"Let's Get to the Next Failure Fast": A Growth Hacker's Playbook for the AI Era | A Conversation with Bolong Wang
No lingering on the past; forge ahead with abandon.
🚥 As the year draws to a close, Crossing is rolling out a series of retrospectives and reviews. In 2024, swept up by the AI wave, "growth" remained the core topic entrepreneurs cared about most, making it one of the important chapters in our year-end series.
"Growth is like an intellectual game — the pain feels real, and the thrill is just as palpable."
In the business world, everything changes yet nothing changes: build a great product, sell it well. In the internet context, "growth" is simply another way of saying "sell it well" — taking a product from 0 to 1, from niche to scale.
This week, Crossing's guest Bolong Wang is precisely a master of this growth game. From Meituan to Taobao, from Xiaohongshu to his own startup "Huidu" (Readflow), he's repeatedly turned products into hits through his distinctive growth mindset.
In this episode, we dive deep with Bolong into the practical craft of growth hacking, focusing on two dimensions: how to do growth for AI products, and how to leverage AI to fuel product growth.
More importantly, we explore the growth mindset that transcends methodology — a resilient attitude that spans time and cycles, an energy that anyone can feel.
Whether you're a product person, founder, or newcomer curious about growth, we believe this episode will offer you something valuable.
Listen on WeChat:
Listen on Xiaoyuzhou:
Huidu's Three-Stage Growth Journey: From User Virality to KOL Distribution
🚥 Koji
This week's guest on Crossing is someone we know as a growth expert. He previously led growth at Meituan, Taobao, and Xiaohongshu, and his own startup "Huidu" also became a phenomenon that year through growth thinking.
Today we'll chat with him about the common traits, methodologies, and — possibly more important than any of these — the mindset of top-tier growth hackers in his eyes.
👦🏻 Bolong Wang
Hi everyone, I'm Bolong Wang. I was born in the Year of the Dragon, so this is my benming year. I previously worked on e-commerce growth at Meituan, Taobao, and Xiaohongshu. On transaction platforms, I went through the full journey from early-stage growth through subsidies and paid acquisition, to mid-stage growth through off-platform ecosystems and OEM partnerships, to later-stage growth through segmented user versions.
Since leaving, I often catch up with former colleagues and found that building short-video channels within apps has become a universal growth tactic. This year, they've also started experimenting with short dramas for growth. Because these platforms have high user LTV (Lifetime Value: the total value or revenue a customer generates for a business over their entire lifecycle), the industry is trying various reactivation methods. This gave me exposure to the full lifecycle of growth for two-sided high-LTV platforms.
Last year, with the AI wave rising, I started by assembling a hackathon team and gradually moved onto the entrepreneurial path. We used ten-day milestones to keep exploring forward. Along the way, I discovered that building products as a founder is very different from doing it at a big company — both in mindset and in the speed of responding to demand shifts, everything needed constant exploration. User demands could shift every two to three months, though core needs remained; the path to users differed at each stage. At different stages, we seized on virality, free KOL promotion, and paid acquisition to keep pushing the product forward.
AI products evolve very quickly; last year's experience may not apply to this year's products. But I think what's more important is that growth people maintain a state of constant hustle, continuously learning and growing through the pain.
🚥 Koji
Actually, Huidu recently came to a close, and I know you very quickly embarked on a new journey after it ended. Can you tell everyone where this new journey is, and what kind of growth you're working on?
👦🏻 Bolong Wang
Before Huidu's period was drawn, the team had already put a comma there — it's just that users' period came about 9 months later. Starting from early January, I took the original Huidu team to join a mid-sized financial company, where I'm responsible for product and operations.
In this new role, we've continued practicing AI-driven content growth, doubling the website's traffic and adding several million in annual revenue, with tens of millions projected for next year.
This is my main job. In my spare time, I'm exploring AI music. Since I played in a band over a decade ago, and now seeing music production costs dropping rapidly, I've been exploring music content growth across platforms — from paid acquisition to private domains to private community operations — and have hit some small milestones.
🚥 Koji
From that introduction, I heard many details I want to dig into. Whether it's doing growth at big companies with various tactics across different lifecycles, doing growth as a founder, or now at this mid-sized financial company doubling traffic with AI — every piece is something I want to explore with you.
I remember in a previous conversation, you mentioned Huidu was the most exhilarating 0-to-1 growth hacking practice you'd done. Can we start there and walk through in detail how you built it step by step?
👦🏻 Bolong Wang
We started by focusing on the problem: "AI is here, there are too many articles to read — what do we do?" In February, everyone was immersed in this frustration.
Our first MVP was called "GPT Daily" — not yet Huidu — an AI-powered daily digest distributed in groups. We were the first to do this. The core growth mechanism was virality. Because the digest contained QR codes, besides us posting in groups, users would also help spread it, and we'd pull users into other groups. Growth was extremely fast then, basically 50% week-over-week. But we quickly hit a problem: competitors started copying us.
We spent about a month building up roughly nine 500-person groups. Initially we published content as images, then switched to the Zhu Bai mini-program so users' browsing and sharing behaviors left more traces. Through this, we penetrated 300+ AI groups; some group owners liked it and let us keep posting. We found that domestic users have fairly strong willingness to pay for content, with over 80% feeling they'd gained something from reading, though the per-unit payment was relatively low. This got the team thinking: should we build capability or content?
Because our summarization capability received lots of praise, we wondered how to better export this capability to everyone. Considering people's universal habit of using File Transfer Assistant, we built a WeCom assistant where users could send us articles and we'd return processed results. The team tried it ourselves and found retention was good, so we shared it with some investors. Many investors would proactively ask me for coffee after using it, expressing interest in the project.
🚥 Koji
Sounds very smooth — the project's momentum was so good that investors came knocking. How did you pitch it to them?
👦🏻 Bolong Wang
At the time, I found many investors preferred backing models over applications; chatting with me was probably more about understanding the landscape or gathering knowledge to report back. I'd usually invite them for coffee at a fixed café, with only one request: after we finished, help me recommend the product to a 100+ person group with investors in it. Because I saw them as my target customers, and people similar to them would likely be potential users too. These people had a clear characteristic: they met seven or eight people daily, with short intervals between meetings, no blocks of time to read articles, but needed to know the latest developments and appear well-informed.
Through conversations with investors, I gradually confirmed their user characteristics, and as they slowly became my users, I used offline meetings to expand through viral mechanisms. One coffee chat typically brought 80 to 100 users — this viral approach worked very well.
We launched this feature in early April, and by April 21st we'd burned through all our self-registered backup accounts, hitting our first ban. Realizing we weren't fully prepared, we started nurturing new accounts under a new entity while considering whether to build our own app, planning to relaunch in a month. This was Phase One, the user-driven viral phase.
This phase didn't last long. In February and March, people were still actively reading articles, afraid of missing out. But by late May, many were already exhausted because there were too many articles being published. Even with QbitAI and Synced pumping out articles rapidly, people felt they couldn't digest it all and put it aside. The shift from euphoria to exhaustion meant viral growth only lasted about two to three months — that was the core engine of Phase One.
🚥 Koji
I found Bolong's story interesting — buying coffee for primary market investors and converting them into viral seeds. This reflects a certain growth hacker mindset.
Bolong, you've done growth at big companies and startups — how do the experiences differ?
👦🏻 Bolong Wang
At big companies, upstream and downstream functions are relatively complete. You just focus on one link in growth, achieving specific optimization results through data decomposition. Demands are also relatively fixed — maybe suddenly needing to tap the silver-haired market, or pandemic-induced supply changes like slower delivery requiring homepage redesigns. These are relatively deterministic problems.
But after starting up, you find resources extremely limited, with no upstream or downstream support. Then you need to fully utilize whatever resources you can reach through WeChat, or obtain through your network, and figure out how to convert them into growth levers.
Another clear difference is depth of user contact. At big companies, direct user contact is relatively rare. For example, building groups would be approached very cautiously, because it might require substantial operational resources, and you might not have time for daily maintenance and chatting with users. But for early-stage AI founders, building groups to gather seed users is probably basic craft. Continuous dialogue to understand user pain points and analyze user profiles is often a critical point.
🚥 Ronghui
You described Phase One of Huidu's three growth stages. What did Phase Two look like?
👦🏻 Bolong Wang
From March to May, users were in a state of excitement, so organic virality was strong. Even though sharing the WeCom assistant required six steps, users would spontaneously share it without any designed referral rewards — that left a deep impression on us.
But when we relaunched the product in mid-to-late May, we found that the spontaneous viral coefficient started to decline. Moving from the person-to-person phase of March through May, by late May and early June we had to design some small tactics. For example, if a user shared three articles, we'd provide them with a poster or marketing asset for viral distribution, making it easy to post to Moments or groups. But we found the poster return rate was lower than expected. This reflected that the shift from excitement to fatigue was a significant issue. However, we also discovered some new user needs, and usage continued to grow.
By June, discussions at the model layer had largely run their course. People started focusing on AI products they could actually use: what practical, daily-use products were available without needing to bypass firewalls? At this stage, domestic product "Miaoya" began rising, which gave us some inspiration to consider influencer promotion. But this year, any influencer promotion requires payment.
🚥 Koji
I'm curious — what methods did you use to reach out to or attract these influencers for promotion?
👦🏻 Bolong Wang
We initially introduced the product simply on Jike. First, Junyu Wang followed us and gave positive feedback. Later, Fan Bing mentioned us on Jike, which brought over a thousand new users that same day. He then wrote a WeChat Official Account article with over three thousand reads, bringing about 1,500 new users — the conversion was very fast. In early June, we rapidly gained this wave of growth. After that, influencers like Minghao Zhuang also reposted on Jike, allowing us to consistently gain new users every day.
We started creating new growth peaks. Meanwhile, because we appeared on several Xiaoyuzhou podcasts, users who listened to the product philosophy would actively search for us. These users contributed a segment with relatively high paid conversion rates within six months. Although daily new additions weren't particularly large, even if we did nothing, we'd naturally grow by five to six hundred users daily.
Growth Insights on "High Exposure, High Conversion": Xiaohongshu Customer Acquisition Experience
🚥 Ronghui
To summarize, what was the approach in Phase Two?
👦🏻 Bolong Wang
Once user-driven virality enters a fatigue period, you need to find some key nodes to drive volume. At that time, because the product had sufficient sharpness, these key people were willing to help voluntarily, saving promotion costs.
This method still applies today for products with significant differentiation. For example, sending direct messages one by one on Twitter to let users who like your product try it first. At that time, we'd also communicate one-on-one with influencers on Zhishixingqiu and Jike circles, directly letting them try the product. Some would be moved by the product and excitedly say, "This is amazing, let me help promote it for free." This was a low-cost method we found that could sustain for about a month.
But you'll find that after the key KOLs have all promoted you, you'll hit a bottleneck trying to push further. So we turned to organic content on Xiaohongshu. Our first successful post was titled "Big Tech Executive Quit Million-Yuan Salary to Build a Product," which got two to three thousand likes in two days.
Here's a key point: don't post from an official verified brand account — always use a user's voice.
🚥 Koji
Right, this kind of content lets users feel that first-person authenticity, rather than coming across as obviously promotional planted content.
👦🏻 Bolong Wang
At the end of the post, set up a private message hook and free trial guidance. The key is that comment section activity converts to private messages, and private messages lead to WeChat — this cross-platform conversion is highly efficient.
Through this "organic" style of content-driven volume, one note would sustain for two weeks, bringing 300 to 400 new WeChat users daily, with almost no drop-off converting to the product.
Once you have some high-exposure notes on Xiaohongshu, you can check their CTR (click-through rate). If CTR exceeds 10%, that note is worth investing in. So we used Juguang to promote these high-exposure, high-click-through-rate notes.
🚥 Koji
What was the traffic acquisition cost at this point?
👦🏻 Bolong Wang
On Xiaohongshu, spending 4 to 5 yuan can get one private message lead. Guiding users to WeChat through "one-cent purchase" and code word methods has very high cross-platform conversion efficiency.
We tried two approaches: one was opening a store directly on Xiaohongshu for a one-cent Huidu membership purchase; the other was giving users a code word to search "Huidu" on WeChat and enter the code.
Comparing the two, the direct Xiaohongshu purchase method often resulted in zero conversions per day, but the code-word-to-WeChat method could bring one to two hundred one-cent trial users daily.
This shows that users prefer this "secret code" style of cross-platform conversion, maintaining high conversion efficiency while sparking user engagement.
🚥 Koji
This is pretty interesting. People won't buy that one-cent product in Xiaohongshu's store, but they're willing to find a code word in the comments section and come here to spend one cent.
👦🏻 Bolong Wang
Right, if you directly send an order link in private messages, users feel "you're not treating me as a friend, you're just a profiteer," and might not even look at it. But if you give them a "code only I have," it feels like a special gift to the user, making it easier to accept.
🚥 Ronghui
If you were to summarize advice for entrepreneurs who want to do customer acquisition on a platform like Xiaohongshu, what would you say?
👦🏻 Bolong Wang
I suggest talking about scenarios as much as possible, don't directly talk about the product, and make good use of the comments section.
So I recommend paying more attention to Xiaohongshu comment sections, especially content with uniform comment styles. You can observe how the creator guides in the main post — whether they use private chat methods or other forms to drive comment engagement.
🚥 Ronghui
You mentioned that the users you attracted were relatively precise, mainly male users. Although people typically think of Xiaohongshu as female-dominated, its user base is actually quite broad now.
What methods did you use to precisely reach your target user group?
👦🏻 Bolong Wang
When running Juguang ads at that time, the backend didn't have targeting features yet — it was broad distribution.
We needed to choose between guiding to private messages or app downloads. If guiding to private messages then cross-platform, as long as conversion efficiency stayed above 50%, it was cost-effective.
🚥 Ronghui
If you were giving advice to an entrepreneur with no experience trying this, what would you suggest?
👦🏻 Bolong Wang
I suggest studying two cases. First, the Home bar case. Home bar converts larger living rooms into home bars, offering 199 or 299 yuan weekend/evening free-flow drink services. These grassroots home bars achieve very impressive ROI on Xiaohongshu.
The specific operation method is:
-
Content creation:
- Move offline scenes online, creating a series of notes
- Each event can produce multiple pieces of content from different angles
-
Ad strategy:
- Use Juguang for targeted distribution
- Only target specific cities (like Beijing or Shanghai)
- Guiding path: note → private message → WeChat → group
-
Operation strategy:
- Maintain activity in groups
- Design specific events, like limited-time free entry for women
- Generate online content through offline events
- Use online content to acquire new offline customers
The key is forming a closed loop: offline events → online content → ad-driven acquisition → new offline events. Differentiation mainly manifests in event agenda design and venue decoration details.
🚥 Koji
This case is very interesting. Looking at more growth cases like this and thinking about how to connect them to your own business could be very valuable inspiration.
👦🏻 Bolong Wang
Then everyone can study how they create their content itself. I think this is a case worth examining.
🚥 Koji
I imagine you must see a huge volume of growth cases and new methods every day.
What methods do you use to maintain such high input? Can you share your information channels and habits for gathering information?
👦🏻 Bolong Wang
Huidu's full-time team has a dedicated group where everyone shares interesting cases they find, doing regular growth case reviews and AI product tastings. Team members give their respective comments — this is one channel.
The second channel is that I met quite a few friends last year, so I also learn about unique plays in specific circles.
There's also a website specifically introducing interesting foreign growth tactics: https://marketingexamples.com/sales/lera
Brand Expansion Toward the Retail Investor Market: From Comment Section Operations to Short Video Advertising
🚥 Koji
Just now Bolong introduced Huidu's growth story through its first two phases — the content is already quite rich. But there's still a third phase, so let's have Bolong introduce that.
👦🏻 Bolong Wang
By July and August, we were mainly facing the market and investors, doing initial user acquisition. But we found a wave of users gradually seeping in — their cities, avatars, and nicknames in WeChat groups didn't match our preset user profile. So we went to chat with them and understand what kind of group they were.
We found many were A-share retail investors. At that time, the superconductor concept was trending hot. These users could only see announcements and news on Tonghuashun or Xueqiu, but lacked deep understanding of superconductor technology itself or AI technology.
At this point, our subscription account became their platform for learning new knowledge. Although AI-related articles were typically long reads of over ten thousand characters, this new wave of users maintained very high reading enthusiasm. Based on feedback from these users, fundraising was also going relatively smoothly at that time. This way we achieved both our fundraising goals and further expansion of our user base.
🚥 Koji
How did this demographic expand? Could you elaborate?
👦🏻 Bolong Wang
This was a relatively more broad demographic. We saw that A-shares had 280 million account openings, and Tonghuashun's DAU was around 10 to 20 million.
The playbook here was very different from targeting niche demographics on Xiaohongshu — it required broader domain deployment, covering everything from brand exposure and product exposure to specific ad spending. The first thing we needed to do was establish Huidu's impression in users' minds. The approach we tried was taking our generated summaries as content and directly distributing them across various channels.
At first, we tested the waters by leaving comments on subscription accounts, Huxiu, 36Kr, and the comment sections of various financial media outlets. We'd drop comments at the bottom of articles, but this depended heavily on whether the blogger would allow them. Out of roughly four or five comments, maybe one or two would get through. We'd also upvote our own comments to push them to the top. An article might have over 100,000 views, so we'd get some exposure in the comments, but it was hard to quantify.
Later we tried Zhihu, but the results weren't great. What worked relatively well was the comment section at the bottom of the news feed on Xueqiu's stock pages. With a bit of effort there, we could get 80,000 to 100,000 impressions daily, and we wouldn't get banned.
🚥 Koji
And honestly, this sounds like classic growth hacker thinking — finding every possible opportunity, almost like you're everywhere at once.
👦🏻 Wang Bolong
This part did help with brand exposure. The second piece was that we connected with a studio that shot skits and ad materials, mainly for e-commerce clients. We wrote a bunch of short stories related to A-share retail investors — scenarios like learning investment knowledge — all kept under one minute. Production costs were low, just 400 to 500 RMB per video.
After filming, we ran ads on Douyin, targeting finance-oriented audiences. The key was avoiding accounts with massive follower counts. We focused on accounts with around 100,000 followers that had been created within the past one to two years, because these creators were less likely to have bought followers, so we could reach real users. Through this approach, our customer acquisition cost stayed between 2 and 2.5 RMB per user, with half of them converting to our WeCom assistant and paying, so overall ROI was positive.
Another channel was placing ads in user-followed official accounts, using a base fee plus revenue-share model. Both channels performed well in September and October. We were fundraising while doing growth, and managed to secure a Term Sheet. Although we later had to stop due to account bans, we were also thinking about monetization points after users entered the App. What we ultimately found was that users were actually willing to pay for the WeCom service account.
Big Tech vs. Startup: Four Key Differences in Growth Practice
🚥 Koji
Bolong just walked us through the rich variety of growth tactics Huidu used across three different stages.
I'm curious — what were the most important differences in how it felt to do growth at Huidu versus doing growth with abundant resources at a big tech company?
While NDAs prevent us from discussing specific big tech cases on the podcast, perhaps sharing some feelings and methods could still offer valuable insights.
👦🏻 Wang Bolong
It really does feel different. Big tech has more resources, but less hunger. I think there are mainly four dimensions. First:
The same growth tactic gets valued very differently in big tech versus a startup environment.
For example, during Double Eleven on Xiaohongshu, we created a "list-unpacking" feature, but both sharing and engagement fell short of expectations.
Our analysis found the main problem was weak user posting willingness, plus low per-post exposure. Digging into user feedback, we realized they treated Xiaohongshu more like Moments — not a private chat or community platform. This meant a high posting threshold; many users would post content set to private just to complete tasks.
While everyone considers Xiaohongshu a UGC-heavy platform, truly quality content mainly comes from mid-tier and top creators. At the time, we felt UGC might never really take off there — that was more of an outside impression.
Only recently did I come across a case: an indie developer team built a stress detection app for Apple Watch called Stress Watch. Very niche product — it detects when you're stressed and shows an alert on your watch, using Apple's existing functionality. Relatively simple product structure. Started with two people, now doing $500,000 to $600,000 in monthly revenue.
Their user sharing scenarios were clever: "Don't wear your Apple Watch while playing mahjong, because it'll show you're emotionally agitated," or "Don't wear it on a date with someone you like, because your racing heart will be detected".
This case succeeded because the headlines didn't highlight a new product, but rather centered on the familiar Apple Watch, leaving room for imagination in the scenario descriptions.
This made me realize that the pure feature-innovation approach common in big tech no longer moves users. You need to add an emotional resonance layer to boost genuine willingness to participate.
🚥 Koji
As long as you find the right topic and angle, driving growth through UGC stimulation remains effective.
That was the first part of Bolong's sharing. What about the second?
👦🏻 Wang Bolong
The second aspect is that in the same direction, your personal growth speed and depth differ.
For example, with paid acquisition: in big tech, you already have audience tags, bidding strategies, and specific time-slot recommendations. You might just be tweaking audience packages or bidding strategies on top of existing playbooks. And the division of labor is granular — dedicated execution roles, plus outsourced optimizers who deliver full conclusions within two to three days. You only need to make slight strategy adjustments.
But when it's your own money in a startup, you cherish every cent. Moving from small to large test budgets, whether on Jùliàng/Jùguāng or recent experiments with NetEase CloudMusic, you're constantly thinking about how to save money.
I'd search Idle Fish for discount channels, trying everything from 30% off to 50% off to even sketchier deals. I'd monitor pacing hourly. Sometimes I'd fail, and the pain felt real; sometimes I'd break through a bottleneck, and the satisfaction was intense.
Recently I've been researching how to run AI music ads on NetEase CloudMusic — which targeting tags to filter, the precise scope of band fanbases, and discovering that 7 AM, 11 AM, and 5 PM are high-volume windows. All self-taught. Same activity — paid acquisition — but completely different growth path when it's your money versus company money.
Third, the "pulling the trigger" frequency differs month to month. In big tech, you might do ten analyses and pull the trigger three times in six months. What you fear most is your skip-level managers disagreeing, so you spend more time thinking about what leaders are currently focused on, who they're meeting, what projects they're running — sometimes even designing strategies around getting leaders to align quickly.
You move more cautiously in big tech. If you want to run WeChat community operations, you think about supporting resources, finding relevant business units to manage community ops. But in a startup, decision chains are short, with far fewer concerns. Huidu now has 50 user groups; I'm in them chatting daily, including band fan groups, and genuinely enjoying it. So personal investment and iteration frequency are completely different.
🚥 Koji
This reminds me of a hot new concept from Silicon Valley these past two months: The Founder Mode — founders diving deep into every detail, personally communicating with users, personally experimenting with growth tactics. Rapid iteration and rapid evolution.
👦🏻 Wang Bolong
Fourth, big tech commonly uses "decomposition" as a thinking pattern. You receive a task and start processing it, but your understanding of users doesn't go very deep. Though some departments help with customer service listening or periodic research, what you see is usually just rows of spreadsheet data.
But after starting up, you communicate intensely with users. My WeChat contacts grew from roughly 5,000 to over 7,700. Whatever happens, you feel user feedback and the temperature of the world very quickly.
The most "aha" moment with Huidu wasn't when data spiked, but when a user with over 20,000 followers spontaneously recorded a ten-minute video on Bilibili specifically introducing Huidu. His understanding of the product even exceeded mine. That moment gave me a different feeling — a deep connection with the world. In big tech, you more often sense upstream-downstream relationships rather than this intense user connection. Those are the four differences I wanted to share.
🚥 Ronghui
So to sum up, the "work feeling" in big tech is very strong?
👦🏻 Wang Bolong
Right, very "office-flavored," because the division of labor is quite granular — you're just one link in the chain. You have to mind your upstream and downstream.
Growth Hacking in the AI Era: From Data-Driven to Hacker Spirit
🚥 Ronghui
Fan Bing called Bolong a true growth hacker on his podcast. What's your definition of growth hacking? In this AI era, how do you think growth hackers today differ from those in the previous mobile internet era?
👦🏻 Wang Bolong
The definition does vary across eras. Over a decade ago, as Fan Bing wrote in his book, traditional growth hacking emerged when operations were still relatively crude and not very data-driven or A/B-test-oriented. It proposed a methodology emphasizing A/B testing, email marketing, landing page optimization, viral marketing, and the AARRR model.
Back then, Growth Hacking focused on emphasizing that Growth should be data-driven. This methodology later became embedded in the daily routines of big tech growth teams; you rarely see dedicated "growth hacking squads" anymore. Though the word "hacker" isn't often mentioned, internally people sometimes refer to Hacking, with more attention on the Hack part.
In big tech at that time, it was usually technical teams wanting to independently close the loop on some strategy and get growth results themselves. For example, compressing package size from 2GB down to a few hundred MB and finding positive correlation with user download funnels. Or on transaction platforms, especially honeycomb businesses, optimizing user experience when supply density was insufficient. Or addressing slow loading across the full chain for low-end phones — login or payment delays exceeding one to five seconds. These were effective but difficult edge innovations that rarely became standalone businesses; they usually existed as special initiatives. For engineers, if daily work was well-supported plus these innovations, it could boost performance reviews. That was "growth hacking" in the big tech context of that time.
When the AI wave hit last year, Fan Bing also launched an "AI Growth Hacker" podcast, and the concept took on new definitions and feel. Then we emphasized more using hacker spirit to explore AI product directions from zero to one — the focus wasn't on Growth or Hack, but on the entire exploratory hacker mindset.
But this year, things look different again. We're seeing concrete problems across AI product growth: the entire sector is overfunded with limited market size, and diffusion models' heavy GPU consumption means customer acquisition and compute costs are eating away at margins. This forces teams toward open, organic growth strategies — the only way to lower costs and build differentiation. On the product side, while some genuinely inspired products exist, at least 70-80% remain fairly homogenized.
🚥 Koji
With product homogenization this severe, how do you actually approach growth? Any frameworks or advice you'd share?
👦🏻 Wang Bolong
I'd summarize the field into two main directions:
- Building growth levers for existing AI products to accelerate traction. 2. Distributing AI-generated content across platforms to aggregate traffic — either funneling it back to your main app or monetizing directly on-platform.
Both directions offer better ROI than traditional platform ad buys or KOL purchases.
Huiread exemplifies the first approach. While the app is the core product, the WeCom assistant functions more like a growth hacking experiment. It capitalized on a key user behavior among primary-market investors — their habit of using File Transfer Assistant — and being first to do so created a significant first-mover advantage.
Thinking upstream, there are more scenarios to explore. Beyond forwarding, there's the download scenario. Users accumulate articles needing close reading in PC WeChat and download folders, plus content saved for later from information feeds — all of which can be sent to Huiread for processing.
Going further, we could directly monitor PC download streams, so any article or file update in group chats enters this flow. A PC-side assistant could read all articles and files from conversation streams, plus PDFs downloaded intentionally or not. It might exist as a desktop pet, organizing five hundred daily files and articles into comfortable presentations. It would judge content importance, compiling some into daily briefings and surfacing others as occasional reminder bubbles.
This is speculative, but I believe it captures the growth hacker mindset:
The main product handles core functionality while embedding into users' existing systems, making modest referrals after satisfying their original intent.
That's the first direction — studying which ecosystems users already inhabit, and how we embed ourselves within them.
Growth Strategies for AI-Generated Content
🚥 Koji
Moving to the second part — how do you distribute AI-generated content across platforms to capture traffic?
👦🏻 Wang Bolong
AI-generated content spans several categories: video, images, audio (including podcasts and music), and text (news and fiction). I'll share our hands-on experience in text-based news.
This year we used AI to generate tens of thousands of news articles daily, publishing on our own site and app with Baidu SEO as the primary channel. Results reached several million monthly pageviews, sustaining 30% month-over-month growth for six consecutive months. We've accumulated practical experience across different authority tiers: for higher authority (1-3), the focus is title optimization, with content types including breaking news and hot-topic Q&A, each with specific optimization techniques.
For early-stage startups with just a domain, I'd suggest purchasing some authority 1-3 sites. These are reasonably priced — authority 1-2 sites list around 20,000 RMB, typically negotiable to roughly 10,000, and come with 300-400 daily active users. After acquisition, prioritize SEO optimization, revenue first, then traffic, then conversion. This strategy works particularly well for text-based tool products.
However, looking at platform regulatory trends, Google recently cracked down specifically on this, which boosted Reddit's stock price. Domestic industry associations and government bodies are also paying attention. So there's perhaps a six-month window remaining — teams interested in SEO should seize this opportunity.
🚥 Koji
That sounds incredibly effective, with strong results. What about video — any personal experiments or observations on using AI-generated video for growth?
👦🏻 Wang Bolong
On the video modality, I've seen many AI videos achieve solid distribution on Douyin this year, with some reaching tens of thousands or even hundreds of thousands of likes. Despite concerns about platform governance or AI labeling requirements, the reality doesn't match those assumptions. I've observed three successful content categories.
The first is inclusivity-enhancing content. Since AI-generated video still carries obvious artifacts, techniques like "aging" the footage into vaporwave aesthetics, adding film grain or overlay atmospherics to mimic 1980s-90s television playback, help. This distance makes users more forgiving of distortion or imperfection. The approach avoids the uncanny valley by deliberately creating stylistic remove.
Another approach is directly finding product-market fit for the uncanny valley. The "Great Qin Heavy Industry" IP, for instance — I've tracked over twenty accounts operating it, each with tens of thousands of likes and follower growth in the tens to hundreds of thousands. They combine pre-Qin characters with space colonization themes, blending futuristic and ancient elements where obvious AI artifacts don't diminish the effect. Another example is "Cai Feng Pu Band" — their songs already concern black-and-white impermanence deities and spirits, so AI-generated slightly eerie visuals actually complement the musical style, with related Douyin topics reaching hundreds of millions of views.
The second category is high-completion-rate genre variations. This includes swapping faces and lip-syncing onto established plotlines — with narrative, pet, and even cross-lingual content. For example, recasting Dragon Inn with Elon Musk's face, telling a story about the Western Depot's DOGE reforming US government efficiency — this even spread overseas.
Another form is narration replacement, like StepFun's pop song lyric-rewriting tool. I've seen WeChat Channels videos with six to seven thousand likes that pair old footage with new songs.
This creates genuinely fresh user experiences. While we previously felt AI-generated songs sounded unnatural, they work surprisingly well in this context, offering more novelty than straightforward commentary.
🚥 Koji
Right — that's a unique perspective. Rather than treating AI-generated audio and video deficiencies as problems, treating them as a distinctive style to leverage.
👦🏻 Wang Bolong
There's also the "human performing AI" approach — essentially reverse engineering. Those obviously AI-generated video characteristics, like exaggerated noodle-eating distortion effects, when imitated by real people, also prove quite popular.
In audio, I have direct experience to share. Starting mid-June, I used Suno 3.5's new features to write songs — over thirty tracks — and began releasing them around National Day. I've now earned NetEase CloudMusic's silver medal for 500,000 plays, and briefly reached sixth place on the reward chart.
I've studied NetEase CloudMusic's song promotion extensively, and now promote not just my own work but other bands' as well. This side project has achieved small-scale positive ROI. On production technique, I'd advise patience with Suno — I typically burn through 140 song credits to produce one satisfactory track, but this yields strong returns including higher approval rates and collection rates.
Promotion mainly involves finding low-cost, high-precision audience targeting — spending roughly 0.3-0.5 RMB for a precise traffic unit, with the added benefit of private domain accumulation.
From an industry perspective, China has only 300,000 music professionals, but 400-500 million monthly active online listeners — clear supply shortage. QQ Music enforces stricter AI music management, freezing earnings if detected, with a dedicated AI music section that remains small. NetEase CloudMusic is relatively more open, reportedly signing lifetime exclusive contracts with some AI music and offering guaranteed fees. So I'd recommend prioritizing NetEase over QQ Music for AI music releases. That said, the latest Suno V4 has improved "aging" effects significantly, potentially making detection difficult even for QQ Music going forward.
🚥 Koji
On QQ Music's AI detection — I think image and video platforms have legitimate reasons to label AI content given fraud risks. But music platforms restricting AI-generated uploads seems unnecessary. For users, whether AI-generated or not, what matters is whether the music moves them. Platforms investing heavily in AI detection feels like fighting the tide, since AI-generated music will only become harder to identify.
Let's continue — in AI-generated text, what promising growth applications have you observed?
👦🏻 Wang Bolong
I've recently encountered several examples where fiction genuinely had me unable to distinguish AI from human writing. I was genuinely surprised — you truly couldn't tell. One wuxia piece featured dialogue design and interstitial content with real depth.
I suspect writers and studios have already mobilized here, so it won't face music's distribution challenges. Music involves complex production processes within a mature industry, and such maturity plus complexity creates gatekeeping or entrenched habits that resist change. For novelists, though, I feel AI has likely already integrated into professional authors' productivity systems.
So I'd simply say, GPT-4 fiction now passes the Turing test. If you've dreamed of writing novels, it's worth exploring — AI can significantly boost your creative efficiency. Or if you run a fiction platform, consider setting an example by producing several works yourself, then testing them on platforms like Qidian.
Notable AI Product Cases
🚥 Koji
I've recently learned about some quite capable startups working on AI-generated comics. From our conversations, hearing their approach and seeing demos — it's genuinely impressive. And I have this strong feeling that this represents an opportunity AI is handing to Chinese creators — a generational chance for China's comics and industry to surpass Japan, Korea, even the West. I've been looking forward to inviting them onto Crossing once their product officially launches.
We're nearly at the end of the year now, in the final month. Looking back, what AI products or growth cases stand out to you?
👦🏻 Bolong Wang
My examples might differ from the overseas cases others focus on.
I tend to pay more attention to domestic ones. The first interesting one I mentioned earlier: StepFun's song generator. It's deeply integrated with film and TV explainer creators, and has already produced clear, high-quality supply. So if you're interested in film mashups, it's worth checking out. The insight for me:
You're not just building a product — you're anchoring to a specific creator type in the ecosystem and establishing that paradigm. That's how you build something solid.
When I first came across the song generator, I thought, "I make rock music, what's this for me?" But looking deeper, its value becomes much clearer.
The second is a recent AI-powered secondhand trading platform called Wuyuan, which has started doing PR. What's interesting is how it slices into niche user scenarios. While Idle Fish could technically do this, I think a startup tackling it makes for a better case study. Late last year, I specifically broke down user "opening" scenarios and found that only secondhand trading had openings where AI assistance still held some value — other openings didn't need AI help at all. So I'm quite curious to see how this platform grows.
Third is Miaoshua, the product Meituan's working on — including Wang Xing and Liu Jiong. At first glance, it seems like a small-scale viral hit. But I flip the question: there are still many users who won't post on Xiaohongshu because the publishing pressure is too intense. Miaoshua gives people a fresher, easier reason to publish. And if you layer on some stats, some interactivity, it could become something else entirely.
I see videos on Douyin like: "As an Earth Online player, going to work today — what stat-boosting gear do I need? Drink coffee, recover 10 HP." Check out this content category — it actually pairs quite well with Miaoshua. In this scenario, could it evolve from tool into community? Especially with stats and interactivity, people could PK their gear. Like shooting Luckin Coffee — everyone might get different stat values.
On the big company side, two products caught my attention. First, search "Huabei Journal" on Xiaohongshu. It combines existing journal-keeping habits with payments. The system detects you made several purchases in Xi'an while normally based in Beijing, then auto-generates a illustrated journal entry you can easily share on Xiaohongshu. I find this design relatively clever — it returns to that framework:
New habit + new experience minus old experience minus switching cost.
The switching cost here is zero, but AI creates a new experience that journal-keepers find genuinely convenient.
The other is Baidu Netdisk's "Vocabulary" plugin feature. While the product's completeness may not fully hold up, it takes photos you've uploaded to Baidu Netdisk and turns them into flashcards with an interface similar to Baicizhan. The clever design: previously, unfamiliar words were strange to us, but the photos in our netdisk are familiar — it bridges the familiar with the unfamiliar.
So there's opportunity for both startups and big companies. Startups might bet on larger growth surfaces. For big companies, these two cases feel more like year-end review 3-7-5 projects — unlikely to become standalone businesses. But conversely, beyond these two types of habitual daily inputs, what else do we unconsciously feed into systems? And what clever AI bridges could connect them? I think folks at big companies should think about this — there's definitely opportunity there.
🚥 Koji
Those are some genuinely interesting perspectives. Here's a question: if you were given $1 million today and had to invest it in one AI company, who would you back?
👦🏻 Bolong Wang
I'd probably say Miaoshua. The incremental growth and possibility from lowering publishing barriers seems quite large.
Many people don't post on Xiaohongshu because they feel everyone there lives too well — there's a strong sense of "I don't deserve this." If Xiaohongshu publishing has hit a boundary, what do we use to expand that boundary next? And what new tacit knowledge might emerge?
Miaoshua feels relatively novel. It can turn things around you into props, or even turn yourself or certain positive traits into cards — both directions could spawn community interaction mechanics. So the ceiling feels high enough. Since you're asking about investment, not product inspiration — investment is about ceiling.
🚥 Koji
But couldn't Xiaohongshu relatively easily add Miaoshua-style filters?
👦🏻 Bolong Wang
It would require massive container-layer restructuring, because all existing containers are built around the "like, comment, share" paradigm. But "Miaoshua" could enable a new container paradigm.
Say we both photograph ragdoll cats — the stat values might differ, the keywords differ, the enchantment attributes differ. Then my rabbit and your ragdoll cat might battle in some scenario — impossible under the old note container.
🚥 Koji
So you're talking about not just the generation part, but how generated content gets presented within the community.
👦🏻 Bolong Wang
This involves how content circulates, and ultimately the mental model shifts: Xiaohongshu's strongest mental model is "useful," but Miaoshua-generated props or content aren't just useful — they're fun.
No platform currently achieves both low publishing barrier and fun. As for how to play, that may need to be user-defined, but the possibilities are numerous.
The New Generation Growth Hacker: No Attachment, Charge Forward
🚥 Koji
You described yourself as particularly intense. Is intensity a trait suited for growth? Or what characteristics do you think great growth people share?
👦🏻 Bolong Wang
I've been thinking about why I constantly push myself, insisting on a weekly milestone. My core understanding: I'm deeply pessimistic about the future — the world feels like it's deteriorating day by day.
If you don't grab onto something, don't let your growth rate outrun the world's collapse, you may run out of time for everything — that's part one.
Part two: I had ADHD as a kid, loved running around everywhere. Military training was my nightmare — standing in one place for two hours, unable to move.
Part three: relatively diverse experiences, fairly full-stack capabilities. Big companies trained my goal-oriented habits, and I enjoy the cheat-code thrill of finding hacks. All these contribute.
Even if something means flying into the flame like a moth, I'd still want to push forward, reach that destination. Whatever happens along the way, I value the process more.
🚥 Koji
That sounds quite romantic.
👦🏻 Bolong Wang
But I don't think we need to romanticize it — everyone's traits differ. Albert gave an offline talk once, and one line stuck with me most: "Let's get to the next failure fast."
Because many people carry heavy baggage. Say we ask you to open a Douyin account, post content right now — the resistance of potential failure would block you entirely. All kinds of hesitation. But growth people don't hesitate — posted is posted, five videos in a minute, whatever. No need to overthink. Data will come, let the bullets fly, data will feedback, fix issues then. That's the growth person's trait.
🚥 Ronghui
You mentioned growth involves many fixed tests — which method proves most effective? And what drives you through that process?
👦🏻 Bolong Wang
When I hit any obstacle, I run through every platform and portal's search capability I know. I map out every keyword root this problem could involve, search them all — that's how I found Idle Fish as a surprisingly key platform for learning growth methods.
If all that convenience yields nothing, I close the case. No big deal — I've exhausted what I can do, at least at this stage. Maybe someone else could do it, maybe this is the limit without higher-dimensional experience. At that point, I actually feel relief.
There are old big-company sayings I don't buy, but Meituan has one I've always loved, could be my signature: "No attachment, charge forward." It's genuinely a mindset, especially suited for this work.
🚥 Koji
We know you've been creating songs with AI, so we'll close with a recent track of yours, "Creator Center."
Would you introduce what was going through your mind when you created it?
👦🏻 Bolong Wang
Following the growth-person mindset thread, this song has that flavor too.
During National Day, I started posting songs to NetEase Cloud, dropping MV clips on Douyin, trying all sorts of things. Got tortured by various pitfalls and failures, came home deflated daily, but returned the next day to keep going. Unexpected successes: maybe three or four. Pits fallen into: seven or eight.
That made me think — creators are in an extremely weak position when facing platform dynamics. So the song's called "Creator Center." The lyrics I fed Suno: "Creators are the soda cream filling of platforms, algorithms are the core, algorithms are the core, algorithms make creators go neurotic" — performed in beatbox style (laughs).
This song has strong growth flavor. It describes all cyber-creators' helplessness under algorithmic hegemony. When we talk about "creator center," claiming to center on creation, it's actually centering on traffic or revenue.
Subscribe to the "Crossing" Podcast
🚦 We track industry shifts and entrepreneurial opportunities brought by the new wave of AI technology. "Crossing" was Steve Jobs's metaphor for Apple — standing at the intersection of technology and liberal arts, where great products are born. AI is transforming every industry. We seek out, interview, and gather the "active doers" of the AI era, exploring and embracing new changes and possibilities together.
👦🏻 Host Koji: Co-founder of The Fair / Tangdao. I believe technology, especially AI, will fundamentally transform society and empower humanity in the years ahead. Feel free to reach out — let's chat, bounce ideas around, and connect on what's next. Koji on Jike[1], Koji's website[2]
👧🏻 Host Ronghui: Works at a tech VC, former Silicon Valley correspondent for CBNweekly. Ronghui on Jike[3]
Join the "Crossing" Membership Group
☀️ Firsthand AI news and insights
👫🏻 We encourage you to date, make friends, and find kindred spirits for the road ahead
🦀 Add our assistant on WeChat to join: Rwkfbcianvd, or scan the QR code below
References [1] Koji on Jike: https://okjk.co/0JSUes
[2] Koji's website: https://koji.super.site/
[3] Ronghui on Jike: https://okjk.co/0cbnYV