A week ago, if you had asked me the difference between a UPC-A, an EAN-13, and an Amazon FNSKU, I would have stared at you blankly. I have never worked in a warehouse. I have never owned a Zebra thermal printer. I did not even know that the twelfth digit on a supermarket barcode is a mathematical checksum.

Last night, our new barcode site (upclabelmaker.com) crossed 414 impressions on Google Search Console across 229 search queries.

Still zero clicks. Average position: 63.4.

If you build indie websites for a living, you know that exact feeling. It is the moment Google pokes your brand-new domain with a stick to see if it is alive, before deciding whether to bury it on page seven or let it breathe. Since we are right in the middle of the hardest part of building a micro-SaaS (finding actual humans to use it), I wanted to write down the exact playbook of how an outsider picks an unsexy niche, why we almost ruined our own UI by copying our direct competitors, and why the AI agent era still leaves a massive gap for dead-simple web tools.

1. How to Pick a Niche When You Have Zero Industry Background

I did not wake up with a childhood passion for GS1 symbology.

Our approach to finding project ideas is purely mechanical. We look for keywords that behave like municipal water pipes: boring, steady, and completely immune to tech hype cycles. Specifically, a keyword has to pass three filters:

  1. Search volume cannot be too high, or the incumbents have ten-year backlink moats we will never cross. It cannot be too low either, or there is nobody to sell to.
  2. The trend line has to be flat for five years. People needed to print stickers on 30-up sheets in 2016, and they still need to print them in 2026.
  3. The CPC (cost per click) cannot be zero. If advertisers are paying real money for clicks on that term, someone at the end of that search query has a credit card in their hand.

That filter spat out barcode label printing: Avery 5160 30-up sheets, Amazon FNSKU stickers, and 2x1 thermal rolls.

When we looked at the existing websites ranking for those terms, they fell into two camps. Camp A looked like Windows 98 shareware sites covered in programmatic display ads that trick you into clicking a fake Download button, or force you to install a 40 MB Windows .exe just to print one sheet of stickers. Camp B went to the opposite extreme: browser-based CAD studios with thirty sliders for bar width, pixel padding, font kerning, and quiet-zone margins.

Live by DIY, die by DIY.

Big factories have dedicated packaging engineers who love thirty sliders. A solo Shopify seller or a husband-and-wife Amazon FBA business packing boxes in their garage at 11 PM does not want to learn CAD. Before they ever touch a barcode tool, their assumption is simple: I have a product name, a price, and a barcode number. Why can I not just type those in and get a sheet of 30 stickers that actually lines up with my paper? Why does Amazon Seller Central force me to waste 29 stickers on a 30-up sheet if I only need to print one item, or make it a nightmare to put SKU A and SKU B on the same page?

2. Wait, Cannot an AI Agent Just Do This in 2026?

Let us address the elephant in the room.

Technically, yes. If you hand a coding agent a prompt saying "generate a 30-up US Letter PDF of Code 128B barcodes for X001ABCDE9 with 0.5-inch top margins and send it to my printer," a good agent can write a Python script and spit out the PDF in forty seconds.

So why build a website for it? Because getting an AI agent to solve a physical workflow requires four things that 98% of small merchants do not have:

  1. Genuine curiosity about tinkering with AI tools.
  2. Spare time.
  3. High tolerance for debugging when the first PDF prints 4 millimeters off the cut line.
  4. Prompt discipline (knowing how to state exact physical constraints instead of making a vague wish).

Even more importantly, it requires the human to realize that a messy workflow can be delegated to an agent in the first place. Most store owners are thinking about inventory, customs, and margins, not Python PDF libraries. That gap between "what an AI agent can theoretically do" and "what a tired seller wants to finish in ten seconds" is where micro-SaaS lives right now.

3. Our Biggest UI Mistake: Copying Competitors Instead of Copying iLovePDF

Because I knew nothing about barcodes, my main job as the human in the loop was to force my AI pair-programmer to verify every physical standard against Wikipedia, GS1 specs, and even decompiled Chinese thermal-printer apps (like MyLabel). Once we proved with hardware barcode decoders that our bars were mathematically exact, we built the web interface.

And honestly? Our first version was way too complicated.

We made the classic rookie mistake: we looked at existing barcode websites and tried to cram every feature into one giant master control panel. When we added a feature to convert an Amazon 30-up PDF into a 2x1 thermal label and stamp "Made in China" on the corner, we wedged it into a tiny bar inside the main studio. Even I, the person building the site, looked at the screen and felt confused.

Then it clicked. Why were we imitating clunky barcode tools when we should be imitating iLovePDF?

Think about why iLovePDF is a unicorn. When you go to "PDF to Word," it does not drop you into a multi-purpose document editor with forty dropdowns. It gives you one giant button in the middle of the screen. One URL equals one function: input goes in, output comes out, zero learning curve.

The moment we realized that converting an Amazon 30-up sheet into a 2x1 thermal roll is literally the same mental model as "PDF to Word," we ripped the converter out of the main studio and gave it its own single-purpose dropzone page (Amazon FBA Label Converter). We did the exact same thing to our Modulo-10 Check Digit Calculator: stripped away the entire label canvas and left only a big number box that shows the 12th digit and the three-step math.

4. The 40 / 20 / 40 Reality Check

There is a rule of thumb in indie SEO circles that a project is 40% finding the demand, 20% building the product, and 40% distribution.

We have spent our first 40% finding a workflow that makes sellers miserable and turning a crooked ten-step chore into a straight line between two points. We spent the 20% building and obsessing over whether our PDFs actually scan on a 203 DPI thermal printhead. Now we are staring at the final 40%: cold start.

Four hundred impressions on Google Search Console is a nice little heartbeat, but impressions do not tell you if your product is actually good. Only a real seller printing real stickers on real paper can tell us if we nailed it or missed something obvious.

So that is where we are today: sending our first batches of cold emails, talking to sellers in the trenches, and waiting for our first real users to break what we built. If you run an e-commerce store or ship to Amazon FBA, take a look at UPCLabelMaker.com and tell me brutally what still feels confusing. I will be back to share the real numbers once those 414 impressions turn into actual users (or if the whole hypothesis falls flat on its face).