Covers every action in the plan.
| Code | Anything mechanical and repeatable, where a mistake is cheap and easy to spot: measuring, collecting data, submitting pages, triggering a request at the right moment. |
| AI-assisted | A model drafts or flags; a person approves before anything goes public. For writing and checking at volume. |
| Human — can't automate | Needs your account access, your decision, or a relationship. |
| Human — shouldn't automate | Speaks for you in public, makes a factual or legal claim, or would break a platform's rules if a bot did it. |
Why the line sits there: your own build shows what unchecked AI output does — it invented "1,842 reviews", "89% fell asleep faster" and lab tests. The same models will invent facts in marketing copy, so every public word needs a human check.
| Action | Code | AI drafts, person approves | Human only |
|---|---|---|---|
| 0 Tracking | Funnel events in the apps; daily Partner API pull (installs, uninstalls, charges) into PostHog; server-side install events to GA4; weekly funnel report. | None needed | Creating GA4, Search Console, Bing Webmaster; adding GA4 to the listing. Needs owner access. |
| 1 One name | None | Listing copy variants within Shopify's rules. | The rename decision, final copy, media, submitting to Shopify. |
| 2 Safe output | In the product: never reuse reference images; a validator blocks any number, review count, rating, test result or health claim the merchant didn't provide. Monthly: test builds compared with the reference store's images (perceptual hash). | A checker lists every claim in a test build not found in the merchant's input. | Deciding the defaults and what's acceptable for supplements and other regulated products. A prompt alone doesn't stop a model inventing numbers. |
| 3 Your facts | One facts file (price, trial, features, limits) that every page and the listing read from. | First drafts of Pricing, About, FAQ. | Terms and copy policy (legal), every published number, the Dropmagic request. |
| 4 Indexing | Sitemap on build; IndexNow ping on deploy; monthly indexed-pages check. | None needed | Verifying the sites. |
| 5 Reviews | Trigger Shopify's review prompt after the first successful build; alert on every new review. | Draft review replies. | Sending replies; checking the cancel flow. A reply is your public voice. |
| 6 Where assistants read | Monthly list of cited pages where competitors appear and you don't; weekly mention monitor (Reddit, YouTube, lists) with a strict brand filter*. | Summaries of new threads; suggested factual points. | Pitching lists, creator relationships, every Reddit post (openly, as the team). |
| 7 Comparison pages | Pull questions and competitor facts from the tracker into a brief. | Outline and first draft. | Fact-checking every competitor claim, editing, publishing. |
| Monthly measuring | AI visibility tracker and App Store rank tracker — already built. | Extracting mentions, positions, accuracy and sources from each answer. | Monthly review of what changed and what to do. |
* "Blocky" alone matches unrelated products (blocker apps, a DNS ad-blocker). Require "blockyapps", or "Blocky" with section/theme/Shopify and without fraud/bot terms.
Asks ChatGPT and Claude the 43 questions (search on/off, 3 runs), saves every answer and source, extracts mentions, position, accuracy against the facts file, legal warnings and cited sites, then reports. It produced every AI number in this review.
Reads public search results for 10 queries, 5 pages each, and logs Blocky's organic position and the top 3 apps. No cost.
Node 23.6+, no installs.
node ai-visibility.ts run → extract → report
node appstore-rank.ts
| Posting on Reddit or forums | Against Reddit's spam and manipulation rules; reads as astroturfing. |
| Writing or incentivising reviews | Shopify removes or delists for it; fake reviews are what the FTC rule targets. |
| Publishing AI copy unchecked | Models invent numbers, as your build showed. |
| Mass "best X" pages | No evidence it helps; assistants prefer listings and established lists. |
| Scraping the ChatGPT app | OpenAI's Terms forbid it. Use the API, as the tracker does. |
| Auto-replying to reviews | A wrong tone on a 1★ costs more than the time saved. |
| Terms and copy policy | Legal text: a lawyer or the founder signs off. |