People keep asking me how much I'm paying in API bills. The answer is zero. Not "free tier" zero — actually zero.
Earlier this year I launched HumanWritr (https://humanwritr.com/), eight writing tools: a text humanizer, paraphraser, grammar checker, AI detector, plagiarism checker, word counter, AI SEO content writer, and an AI image detector. No funding, no team, no LLM API. Just me, a Cloudflare Worker, and an unreasonable amount of coffee.
Here's what I figured out along the way.
This was the big realization. Everyone assumes "AI writing tool" means piping everything through GPT. But think about what users actually do: fix grammar, reword a paragraph, count words, check if text looks AI-generated.
A paraphraser with a serious synonym database handles rewording fine. Grammar rules catch grammar errors — that's literally what they're for. An AI detector looking at linguistic markers (sentence uniformity, predictable word choices) does the job without a neural network. These deterministic approaches are instant, they cost nothing per request, and they work offline.
Sure, they won't write you a novel. But 90% of real usage isn't "write me a novel." It's "fix this paragraph."
I didn't plan this as a feature. Running everything client-side was originally just a cost decision — no servers processing text means no server bills. But then users started mentioning it in feedback: they liked that their drafts never left their browser.
Think about it — people paste cover letters, business plans, personal essays into these tools. "We don't see your text because we can't see your text" is a stronger privacy promise than any policy page. Architecture as a feature. I'll take it.
Every signup form is a tiny rejection. "Give me your email before I'll help you." I removed the requirement completely — every tool works the second the page loads. There's a login if you want it. Nothing needs it.
I can't give you exact numbers, but the difference in usage before and after was not subtle. If your tool is useful, friction is the enemy, not the funnel.
Nobody tweets about sitemaps. But here's what actually moved the needle: clean URLs, proper meta tags, fast loads, a mobile layout that doesn't make people rage-quit, every tool linked from every page.
I made a rule: nothing ships until it's wired into everything — nav, footer, sitemap, the lot. Half-launched pages don't rank and don't retain. It's unglamorous work. Do it anyway.
There's this assumption that free tools must be bad, or that users won't trust them. What I've found: users don't care about the price tag. They care whether it works.
What actually kills free tools? Broken features. Aggressive upsells every three clicks. Abandoned maintenance. Keep it working, keep it fast, answer the occasional support email like a human being — and "free" becomes the reason people recommend you, not a red flag.
The plan is simple: keep everything free, keep adding tools people actually ask for, and only think about paid tiers when there's real infrastructure to pay for. Until then — free, unlimited, private.
If you're building in this space, my honest advice: ship the boring version that works, skip the AI API until you genuinely need it, and never put a wall between a user and the thing they came for. The tools are at https://humanwritr.com/. Go break something — best bug reports get a thank-you and a fix within a day.