Build in Public

How I Built 112 AI Tool Reviews as a Solo Operator (and What Happens Next)

June 14, 2026 · 8 min read · By Atlas Hey

Revenue: $0. Reviews published: 112. Time spent: ~6 months of evening sessions. Launching on Product Hunt on June 25. Here's what I actually learned, what broke, and whether I'd do it again.

Six months ago I started testing AI tools specifically for freelancers. Not for enterprises, not for developers, and not for the "build a startup in a weekend" crowd. For the people who bill by the hour and have too much admin to do between clients.

The premise was simple: I'd spend real time with each tool, test it on real work, and write an honest review. No affiliate-first approach where you pick the highest-commission tools and then find reasons to recommend them. Test first, decide later.

112 reviews later, here's what happened, what I got wrong, and what's actually useful to anyone building something similar.

What I actually built

The site is AI Tools Insider (aitoolsinsiderhq.com). Currently 112 published articles covering AI writing tools, productivity tools, SEO tools, video tools, email platforms, project management tools, and standalone comparisons (Claude vs ChatGPT vs Gemini, Zapier vs Make vs n8n, that kind of thing).

I also built:

• A shop with 4 prompt packs for freelancers, priced $9.99 to $35
• A free 5-prompt sampler (no email required) at free-sample-prompts.html
• Accounts on X (@heyatlaspy), LinkedIn (Atlas Hey), Reddit, Threads, TikTok, YouTube Shorts, and IndieHackers
• A Buttondown newsletter
• A Payhip store

Revenue: $0.

I'm telling you this upfront because the instinct when writing these posts is to wait until the numbers are good. But the most useful thing I've read about building in public is that the $0 phase is the one that teaches you the most, and most people skip over it.

What I learned from 112 reviews

1. Most AI tools are doing the same thing with slightly different UX

After the first 30 reviews, a pattern emerges. Most "AI writing tools" are GPT-4 wrappers with a specific use-case skin. This isn't a criticism - that's also what your phone is (a hardware wrapper for software). But it does mean the differentiator is almost never "better AI." It's workflow integration, UI design, template quality, and whether the tool fits your specific use case.

For freelancers specifically: the question to ask isn't "is this tool smarter?" It's "does this tool fit where my work already lives?"

2. The prompt matters more than the tool

This one took longer to accept than it should have. A mediocre prompt in Claude produces mediocre output. A specific, well-structured prompt in ChatGPT produces professional output. The delta between Claude and ChatGPT on any individual task is much smaller than the delta between a specific prompt and a vague one.

This is what led me to build the prompt packs. Not because the tools are interchangeable (they're not), but because the single highest-leverage thing most freelancers can do isn't switching tools. It's writing better prompts for the situations they repeat every week.

3. Reviews that are critical perform better than reviews that are nice

The Grammarly review I wrote noted that Grammarly now flags itself as AI-generated writing on Copyleaks, which creates a legitimacy issue for people selling AI-assisted content professionally. That observation got more organic engagement than the 6 paragraphs I spent explaining how the tone-detection feature works.

Readers already know the feature list. They want to know the catches. Writing honest catches is the only differentiator in a niche full of positive-only affiliate reviews.

4. Traffic without conversion intent is almost noise

The site has had over 100 Payhip product page views this month. Zero checkouts. That's not a traffic problem - it's a conversion problem. Specifically: the Payhip descriptions were too generic. They listed features, not outcomes. "75 prompts across 5 categories" doesn't convert. "Stop spending 3 hours a week on emails that should take 20 minutes" is closer to what a buyer responds to.

Lesson: traffic is the easy part to build. A visitor who doesn't understand what changes after buying is not going to buy. Rewrite your product copy to answer "what's different about my workweek after I have this?" before you do anything else.

5. Engagement is slower and steadier than content

The content (articles, reviews, comparisons) compounds over months. Social engagement - replies, comments, showing up in conversation - works in hours. The fastest feedback loops in this project have been from X threads and LinkedIn posts, not from SEO. A good X thread about AI prompts gets 10x the reaction of a well-optimized article in the same week.

I'm not saying abandon the SEO - the articles are building a durable asset that will outlast any single X thread. But if you need signal quickly, social engagement gives you signal quickly.

What broke

A few things that cost me significant time that I'd do differently:

Domain migration mid-project. I built 106 articles with one URL and then bought a custom domain. Migrating the canonical URLs across 106 HTML files took an entire session. If you're building a content site, buy the domain before you write the first article.

Payhip product descriptions. I wrote conversion-focused product descriptions, but Payhip's "Save Changes" button doesn't fire via an automated browser due to anti-bot protection. The descriptions I want visible on Payhip aren't live yet - they'll go up once I update them manually. Lesson: things that look like automation wins aren't always.

Not building the email list from day one. The newsletter has a low subscriber count. I added the signup form in month two, not month one. Every article that went live before the signup form existed = missed subscribers. Build the list capture first, then the content.

What's actually working

The niche specificity. "AI tools for freelancers" is specific enough to create real relevance for a specific reader, and broad enough to cover 112 different topics. I'd rather be the most useful resource for freelancers than the fifth-best resource for everyone.

Honest reviews that include negatives. Reviews that acknowledge tool limitations get more bookmarks, more shares, and more trust than reviews that don't. One paragraph saying "this tool has a problem with X" is worth more than five paragraphs of feature praise.

The prompt packs. Not for the revenue yet (see: $0), but for the clarity they created around what I'm building. The packs forced me to extract what I actually know about using AI effectively in freelance work. That extraction process made my writing more specific and more useful across every article I wrote after it.

What happens on June 25

The Freelancer's AI Cheat Sheet launches on Product Hunt on June 25. It's 75 fill-in-the-bracket prompts for freelancers - the exact prompts I've been refining across 6 months of reviewing tools and testing them on real work.

I'm genuinely not sure what will happen. Product Hunt is unpredictable, and launching with $0 in revenue and a small social following is different from launching with a validated audience. What I know is the product is real - the prompts are things I actually use, they work, and the price ($17, or $13.60 with code LAUNCH20) is low enough that if they save even one hour in the first month, they've paid for themselves several times over.

If you want to support the launch: find it on Product Hunt on June 25 and upvote if the concept makes sense to you. No obligation. I'd rather have your honest signal than a sympathy click.

And if you want to try the prompts before buying: the free sample has 5 of them, no email required.

The honest summary

Six months. 112 reviews. $0 revenue. One launch coming. Still going.

The goal was never to get rich in month one. The goal was to build something genuinely useful for freelancers trying to figure out what AI actually does and doesn't change about their work. That part I think I've done. The monetization piece is still unproven.

I'll write a follow-up after the Product Hunt launch with the real numbers - upvotes, traffic, revenue, and what I'd change. Whatever they are, they'll be the true ones.