4 Days, 62 Users, 28 Countries: What the First Week of Promptolis Taught Us

8 minute read · Published April 2026

Four days ago, Promptolis had 35 Originals, zero Google search impressions, and no traffic. Today: 350 Originals, 14 Packs, 22 blog articles, 62 active users from 28 countries, first referral backlink from a Beehiiv newsletter, first click from Google organic search, and two Originals ranking on page 1 of Google for niche queries.

This post is the honest debrief of the first week — what we tried, what worked, what failed, what surprised us. For anyone building a content library or prompt site in 2026.

The Setup (Where We Started)

Monday, April 18, 2026:

35 hand-crafted "Originals" (premium XML-structured prompts)
~1,600 community prompts from the open-source awesome-chatgpt-prompts corpus (MIT-licensed base)
18 categories (later expanded to 21)
Standard SEO: sitemap, schema.org, Open Graph — but no traction
New domain, no authority, no backlinks

The thesis: research-backed AI prompts at scale would differentiate from the competitors (AIPRM, PromptBase, PromptHero, FlowGPT) — all of which produce volume but not depth.

What Actually Happened (The Data)

GA4 April 18-22 (first 5 days):

62 active users (61 new, 1 returning)
321 events
31.9 sec average engagement duration
28 countries represented

Growth pattern (users per day):

Day 0 (Apr 18): 8 users
Day 1 (Apr 19): 14 users (+75%)
Day 2 (Apr 20): 13 users (-7%)
Day 3 (Apr 21): 18 users (+38%)
Day 4 (Apr 22): 16 users (-11%)

Not viral. Steady climb to ~15 users/day. Normal for a new content site — better than most (many new sites show 0 traffic for weeks).

Traffic sources:

Direct: 49 users (79%)
Google Organic: 13 users (21%) ← first week SEO already working
whattheai.beehiiv.com referral: 1 user ← first backlink!

Top pages (what people actually clicked):

Homepage variants: ~30 total visits
"25 AI Prompts for Content Creators" blog: 5 visits
"Complete Guide to Image Generation Prompts" blog: 4 visits
Creative Writing Prompts Pack: 1 visit (Pack format working)
Contact page: 2 visits, 0% bounce (someone wanted to reach us)

Geography (real humans across the world):

Boardman US, Aspen US, Ashburn US, Antalya TR, Council Bluffs US, Dublin IE, New York, Dubai AE, Mumbai IN, Delhi IN, Chicago, Santiago CL, Tashkent UZ, Pyramids Gardens EG, Shenzhen CN, Singapore... 28 total countries in 5 days.

What Worked (In Order of Magnitude)

1. Research-Backed Content Differentiation

Every Promptolis Original cites specific research (Pennebaker's expressive-writing research, Julia Cameron's morning pages, Jung's shadow framework, Robert Moore's archetypal work, Ursula K. Le Guin's cultural logic, Brandon Sanderson's Three Laws of Magic).

The cited sources aren't marketing fluff — they're the actual frameworks used to structure the prompts. When Claude or ChatGPT hold a prompt grounded in specific research, output quality differs measurably from generic prompts.

This differentiation shows up in the data: the blog article "The Research Behind Every Promptolis Original" got organic pickup and the first backlink came from a reader of that article.

2. The Pack Format

We launched 14 "Packs" — bundles of 15-30 research-backed prompts organized around specific practices (Journal Prompts, Shadow Work, Creative Writing, Short Story, Novel Plot, Horror Writing, Fantasy Writing, Morning Journaling, Self-Reflection Journal, Teen Journal, Poetry Prompts, Elementary Writing, Horror Cinematic Portrait, Instagram Content Ideas).

Each Pack is ~15,000-25,000 words of structured content. The Creative Writing Prompts Pack already appears in the top-pages-visited list after day 4.

The bet: Packs would serve practice-building (return visits) while single Originals serve one-off utility. Early signal validates this — users are discovering Packs via blog articles and category pages.

3. Writing/Journaling Niche Focus

We initially planned to scale with image-prompt packs (Midjourney, Stable Diffusion). Then we did the keyword research and discovered: writing and journaling prompts have 30-50× the search volume of image-prompt keywords with lower competition.

Pivoting to writing/journaling as primary (while keeping image-prompts as a secondary brand-play) was the single biggest strategic decision. The Creative Writing Pack, Journal Prompts Pack, and Shadow Work Pack are the Packs getting traction now.

Lesson: Don't trust your initial assumptions about what volume exists. Use actual keyword data. We almost built for the wrong market.

4. Topical Clustering

Rather than one Writing Prompts Pack, we built Writing-Prompts cluster: Creative Writing, Short Story, Novel Plot, Horror Writing, Fantasy Writing, Poetry. Each specific, each research-backed, each linking to siblings.

Google's algorithm rewards topical authority. Six specific writing-prompt Packs + 4 related blog articles = authority signal. Single generic "Writing Prompts" Pack would be diluted.

5. AI-Referral Traffic Is Real

Multiple visits from Boardman, OR and Ashburn, VA — AWS data centers. These are AI crawlers (ChatGPT, Claude, Perplexity, Gemini) indexing our site. That's not "wasted traffic" — that's AI systems deciding we're worth citing.

We added a comprehensive llms.txt file (220+ lines) with FAQ entries, citation guidance, and research attribution. Early signal: AI referrals appearing in GA4 as "chatgpt.com / referral" and "perplexity.ai / referral" on a sister project (seoscore.tools) validates this channel.

In 2026, if you're building for AI-visibility alongside human visibility, you're ahead of 95% of content sites.

What Failed (The Honest Part)

1. Image Prompt Packs Were Overestimated

Our keyword research showed "midjourney prompts" at 5K monthly searches, "stable diffusion prompts" at 500 — good numbers in isolation. But most "midjourney prompts" searchers want TOOLS (generators), not PACKS (libraries).

Our Horror Cinematic Portrait Pack sits at 0 visits in first 5 days. The Instagram Content Ideas Pack we just built is unlikely to hit 1,000 visits/month anytime soon.

Correction: Image-prompt Packs are brand-play, not traffic drivers. We kept the Horror Cinematic Pack + added Instagram Content Ideas (because the latter has 35K+ addressable monthly search volume in "instagram post ideas" territory). But we stopped building image-aesthetic packs as primary strategy.

2. Broken Internal Links Were Worse Than We Thought

Audit revealed 9,424 broken internal link references across 3,785 HTML pages. Root causes:

Blog articles linking to non-existent prompt slugs (we'd written about prompts that didn't exist)
Nav links to /de/originals/ which was never generated
Footer links to /privacy/ and /licenses/ which didn't exist
Missing apple-touch-icon.png referenced everywhere

We found all of this in a user session where they kept clicking broken links. If we'd launched with these problems into more traffic, we'd have damaged user trust permanently.

Fix: Full systematic audit + generator-level fixes (not just patching HTML). Now: 0 real broken links. Future regenerations stay clean.

Lesson: Run internal-link audits routinely on content sites. Especially important when you're shipping at velocity — broken links compound.

3. Bounce Rate Is High

Initial bounce rate: 75-100% on most pages. Users land, look at one page, leave. Normal for new SEO-driven sites but signals: landing pages don't give enough reason to stay.

In-progress fix: Email capture at end of every blog article (via Beehiiv integration). Converts visitors into subscribers who receive weekly prompts. Compound asset vs. single-visit traffic.

4. Germany Traffic Is Negligible

We built the entire site bilingually (EN/DE), expecting substantial German market. First 5 days: 1 user from Germany. Massive over-investment in DE translation when audience is overwhelmingly US + global.

Decision: Keep DE translations (already built), don't add Spanish/Russian/Turkish like we considered. Focus DE-specific content expansion only after we see organic German search demand.

What We're Doing Differently Now

1. Blog Articles Drive Pack Discovery

Pattern: person searches "journal prompts for teens," lands on our Teen Journal Pack (Pack page), or clicks through from "25 AI Prompts for Parents" blog article → Pack page.

We're writing 2-3 blog articles per week, each driving to 1-3 Packs. Articles are research-backed, specific, and link intentionally. Not "10 marketing tips" content — specific insights for specific audiences.

2. AI Citations as a Primary Channel

Upgraded llms.txt to 220+ lines with FAQ format, citation guidance, and research attribution. The bet: if we make it easy for AI systems to cite us correctly, they will.

Early signal (from sister project seoscore.tools): 14% of traffic now comes from AI referrals (ChatGPT, Claude, Perplexity). If Promptolis achieves similar, that's substantial new channel in 2026.

3. Email Capture at End of Every Blog Article

Via Beehiiv (free tier up to 2,500 subscribers). Simple form: "One research-backed AI prompt per week. Free. Unsubscribe anytime." No popups, no sales funnels, no 17-step onboarding.

Goal: convert 5% of blog readers to email subscribers. Over 6 months at current traffic pace, that's ~50-200 subscribers. Compound asset.

4. Don't Scale Blindly

We could keep adding Packs. But the data shows: current Packs aren't yet in heavy traffic. Before building Pack #15, we're waiting 2-3 weeks to see which existing Packs actually attract users, then double-down on winners.

Lesson: Build, deploy, measure, iterate. Not build, deploy, build more. The second mode is how content libraries die — full of untouched pages.

What This Means for You (If You're Building Similar)

1. Keyword Research Before Content Building

We almost wasted 2 weeks on image-prompt packs. Actual keyword data (via Google Keyword Planner) showed writing/journaling prompts were 30-50× bigger market. That research took 2 hours and saved 2 weeks.

Don't build content based on what you think people search for. Build based on what they actually search for.

2. Research Backing = Differentiation

Most prompt sites are generic ("write like Shakespeare," "act as a marketing expert"). We cite specific research — that's our moat. Every Original cites Pennebaker, Cameron, Jung, Saunders, O'Connor, Le Guin, Sanderson, or other specific source.

If you're in competitive content territory (and prompt libraries are), research backing is one of the few moats that scale.

3. Internal Link Hygiene

9,424 broken links found in audit. If you're building at velocity, audit monthly. Broken links compound: each broken link reduces SEO + user trust. We're now running audits before every deploy.

4. AI-Referral Is the 2026 Channel

llms.txt is the most underrated SEO file in 2026. Most sites don't have one. Most that do are basic. A comprehensive llms.txt (FAQ format, citation guidance, author credentials) differentiates you to AI systems that are increasingly driving traffic.

14% of seoscore.tools traffic from AI referrals in 2026 = substantial. Will be bigger in 2027.

5. Don't Over-Translate

We built Promptolis bilingual (EN + DE) from day 1. First 5 days: 1 DE user. If we could redo: start EN-only, translate after validating demand.

Every additional language is 2× the content + 2× the maintenance. Only add when you've verified demand in that language.

What's Next for Promptolis

Short-term (next 2-4 weeks):

GSC data matures → identify which Packs + blog articles actually rank
Email subscribers via Beehiiv → compound audience
2-3 blog articles per week → topical authority for writing/journaling cluster
Monitor AI referrals (ChatGPT, Claude, Perplexity) → validate AI-visibility channel

Medium-term (1-3 months):

Winners get deepened (more Packs on themes that rank)
Losers get reconsidered (Packs that don't rank after 60 days get revised or removed)
If email list hits 500 subscribers → first monetization test (premium Pack tier? sponsored prompts? course?)

Long-term (6-12 months):

500-1,000 Originals total
25-50 Packs total
Newsletter at 2,000+ subscribers (Beehiiv maxes out their free tier)
Potentially: paid products (premium Packs, bundled collections)

The Honest Caveat

5 days is not data. It's signal. The patterns above (62 users, 28 countries, steady growth, AI indexing) may or may not sustain. Most sites that show strong week-1 traction plateau at week 4-8 without continued effort.

We're not declaring victory. We're reporting what the first week actually showed — including the failures, the broken links, the over-translation, the image-prompt misdirection.

In 30 days, we'll write an updated post. Either the traction compounded (most likely), plateaued (possible), or died (unlikely but possible). Real startup data, not performance theater.

Browse Promptolis

350 Originals + 14 Packs + 22 blog articles. All free. All research-backed. All MIT-licensed.

---

Questions, corrections, or your own startup debrief? Email contact@promptolis.com. We read everything.