WeChat's Mini AI: Say a Sentence, Skip the List, Land Inside Your Mini Program
Getting to Know WeChat's Mini AI: One Sentence Can Directly Invoke Your Mini Program's WeChat AI, and Most People Misunderstand It at First Glance
Series: "WeChat Mini AI (Mini Program AI) in Practice" Part 1
⚠️ Version note: This article is based on the internal testing version of WeChat Mini AI (official docs and team testing as of 2026-11-18). WeChat has not released an official version. Capabilities, entry points, and rules may change as the version iterates. After the official version is released, refer to the official documentation.
Foreword
Recently our team integrated WeChat's AI capability (Mini AI) into our mini program, going through application, development, evaluation, submission for review, and gradual rollout. During the process we found that online, there are many different claims about "Mini AI," and many colleagues even have misunderstandings about "what Mini AI actually is."
This is the first article in the series, focusing on the most basic and important questions:
- What is Mini AI?
- Where does a user encounter it?
- After a user says a sentence, what happens behind the scenes?
- Two common cognitive misconceptions.
Later articles will cover integration modes, engineering implementation, writing a real skill, evaluation and submission, and gray rollout troubleshooting, but everything builds on the mental model established here.
1. What is Mini AI?
Mini AI is WeChat's built-in AI assistant, not a mini program list, not an app store, and not a third-party robot.
One-sentence positioning: A user in WeChat expresses a need in natural language, and Mini AI determines whose capability to invoke to satisfy that need.
For developers, it is a new form of traffic entry: users no longer go from "search → browse list → tap into your mini program," but instead "say a sentence → AI judges → directly enter one of your capabilities."
2. Where does a user encounter Mini AI?
Mini AI has many invocation entry points. Officially, about 16, while third-party testing shows over 28 (internal testing version data; the official version will follow official announcements). Typical ones include:
Mobile:
- Top left of the message list (green robot / "small eye" icon)
- WeChat search
- Official account article pages, video channels
- Inside a mini program / top-right menu invocation
PC:
- Homepage sidebar quick entry
- Top-left independent icon (opens what many call the "left popup window")
- Chat session right-click "Ask AI"
- Built-in browser "Ask AI," file preview page
Key insight: these entry points all lead to the same Mini AI assistant. No matter where the user enters from, they are chatting with the same AI and see the same capability list.
Another important premise: Mini AI is also being rolled out gradually on the user side. Not all WeChat users already see these entries; entry coverage and capability range differ between mobile and PC (currently PC has fewer entries and less capability coverage).
3. What happens after a user says a sentence?
Use a real case that our project ran through (anonymized). A user says to Mini AI:
"Help me search for XX Pioneer"
The complete pipeline is:
① User says to Mini AI: "Help me search for XX Pioneer"
↓
② Mini AI determines: this is a "find a specific fund" need
Which mini program's capability should be invoked?
↓
③ The platform finds your mini program in the "invokable list"
(precondition: your capability is live and has been rolled out to the current user)
↓
④ Invokes the atomic interface you registered, searchProduct
↓
⑤ The conversation flow presents a sentence + a "account card"
"Found XX Pioneer for you. Tap the card below for details."
↓
⑥ User taps the card → enters your mini program's "relay page"
(lands directly on the product detail page, not the homepage)
Note the user's perspective: "I said a sentence in Mini AI, and then I entered some page in your mini program."
There is no search box, no result list, no "open mini program" action. The user is sent by AI by intent to a specific page in your mini program.
Here are two key concepts that later articles will expand on:
- Atomic interface: the smallest capability unit you expose to the AI (for example, "search funds by keyword"), which the AI can invoke directly.
- Relay page: the page the user lands on after tapping the card, which receives the user sent by the AI and completes the flow.
4. Two most common misconceptions
Almost every new colleague on our team hits these two. They must be cleared up first.
Misconception 1: "The PC left popup window = a left-side mini program"
Many people have tapped the AI window on the left side of PC WeChat and thought that "WeChat has put mini programs in the sidebar."
No. That is Mini AI's AI conversation window, not a list of mini programs. Your mini program will not appear there, whether or not you have integrated AI capabilities. What you can do is have Mini AI recommend your capability in the conversation at appropriate moments.
Misconception 2: "Mini AI will open my mini program"
Before integration, the most common expectation is "When a user says 'open XX mini program' to Mini AI, it will open my mini program."
The real pipeline is not like that. Mini AI does not look for a mini program by name or keyword. Instead, it:
- Understands the user's intent (what they want to do)
- Among all integrated capabilities, determines which one can satisfy that intent
- Invokes the atomic interface corresponding to that capability
- Produces a card → user taps the card → enters the relay page
Two corollaries:
- The name does not participate in matching. If a user types your mini program's name, Mini AI will not "find you" because of it. Matching is based on intent and your capability description, not the name. (Does that sound a bit like SEO? Half right. Article 10 will explain.)
- The landing is a specific page, not the homepage. When a user taps the card, they go directly to the product detail or target page. Your mini program is essentially "broken into parts" and invoked on demand by the AI.
An extended misconception: "If I develop AI code, Mini AI can use it, right?"
No. Code is just step 0. The complete pipeline is:
Apply for access → Develop → Evaluate → Submit for review → Release → Gray rollout → Only then can it be invoked
In our testing, we encountered: all code complete, version released, but on a colleague's phone it never hit. It turned out to be a gray rollout and client version issue. This process and troubleshooting approach is the topic of Part 9.
5. Application scenarios for Mini AI
What kind of mini program capabilities are suitable? From our practice and official examples, the typical scenario is "a single point task that can be described in one sentence":
| Scenario | User says | After integration |
|---|---|---|
| Product/content search | "Help me search for XX fund" | Search, show card directly → tap into detail page |
| Service direct access | "I want to check my orders" | Go directly to order center/order page |
| Content/information | "See today's market views" | Guide to information page |
| E-commerce/ordering (official example) | "One iced Americano" | Show spec selection card → complete order |
| Query type | "Check my orders" | Show results directly in the conversation |
Not suitable: flows requiring complex multi-step operations, strong graphical interaction, or intents that cannot be described in one sentence. (Some can be split into multiple single point capabilities and combined.)
6. Summary
- Mini AI is WeChat's built-in AI assistant, behind 16+ entry points is the same assistant, and both user and developer sides are still in gray rollout (internal testing).
- The pipeline is "invocation by intent": speak → AI judges → invoke atomic interface → produce card → tap card to enter relay page.
- Two misconceptions: the left popup is not a mini program list; Mini AI does not "open your mini program" but "invokes your capability."
- Code is only step 0: apply → evaluate → submit for review → release → gray rollout, not one step less.
- Name doesn't matter; intent matching is what matters.
Next article covers integration modes: the difference between automatic mode and AI development mode, how to enable it, and the version thresholds you need to confirm in advance or you'll get stuck.
This article is compiled from practical experience (based on internal testing version). Official capabilities and processes are subject to the WeChat Open Documentation and the official platform backend.
Top 2 of 5 from juejin.cn, machine-translated. The original thread is authoritative.
Apply for activation → develop → evaluate → submit for review → release. How long does this usually take before it takes effect? That is, how long before it might appear in search on XiaoWei?
It's now in internal beta; the public version should be out soon. You can integrate the feature first, so you can be among the first batch to officially launch.
No, I mean even if it's internal testing and the mini program has been gray-released, do we have to wait a few days before it can be found on XiaoWei? Do you know how long those few days typically are?
As long as your XiaoWei feature is up, add the WeChat team group, submit for review and pass, enter gray testing, and it can be called immediately in the AI test version.