I originally got interested in TikTok for one simple reason: Everywhere I looked, founders were saying it was working.

Founders were growing like crazy and getting traction there. So I decided to try it myself and see if there was something real underneath the hype.

The product we ran this with was AIFlyer.ai.

AIFlyer is an AI design tool that helps people create flyers, social posts, and simple marketing visuals in seconds. It’s a general utility product. More consumer and SMB than enterprise. That context matters because a lot of what I’m about to share probably won’t work if you’re selling complex B2B software to a narrow ICP.

This post isn’t a guide on everything we tried. We tried a lot of things that didn’t work. I’ll focus on the only thing that worked, and why it worked. I’m happy to answer questions in the comments if you want to go deeper.

Like most people, our first instinct was influencers, just find people in that niche, and leverage their audience. But we failed, for several reasons.

One big reason was the budget. Influencer marketing only really works when you can place many bets. You need enough volume to discover the 10% of creators that will drive 90% of results. That usually means spending $10k–$15k a month just on experimentation and wasting 90% of the budget in the first couple of months. I couldn’t afford that luxury.

So we tried what most early founders do. We placed one or two bets and hoped we’d get lucky and we didn’t. Even when one video went viral, the ROI to paid was zero. Out of curiosity, I cross-referenced one of the creators with another YC founder to see if the post they made for them worked. Same story there.

We even tried bringing an influencer in-house for $1,000 a month to post once daily. The average video got about 200 views, and there were no breakouts. We also tried Whop and clipping, too, and it didn’t really help.

Eventually, we copied a carousel concept we saw from another brand that was basically like a meme. It had 157k views.

Massive engagement. And absolutely nothing happened to our business.

Two things became clear:

One thing we did early that ended up mattering a lot was adding a simple “How did you find us?” question during signup.

This gave us a baseline. If TikTok showed up consistently as a source, we’d keep going. If not, we’d kill it. That decision alone probably saved us months of guessing.

We kept posting anyway.

One day, I noticed something strange. A post that had completely flopped on day one, ~100 views, started climbing. A week later, it was at ~2,000 views.

That wasn’t normal, so I checked the analytics.

Almost 98% of the traffic came from search, not the For You Page (The feed curated by the TikTok algorithm where most viral videos get their views from).

That’s when it hit me: some people are using TikTok like Google.

That one post started bringing in ~20 signups and the occasional paid subscription every couple of weeks. Not huge, but real.

So we dug deeper.

That one post changed how we thought about the platform. Some people also use TikTok the same way they use Google. They search with intent.

So we leaned into that.

We started looking up keywords relevant to us and creating carousel posts specifically for them. Since we were building a design tool, this was straightforward. We’d write the content for the carousel post, then use AIFlyer to generate the designs for the carousel.

The next week, we had our next “viral” hit. But again, almost all the views came from search. Around 96% of traffic was from people actively looking for something.

We started looking up keywords relevant to us and creating carousel posts specifically for them. Since we were building a design tool, this was straightforward. We’d write the content for the carousel post, then use AIFlyer to generate the designs for the carousel.

The next week, we had our next “viral” hit. But again, almost all the views came from search. Around 96% of traffic was from people actively looking for something.

This time, the conversion was noticeably better.

A keyword like “How to make a flyer on iPhone” isn’t entertainment. It’s intent. This is the same type of keyword people fight over with SEO and pay top dollar for on Google Ads. On TikTok, it was free.

All we needed was a relevant carousel that subtly showed our product solving the problem.

Compounding Instead of Spiking

The beautiful thing about these posts is that they don’t need to win on day one, compared to viral videos where the video will max out in a week, and once that trend is gone, you have to start hunting for the next miracle.

Most of the screenshots I shared above had 100–300 views on the first day. Then they slowly accumulated views every day as people searched for that keyword.

When you build a library of these posts, each serving a specific search intent, it starts to compound.

At some point, something even crazier happened.

TikTok’s AI started mentioning us in the LLM response.

A friend sent me a screenshot where TikTok’s LLM answered a query and mentioned AIFlyer right after Canva. We hadn’t optimized for this. We didn’t even know it was happening. The model was just pulling from our content and sending people our way.

That was when it clicked that we weren’t just posting content anymore. We were feeding a system.

Once we understood the pattern, the next step was obvious.

I built a small AI agent that used the prompts we had manually used to generate all the contents for the carousels that worked, generated content across multiple relevant topics, used AIFlyer to create the designs with templates that converted via function calling, and automated posting.

We ran ten accounts, posting four times a day. TikTok’s automation API doesn’t allow adding sound, so I assumed this would fail but on average the posts did 4 times better on day 1 than what we had an undergrad student create the contents manually.

At that point, the math became interesting.

What does this look like when you have 1,000 posts per account, each serving a different search keyword, across ten accounts? Even if a single post brings one paid subscriber per day, that’s a completely different growth engine.

Especially if you’re in the discovery phase, where you just want people to try your product and later identify the power users who cover the economics, the way companies like Lovable or Replit do.

I tied the prompts, the design workflow, and the posting automation into an AI agent.

I remember telling a few founder friends who helped me early on that if this ever worked, I’d share it. This is me keeping that promise.

If you’re a founder building a consumer or SMB tool and want help setting something like this up, I’m happy to help you think through it.

If it makes sense, we can talk about using AIFlyer as the design and automation layer. If not, you’ll still leave with something useful.

You can reach out to me via email [email protected]. I’m genuinely happy to help.