First, the data: I bought the domain on the afternoon of October 9th. The first sale came at 2:38 AM the next day. In 72 hours, revenue hit $5,480. By the end of the first week, it was over $6,700, and now it's close to $7,200 (total Stripe transactions, including tax). No paid ads, no purchased backlinks.
If there's one secret to this success, it's just one word: Early.
Let me briefly explain what I built. In early October, a freestyle rap by Dominican rapper Bulin 47 went viral on TikTok. Many people used AI to insert themselves into that video clip. My website does exactly one thing: upload your photo, generate a 19-second video of you rapping in that scene.
Why I Chose a "New Keyword"
Let me clarify "new keyword": search terms that have just emerged and aren't yet targeted by many websites. For example, when a meme goes viral on TikTok, people start searching for "xxx ai." Whoever has a functional site first captures the initial wave of search traffic and AI-driven referrals.
I learned this "new keyword, new site" strategy from @gefei55's community group, covering how to select keywords, choose domains, and create SEO-friendly landing pages. People in our group have been doing this for a while. Previously, AI tool sites focused mainly on new model names, but as competition increased, I shifted to "old keywords" (terms with high existing volume and competition, like "ai image generator"). It wasn't until Ge Fei started sharing sales screenshots from his new keyword sites in the group that I realized this niche still offers significant returns.
In late September, Ge Fei personally built a new keyword site for educational purposes, accumulating $5,345 in revenue so far. During that period, he frequently shared new keywords in the group, and many members followed suit, achieving decent earnings. I was tempted but traveling abroad, so I didn't act until returning on October 8th.
October 8: Adding a Page to an Existing Site
On the morning of the 8th, Liang Chenmei sent me a list of new keywords, including "ai raindance" and "ai zombie" (AI video memes popular in early October). I thanked her and said I'd look into them, but I was busy with existing sites and didn't check immediately.
That afternoon, I considered building a new site for the keyword "hotel lobby ai." Thinking it was too late and wouldn't get traffic, I decided to just add a page to one of my existing sites to test. Around 6 PM, another friend in a small group reported their new "hotel lobby" site had made a sale. Realizing it was still possible to earn even days later, I quickly published that page. (It also generated sales the next day, serving as a catalyst.)
October 9: Finding the Next Keyword on TikTok
On the morning of the 9th, I seriously reviewed the keywords Liang Chenmei sent. Most already had significant competition. Entering now would mean fighting for scraps, not the main meal. Instead of chasing crowded trends, I wanted to find emerging ones.
Observant people noticed this wave of new keywords originated from TikTok. So why not go directly to TikTok to find them earlier?
With this idea, I fed "hotel lobby," "ai raindance," and "ai zombie" into Claude Code, asking it to find similar memes on TikTok where users want to replicate viral videos, and the replicated videos are also trending. It found several, including "burning bridges," which many later pursued. I chose another: Bulin 47. Checking Google Trends, it was just starting to rise, already surpassing "hotel lobby." The viral video was abstract and hard to understand, causing some hesitation. (Lesson: Trust data, not personal preference!)
I delayed buying the domain until the afternoon—a mistake to fix. Several candidate keywords existed; I compared them in Google Trends to see which had the highest volume and looked most like a tool-seeking query. I even stopped for lunch. Why emphasize "early"? When I registered, the desired domains were still available, making acquisition smooth.
Launching Same Day
Building the site itself was manageable. Testing took hours because I wasn't satisfied with the generation quality. Around midnight, my wife urged me to sleep, so I launched the imperfect version, adding restrictions and logic (e.g., auto-refund if the result didn't look like the user). Later, these proved counterproductive.
First Night: First Sale in 8 Minutes, Then Constant Bug Fixes
I should have slept, but the "hotel lobby" page added the previous day had two sales. Checking, I saw both users failed to generate. Debugging, fixing, and emailing apologies took until 2:19 AM. Meanwhile, visitors entered the new site, but no payments. I asked Claude to add funnel tracking in Plausible to see where users dropped off.
After handling that, I noticed the pricing section defaulted to subscriptions. Users wanting one-off videos won't subscribe. I switched to credit packs. Eight minutes later, someone bought. Twenty minutes later, three paid. Seeing rapid sales, I stayed up monitoring the backend, checking generation success and visitor counts.
Then I found a user who successfully generated a video but received an automatic refund. The video looked fine. After consulting AI, I realized the "fail if face doesn't change" logic Claude added earlier misjudged valid outputs. I loosened the criteria, then removed the logic entirely. Next, the site became sluggish and threw 500 errors due to database connection limits being overwhelmed by traffic. I fixed those too. The first two days involved little sleep, constantly debugging and improving UX. Here is a bug log from the first two days compiled by Claude Code:
Steps Taken to Reduce Customer Support Tickets
- Email users immediately upon video generation completion, inviting replies for issues.
- Refund all requests. If users feel quality is poor or want more credits, grant them.
- Monitor generation tasks closely. Contact users directly if problems arise. Trend-based site users expect instant results. Immediate failure detection allows quick response.
Data Insights
The $5,480 headline figure represents order prices within 72 hours of the first sale, excluding checkout taxes and refunds. Week one totaled 558 orders, $6,434 (order price), with 393 unique payers and 22 refunds.
From meme emergence to Google Trends peak takes about 4 days. 85% of revenue occurred in the first 72 hours. By day four, order volume dropped to one-fifth of the peak.
Surprising metrics:
- iPhone users converted at over 50% after opening the payment page; Android users only 17%. Likely due to Apple Pay's one-click ease versus Android's manual card entry without Google Pay setup.
- Traffic split: 50% direct, ~30% Google. ChatGPT referred users on day one, converting at 1.7x Google's rate. I implemented llms.txt and IndexNow, though unsure of impact. Matching domain/page titles to search queries seemed more critical.
- US visitors were only 25% of traffic but nearly 50% of payments. The meme originated in the Dominican Republic (second-highest traffic), but only 1/5 of purchasers completed payment.
- Refund rate: 3.9%, mostly due to "doesn't look like me."
Reusable Takeaways
Speed and timing are paramount, plus active community engagement. Other points:
- Act on signals immediately. Delaying domain purchase until 2:30 PM wasted hours.
- Prepare templates in advance. Actual coding took only 5 hours because payment, credits, login, and analytics were pre-built from other sites.
- Launch before perfecting. I deployed 23 times on day one. Issues surfaced via real users, impossible to predict beforehand.
- Don't let AI add arbitrary limits/auto-judgments. "300/day limit" and "fail if face unchanged" blocked paying users.
- Sell per-use, not subscriptions. Trends last <1 month. $9.99 single-video purchases accounted for 92%.
- Refund complaints about likeness instantly. Offering credits for retries usually ended in refunds anyway.
Common Questions
- Is this method reusable? Yes. A friend saw my share on Web.cafe (Ge Fei's community) on Friday, acted immediately, and made two sales by Sunday. My second site is also earning (~$500).
- API costs: Over 10,000 RMB so far.
- Free tier? None.
- Backlinks? None.
- Paid ads? Tested small amounts, then stopped as ROI was negative.
- Cart abandonment emails? Sent to 100 non-payers with a $6.99 discount. Zero conversions.
- Model used: Seedance 2.5.
Final Thoughts
I don't claim expertise. Success stems from Ge Fei's new keyword methodology, his consistent sharing of sales to motivate action, Liang Chenmei's keyword lists, and group discussions. The rest is luck.
Feel free to discuss further in the comments.
