OptinMonster

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the email capture I actually need instead of OptinMonster: a well-aimed offer, tested, that does not make people hate the site.

STACK
- Node 20+ with Fastify
- SQLite through better-sqlite3, WAL mode
- One embeddable script under 10KB gzipped, from my own domain
- Caddy in front

BEFORE THE SPECIFICATION, THE THING THAT DECIDES WHETHER THIS WORKS
- A popup is an interruption. It converts because it interrupts, and it costs goodwill for the same reason
- So the whole design is about aiming: shown to the right person, at the right moment, once. A tool that shows everything to everybody converts worse and annoys more
- Never a popup on the first few seconds of a first visit, never one that covers the content on a phone, never one whose close button is hidden or delayed, never a second one after the first is dismissed
- Google penalises intrusive interstitials on mobile, which is a search ranking consequence on top of the human one
- Build the frequency rules before the templates and default them conservatively

THE DATA MODEL
- campaigns: id, name, kind, status, content_json, theme_json, position, priority, starts_at, ends_at
- variants: id, campaign_id, name, content_json, weight — for split testing
- rules: id, campaign_id, kind, operator, value — page, referrer, country, device, new or returning, UTM, scroll depth, time on page, exit intent, cart value
- frequency: id, campaign_id, max_per_session, days_after_dismiss, days_after_convert, global_cooldown_days
- impressions: id, campaign_id, variant_id, session_hash, shown_at, dismissed_at, converted_at
- conversions: id, impression_id, email, name, fields_json, created_at
- subscribers: id, email, token, confirmed_at, unsubscribed_at, source_campaign_id, consent_text, consent_at, ip_hash
- Every impression recorded. A conversion rate without a denominator is a marketing claim

KINDS
- A bar at the top or bottom that pushes the page rather than covering it — the least intrusive and often the best value
- An inline block within the content, which converts well and costs nothing in goodwill
- A slide-in from a corner, triggered by scroll depth
- A modal, which converts best and annoys most. Offer it, with the trade written next to it in the interface

TRIGGERS
- Scroll depth as a percentage, time on page, a number of pages in the session, a click on a chosen element, and exit intent
- Exit intent on desktop is the cursor leaving through the top of the viewport. On a phone there is no cursor, so it is a fast upward scroll or a back gesture — both are guesses, and both misfire. Say so in the interface, and default mobile exit intent to off
- Trigger on a custom event my own code sends, which is how 'they added to the basket then stopped' becomes a rule

TARGETING
- By page pattern, referrer, country, device, new or returning visitor, UTM parameter, day and time
- Rules read as sentences in the admin, with a preview of how many recent sessions would have matched
- One campaign shows at a time, chosen by priority, and the admin flags colliding rules before saving

THE FORM
- Email, and nothing else unless there is a real reason. Every additional field costs conversions
- A honeypot and a minimum time on form
- Double opt-in available and off by default, with the consequence of each stated
- The consent sentence stored with its timestamp, exactly as the person saw it
- One-click unsubscribe that needs no login, from every email that follows
- Success replaces the form in place; a redirect is optional

TESTING PROPERLY
- Two or more variants on one campaign, split by a hash of the session so a visitor always sees the same one
- The split, the sample size and the confidence shown together, and no winner declared before the sample size set in advance. Calling a winner at forty conversions is how a tool convinces you of something untrue
- One test at a time per page, or the results mean nothing

MEASURING
- Impressions, dismissals, conversions, conversion rate per campaign and per variant
- Bots excluded from published figures and recorded with a flag
- The number that matters is subscribers who stay subscribed and open something later — track it, because a popup that captures a thousand addresses which never open an email has achieved nothing

PRIVACY
- No cookie beyond a session-scoped identifier, no cross-site tracking, no third party
- Country from a local database file
- Addresses exportable and deletable by address

OPERATIONS
- .env: DATABASE_PATH, BASE_URL, HASH_SALT, SESSION_SECRET, SMTP_URL, ALLOWED_ORIGINS
- Migrations on boot, each once
- The payload endpoint answers from memory with an ETag; it is hit on every page view
- Nightly backup off the machine, restore script
- Health endpoint

WHAT MATTERS MOST
Frequency rules and honouring a dismissal. Build those first and test on a real site for a fortnight, watching bounce rate as well as conversions. The tool that converts best over a year is the one people do not resent.

Give me the repository, the embed script, migrations, .env.example, a seed with a bar and a modal, and a README with deploy steps behind Caddy.