Section 4 · You asked: "a plan to reach our goals, with reasoning"

The plan

Goal 1: recommended by ChatGPT and Claude · Goal 2: App Store discovery and conversions. The Blocky transition is section 6.

Short answer — why this order
  1. Measure first (0). Without a baseline, no later result can be credited to anything.
  2. Then fix what's true about you (1–3). Sending AI or App Store traffic to a confusing listing and a risky product wastes it.
  3. Then make sure you can be found (4).
  4. Last, the slow reputation work (5–7). It takes months and only pays off once the first steps are in place.

The plan on one page

#ActionGoalWhenEffortWeeks 1 → 13 (~90 days)
0Turn on tracking1 + 2 (prerequisite)Week 1~3 h
1One product, one name2, then 1Decide wk 1; live wks 4–6S
2Safe build output by default1 + 2Weeks 1–4S–M
3Publish your own facts1Weeks 1–4S–M
4Get indexed1Week 1~2 h
5Fix Blocky's reviews2 (and 1)Weeks 2–12M
6Be where assistants read1Days 15–90M
75 honest comparison pages1Days 30–90M
+Optional: App Store ads2After step 0Budget

Orange tick on action 1 = the naming decision in week 1. S/M = small/medium effort.

How you'll know overall

Every month I re-ask the same 43 questions (3 runs, search on and off) and re-check App Store rank and reviews. Baseline, 16 Oct 2026:

0/408
category answers naming Blocky
4/408
naming Clone.Store
#45
Blocky for "theme sections"
4
reviews in the last 90 days

Each action, with reasoning

0

Turn on tracking

Week 1 · ~3 hYour developer + your account access
What
  • GA4 (with Measurement Protocol) on the App Store listing
  • Search Console + Bing Webmaster for both domains
  • Check PostHog receives events; UTMs on blockyapps.com links
  • "Where did you hear about us?" in onboarding; funnel events
Why
  • Listing → install conversion can't be measured today.
  • Claude's traffic is invisible (no referrer) — asking is the only way.
  • Bing Webmaster has an AI report showing when Copilot cites you — only the owner can open it.
How you'll know
  • Listing views → install clicks → installs → paid, weekly.
1

One product, one name

Decide wk 1 · live wks 4–6Founder + marketer
What
  • Rename the listing "Clone.Store: AI Store Builder", "formerly Blocky" in the description.
  • Lead with the real workflow; more screenshots + a short video (today: 4, the fewest of 16).
  • Say on both sites that Blocky is Clone.Store's Shopify app.
  • Within Shopify's listing rules.
Why
  • Clone.Store installs Blocky, but nothing public says so.
  • The split likely costs reviews: the 7 Mar 1★ "does not build anything with AI".
  • ChatGPT reviews the wrong app (3/3); a competitor tells your story.
How you'll know
  • ChatGPT stops describing Blocky Fraud Filter (baseline 3/3).
  • No new "where's the AI" reviews.
  • Listing → install rate before vs after.

Conditions: decisions 1–2, action 2 live first, and the 3 gates in section 6. If Shopify rejects the name, explain the link on your websites only.

2

Safe build output by default

Weeks 1–4Product team
What
  • Generate new images, never reuse the reference store's.
  • Never generate reviews, ratings, stats, test results or health claims — merchant facts only, marked placeholders otherwise.
  • Preselect an AI palette.
Why
  • My test build would be a liability for the merchant: Grüns's photos (incl. Shrek), "1,842 reviews", invented trial results. The FTC banned fake reviews in 2024.
  • It's what assistants warn about (47% / 21%), and they're unlikely to recommend a tool they frame as risky. My hypothesis.
  • May protect the Blocky listing from Shopify's "content copying" ban.
How you'll know
  • Monthly test build: 0 copied images, 0 invented claims (baseline: 1 of 1 build had both).
  • Legal-warning rate in AI answers falls.
3

Publish your own facts

Weeks 1–4Founder (sign-off) + marketer
What
  • Pricing, About and Terms pages (Terms incl. what can and can't be copied).
  • Fix "launched free" and the old "no cost" meta.
  • Align "Claude or ChatGPT", "store" vs "product pages".
  • Politely ask Dropmagic, with evidence, to update its rating and "no AI" claim.
Why
  • When ChatGPT reaches your site it's right 11 of 15 times; Claude 3 of 15.
  • The pages that answer "is it legit / legal / what does it cost" don't exist (404).
  • Dropmagic's 2.7★ is from Feb 2026; Blocky is 3.8★ today.
How you'll know
  • Correct answers on brand questions (baseline ChatGPT 11/15, Claude 3/15).
  • No "free" or "no cost" in price answers.
4

Get indexed

Week 1 · ~2 hYour developer
What
  • Sitemap + robots.txt on blockyapps.com (both 404 today).
  • Submit both sites to Bing Webmaster and Search Console.
  • Ping IndexNow on every deploy; link the orphaned blog from the homepage.
Why
  • Assistants only cite pages their search providers hold (Claude: Brave).
  • Brave lists 2 clone.store pages and 1 blockyapps.com page — likely why Claude misses Clone.Store 9 of 15 times.
  • Crawler access is already fine: this is submission work, not a rebuild.
How you'll know
  • All key pages indexed in Bing and Google.
  • Claude finds Clone.Store by name (baseline 6/15).
5

Fix Blocky's reviews

Weeks 2–12Developer + support
What
  • Check the cancel flow on a dev store: clear "Delete All Data" wording, confirmation, export. Then reply again on the 1 Jul review.
  • Ask for a review after the first successful build (Shopify's in-app prompt, no incentives).
  • Reply to every review; then check Built for Shopify eligibility.
Why
  • Review velocity is the widest gap: Blocky 4 in 90 days vs 57–286.
  • All top 5 for "ai store builder" have Built for Shopify.
  • With 18 ratings each review moves the score; the cancellation complaint shows up in AI answers (Claude 8/15).
  • Incentivised reviews get apps delisted.
How you'll know
  • Reviews per 90 days (baseline 4), rating (3.8).
  • Rank: "theme sections" from #45 to page 1 within 90 days.
6

Be where assistants read

Days 15–90, ongoingFounder or marketer
What
  • Pitch the 4 independent "best of" lists with a factual note and a demo.
  • Open G2 and Capterra profiles.
  • Answer on Reddit openly, as the team.
  • Creator videos say "Clone.Store" + "AI Shopify store builder" in titles.
Why
  • Claude cites "best of" lists: 0 of 25 mention you; 12 of 25 are vendors ranking themselves.
  • ChatGPT read Reddit in 125 of 129 searched answers; you have 0 threads, 1 comment.
  • Undisclosed posting breaks Reddit's rules — so openly.
How you'll know
  • Category mentions (baseline 0/408 and 4/408).
  • Lists that include you (baseline 0/25).
7

Write 5 honest comparison pages

Days 30–90Marketer writes, founder checks facts
What
  • "Clone.Store vs Shopify AI Store Builder vs Atlas vs Dropmagic"
  • "Build from a store you like: what you can and can't copy"
  • "Building from ChatGPT or Claude: Shopify's connector vs Clone.Store"
  • Real prices and limits, fair to competitors.
Why
  • No page on either site answers a comparison question today.
  • ChatGPT prefers first-party pages: Shopify docs, listings, vendor sites.
  • A fair page gets cited; a hit piece doesn't. Dropmagic shows how much one page gets used.
How you'll know
  • Mentions on comparison questions (baseline 0).
+

Optional: App Store ads

After step 0Marketer, your budget

Buy ad slots on "theme sections" and "page builder" while reviews catch up — competitors already do. Start only after tracking is live, so cost per install can be measured.

Don't spend on

llms.txtGoogle says Search ignores it; OpenAI and Anthropic don't say they use it.
Structured data "for AI"No evidence it gets you cited. Google retired the FAQ rich result in 2026.
Speed work to win AI answersNo assistant links speed to citation. Do it for conversion (funnel step 6).
Mass AI-written pagesNo evidence they help, and they bring the invented-facts risk into your own marketing.
Reddit seeding, paid or incentivised reviewsAgainst Reddit's and Shopify's rules.
Blocking AI crawlersRemoves you from training data and search answers.

Risks

Sources