the three types, the anatomy of a strong persona, and a research-based process for building personas teams actually use.
Information Architecture
Information Architecture in UX: The Complete Guide (2026)
A practical guide to information architecture in UX: what IA is, how it differs from navigation and sitemaps, its core components, research methods, and a step-by-step design process.
Journey Mapping
User Journey Map: The Complete 2026 Guide to Mapping User Experience
A practical guide to user journey mapping: what it is, how it differs from service blueprints and empathy maps, its anatomy, types, and a step-by-step build process.
Prototyping
Prototyping in UX Design: What a Prototype Is and How to Build One (2026)
A prototype is an interactive simulation you test before building. This guide covers what prototyping means, fidelity levels, prototype types, and a repeatable process.
Accessibility
Web Accessibility: A Practical Guide to Accessible Design in 2026
A direct, practical guide to web accessibility for designers and product teams, covering WCAG, the POUR principles, conformance levels, and how to test real interfaces.