The useful AI subscription-ad workflow is approved product → versioned recurring-offer record → textless visual candidate → deterministic terms → product-page/cart/checkout audit → mature revenue readout. AI can explore the scene. It should not decide what gets charged, how often it ships, whether it renews, or how the customer cancels.
We ran that workflow on a fictional NORTHLINE CEDAR candle. The controlled offer is one 8 oz candle for $27.00 every 30 days, compared with a fictional $30.00 one-time purchase. Billing and delivery both recur every 30 days until canceled. No Shopify selling plan, checkout, ad account, or customer is connected, so the release state remains blocked.
Evidence boundary: NORTHLINE, the SKU, prices, discount, selling plan, policy, and every channel state are fictional controlled records. The built-in style-board run, live Masonry background job, source-preserving product placement, deterministic term layer, and consistency audit are real. This article does not report subscription performance, establish legal compliance, or recommend a universal default purchase option.
Why this is a distinct merchant job
Shopify currently defines a subscription as a recurring purchase option at an agreed price and scheduled frequency. Its sales reports can be filtered by the Subscription or one-time dimension. Read Shopify's current subscription overview.
The implementation contract crosses surfaces. Shopify's selling-plan documentation says merchant-facing selling-plan strings appear in admin, on the storefront product page, and at checkout. Shopify's checkout guidance also documents delivery frequency, recurring-shipment information, and a purchase agreement for subscription orders. Read the selling-plan field guidance and checkout considerations.
The merchant pain is qualitative but direct. One Shopify seller reporting recurring “I didn't know it was a subscription” disputes asked how to reduce chargebacks; replies questioned whether both one-time and subscription options were visible and whether the recurring charge was clear. Unverified claims in that thread are not benchmarks and are not repeated here. Read the merchant discussion.
Another merchant asked whether showing subscriptions to first-time ad visitors depressed conversion or pushed add-to-cart too far down the page. Replies described different product and customer contexts rather than one universal answer. Read the first-purchase discussion.
The supplied three-month Search Console export contains no non-brand subscription ad creative, subscription product page, subscribe and save image, recurring offer creative, or Shopify subscription creative query row. This is an adjacent revenue-intent expansion from Masonry's product-photography and ad-creative cluster—not proof of existing search demand.
Step 1: freeze the recurring-offer record
Do not begin with “make subscribe-and-save feel more compelling.” Begin with a record that can be rejected.
| Field | Fictional controlled value |
|---|---|
| Offer / SKU | NL-CEDAR-SUB-30D-V1 / NL-CEDAR-8OZ |
| Sold item | one NORTHLINE CEDAR 8 OZ candle with black lid |
| Market / currency | US / USD |
| One-time alternative | $30.00 |
| Subscription price | $27.00 per delivery |
| Discount basis | 10% below the controlled one-time price |
| Billing / delivery | every 30 days / every 30 days |
| Renewal | renews every 30 days until canceled |
| Cancellation path | customer account before the next scheduled charge |
| Shipping / tax | calculated separately where applicable |
| Legal / implementation status | merchant review pending; no live selling plan or checkout; release blocked |
Download the complete offer and asset manifest. It separates product, price, schedule, renewal, cancellation, source, visual, generation, implementation, release, and measurement authority.
Do not collapse billing and delivery into one field. A prepaid plan can charge on one schedule and deliver on another. Do not render “cancel anytime” unless the actual policy, cutoff, channel, and market review authorize those exact words.
Step 2: give each input one job
The approved source is the only product authority:
The textless board carries only art direction:
The offer record—not either image—controls $27.00, $30.00, 10%, every 30 days, renewal, cancellation, and release status.
Step 3: generate the background without the product or terms
We used the style board as the only reference for a live Nano Banana 2 background job:
masonry image "Create one textless 4:5 ecommerce product-ad background plate from the reference's warm ivory paper, pale stone, muted amber glass, aubergine accent, soft directional daylight, and quiet evidence-board mood. Build a premium studio composition with a clean central product zone and a separate empty lower offer zone. The reference controls visual direction only. Do not include a product, package, bottle, jar, candle, box, words, letters, numbers, price, currency, percentage, discount, subscription, badge, CTA, UI, shopping cart, credit card, legal claim, logo, watermark, or person." \ --model gemini-3.1-flash-image-preview \ --aspect 4:5 \ --seed 2608158 \ --ref ./subscription-style-board.webp
Job ef60f4c8-80aa-49f4-a920-d2b71b6410f9 succeeded in 10.128 seconds and returned a 928 × 1152 file.
This route deliberately cannot damage the SKU or fabricate a recurring promise because neither enters generation.
Step 4: composite product truth and term truth deterministically
The final controlled creative places the approved 1254 × 1254 product source without generative redraw. The price and cadence are HTML-rendered from the manifest:
This is still not a publishable ad. The cancellation path and other qualifiers can sit in primary text, a landing-page disclosure, or another channel-specific location only after the merchant's legal and channel review. The creative cannot substitute for the live purchase path.
Step 5: audit the promise after the click
Download the five-surface consistency audit. Its rows are controlled expected states, not observations from a live store.
| Surface | Required controlled check |
|---|---|
| Ad | exact SKU, purchase type, price basis, cadence, safe crop, approved qualifiers |
| Product page | one-time and subscription choices, selected plan, price, billing, delivery, policy link |
| Cart | subscription line item survives; price, frequency, quantity, and policy remain visible |
| Checkout | selected plan, recurring total, applicable shipping/tax, frequency, agreement, policy |
| Confirmation | next charge/delivery dates, selected plan, merchant identity, direct management path |
| Account | customer can inspect and perform the management actions the policy promises |
Shopify's current guidance says the free Shopify Subscriptions app adds a purchase-options cancellation policy and customer-management blocks, but themes, apps, gateways, shipping profiles, markets, and plan-level checkout editing can alter the actual path. Read the current setup guidance.
Do not claim that a mockup “complies.” The FTC's current Negative Option Rule page says the agency sought public comment in March 2026 about recurring payments consumers did not intend and cancellation obstacles. That changing regulatory surface is a reason for market-specific legal review, not a checklist this article can certify. Read the current FTC rule page.
Step 6: use Shopify fields as the implementation handoff
For a custom selling-plan implementation, Shopify's current developer guidance says the app populates merchant-facing strings such as the selling-plan name, and recommends patterns that include subscription type and delivery frequency. A product can also be marked as requiring a selling plan. Do not hard-code a second set of contradictory terms into generated media or theme copy.
The handoff should map:
offer_id → product / variant / requires-selling-plan state → selling-plan group and selected selling plan → billing policy / delivery policy / pricing policy → product-page widget → cart line item → checkout recurring summary and agreement → confirmation and customer-account management → channel asset version
Verify the live store on desktop and mobile. Test the one-time selection, each subscription interval, direct product URL, cart edit, accelerated checkout, shipping method, discount interaction, checkout wording, policy link, confirmation, customer account, email, and cancellation route. If any surface changes the product, price, frequency, or purchase type, block the asset rather than fixing only its headline.
The AI Black Friday creative workflow uses the same separation for a time-bounded promotion. The one-SKU ad creative matrix defines the broader controlled test, while the return-informed PDP workflow routes downstream expectation failures without pretending creative caused them.
For a non-recurring checkout that goes unfinished, use the AI abandoned-cart email workflow to preserve the selected SKU, variant, quantity, price, eligibility, and recovery destination through the reminder.
Step 7: measure revenue quality, not subscription clicks
Choose the commercial decision before launch. A practical near-term primary outcome is completed subscription purchase or contribution after the offer's real acquisition cost. Use click-through, product-page view, purchase-option selection, add-to-cart, checkout, and payment failure as diagnostics.
Then wait for the outcomes that cannot exist on day one:
eligible ad exposure → product-page visit → one-time or subscription selection → completed purchase → first fulfillment → next billing opportunity reached → renewed / skipped / paused / canceled / refunded / disputed → contribution after service and fulfillment cost
Segment one-time and subscription orders rather than blending them. Compare cohorts only after they have reached the same billing opportunities. Keep support contacts, “didn't know it recurred” complaints, cancellation attempts, refunds, and chargebacks as guardrails. A higher initial subscription rate can be worse for the business if cancellations, service cost, disputes, or contribution deteriorate.
Hold product, offer record, destination, audience, attribution, optimization event, budget treatment, and test window fixed when comparing visual cells. Do not interpret fewer than 20 eligible people or fewer than five downstream conversions as a stable rate. Do not call a creative the cause of retention or churn without a design that can support that claim.
The valuable output is not “a subscribe-and-save image.” It is one source-preserved product, one versioned recurring promise, one reviewable visual layer, one verified buying path, and one mature revenue decision.
