Month 2 Building a Free AI Tool — What the Numbers Actually Look Like
Month 2 of running free AI image generator no account publicly. Here's the honest version — not the curated highlights post, but what the numbers actually look like and what I got wrong in month 1.
The Numbers
Traffic is growing month over month. Not hockey-stick growth — steady, compounding growth. The kind that comes from SEO and word of mouth rather than a viral moment.
What moved the needle most:
Blog content surprised me. I expected search traffic to be slow for months. It started arriving faster than anticipated — specific longtail queries ("free AI image generator no watermark," "stable diffusion alternative online") began ranking within 3-4 weeks of publishing.
The lesson: specificity in blog topics matters more than volume. A post targeting "free AI image generator" competes with thousands of pages. A post targeting "free AI image generator no watermark no sign up" has much less competition and converts better because the user intent is crystal clear.
What I Got Wrong in Month 1
Underestimated cold start impact on user experience.
I knew cold starts existed. I didn't fully appreciate how much the occasional 15-second first generation would affect perception, especially for first-time users who had no context for why it was slow.
Fixed in month 2 by keeping a minimum warm worker during peak hours. The median generation time stayed the same, but the worst-case first experience improved significantly.
Overestimated how much the gallery would drive engagement.
I built a public gallery of generated images thinking it would keep users on the site longer and inspire new prompts. Actual engagement: minimal. Users come to generate, not to browse.
The gallery is still there — it's not hurting anything — but I stopped optimizing for it and redirected that attention to the generation experience itself.
Underestimated prompt quality as a support issue.
The most common frustration isn't the tool failing — it's users not knowing how to write prompts that produce what they're imagining. This showed up in the rare feedback I did receive.
Response: more educational content. Blog posts explaining prompt techniques, style vocabulary, what specific words do. This serves users better than any UI change would.
What Actually Drives Return Visits
Without accounts and without email, return visits come from one thing: the tool being genuinely useful enough that people remember it and come back.
Tracking this (via cookie-based return visitor detection) shows a meaningful percentage of sessions come from returning visitors. This number has grown month over month, which I take as the clearest signal that the product is working.
What seems to drive returns:
Users who learned how prompts work (through the blog content or through experimentation) get better results and come back to generate more
Users who found the no-account experience genuinely refreshing compared to other tools they've tried
Users who bookmarked it for a specific use case (social media content, blog images) and return when they need more
Content Strategy — What's Working
The content approach is straightforward: write posts that answer real questions users have about AI image generation, make them genuinely useful, mention Pixova where it's naturally relevant.
What's working:
Comparison posts ("best free AI image generators") drive traffic with high commercial intent — these users are actively evaluating tools
How-to posts ("how to write better AI image prompts") drive educational traffic that converts to actual usage when users try the techniques
Conceptual posts ("what is text-to-image AI") bring in early-stage users still understanding the space — lower conversion but higher volume What I'm not doing:
Publishing thin content just to hit a posting schedule
Writing about topics I don't actually know well just because they have search volume
Keyword-stuffing posts that are painful to read The bar I use: would this post be useful even if it never mentioned Pixova at all? If yes, it's probably worth writing.
Infrastructure — Month 2 Changes
A few things changed under the hood that improved the experience without changing the surface:
Improved error messaging. Generic "something went wrong" errors got replaced with more specific, actionable messages. Users now know whether the issue is a prompt problem, a temporary load issue, or something else. Support-style messages dropped.
Better aspect ratio handling. The 9:16 portrait ratio had subtle composition issues on some prompts — the model was treating it as a cropped landscape rather than a purpose-built portrait composition. Adding specific compositional guidance for portrait ratio in the underlying prompt handling fixed this.
Faster download. The download flow had an unnecessary step that added 1-2 seconds between clicking download and the file arriving. Removed it. Small thing, noticeable improvement.
What Month 3 Looks Like
The priority is content. More blog posts, more educational material, more coverage of specific use cases (AI images for social media, for print, for commercial projects, for different creative styles).
The tool itself is in a good place — generating well, fast enough, no account friction. The gap between "tool working well" and "many people knowing about it" is a content and distribution problem, not a product problem.
Secondary priority: backlink building. Directory submissions, platform posts on Dev.to and Hashnode and Medium, getting the tool in front of communities who would find it useful.
The boring answer about what drives SEO growth is usually the right one: consistent useful content plus time.
Thanks for following along. Month 3 update will have more concrete traffic data as the content compounds.
Questions about specific decisions or numbers? Happy to answer in the comments.
