The introductory post on Cycle AI covers how the feature works at a structural level: what data it reads, how it maps a description to wizard fields, and what the draft looks like when it loads. This post is the follow-up for merchants who are actually using it and want to get more out of it.

The gap between a prompt that produces a draft you can launch in five minutes and one that produces a campaign you have to rebuild from scratch is usually not the AI’s fault. It’s prompt structure. Cycle AI is working from the text you give it plus your store data. Give it a direction and it makes confident choices. Give it nothing to work with and it defaults to the broadest interpretation of your store, which may or may not be what you wanted.

This post covers what makes prompts work, where the quota sits, and what the AI still leaves for you to handle — so your expectations match what Cycle AI actually does.

Prompts that produce tight drafts

Cycle AI reads your prompt, your catalog, and your store intelligence together. A prompt that references concrete, verifiable things from your catalog — category names, product traits, timing constraints — gives the AI enough signal to make precise choices. Here is what that looks like in practice.

Name a category or product type

If your prompt names a category that exists in your store, Cycle AI will target it. “20% off all jackets this weekend” will produce a campaign targeting your Jackets category, with a percentage discount, running Friday through Sunday. The AI uses the actual category IDs from your catalog, not a guess. If the category name does not exist in your store, it will tell you in the reasoning text before the campaign loads.

Be specific about timing

Cycle AI resolves relative dates correctly. “This Friday” resolves to the correct calendar date. “For 10 days” calculates the end date from today or from the start date you specify. “6pm to midnight” converts to 18:00–00:00 in your store timezone. The more specific your timing, the less the AI has to assume.

Prompts without any time reference produce campaigns that start immediately and run for a default duration. That is fine for spontaneous promotions, but if you have a specific event in mind — a long weekend, a product launch, a holiday — say so explicitly. The AI reads upcoming seasonal events for your store’s region, but it will not apply them unless your prompt makes the connection.

Name a discount mechanism when you have one in mind

Cycle AI picks the discount type that best fits the goal. Clearance language maps to percentage. “Buy more, save more” maps to tiered pricing (Pro). “Spend $100 get $15 off” maps to spend threshold (Pro). But if you have a specific type in mind — BOGO, fixed amount, bundle — naming it produces a more accurate draft. “BOGO on all running shoes” will use the buy-one-get-one discount type. “Buy any jacket, get matching gloves at 50% off” maps to cross-product buy-X-get-Y.

Use store intelligence signals in your language

Cycle AI knows your bestsellers, slow movers, new arrivals, and low-stock items. Your prompt can reference those signals directly. “Clear out my slow-moving products” produces a campaign targeting smart selection with the low-stock or slow-mover criterion. “Flash sale on my bestsellers” targets smart selection with best sellers. You do not need to list product names — the AI picks from the relevant set automatically.

Prompts that produce vague drafts — and why

Some inputs produce campaigns that are technically valid but not particularly useful. Understanding why helps you either fix the prompt or know to expect a draft that needs more editing.

The single-word or single-phrase prompt

“Run a sale” will produce a storewide percentage discount starting today with a default duration. It is not wrong — it is the safest interpretation of an undirected request. But it is unlikely to be what you wanted. Storewide discounts with no time constraint and a default discount value are where you end up when the AI has nothing to work with.

If you find yourself editing almost everything in the wizard after a vague prompt, that is not an AI limitation you need to work around — it is a signal to add a bit more to the prompt. Even adding a category name and a duration will produce a meaningfully tighter draft.

Prompts that reference things not in your catalog

If you mention a category that does not exist in your store, the AI will note this in the reasoning text. It may fall back to all products, or it may ask you to clarify. It will not silently target a wrong category — but it also cannot invent products or categories that are not there. If your catalog uses non-obvious category names, use the exact names when you prompt.

Pro-type requests on the free plan

If you ask for tiered pricing, a spend threshold, or a bundle discount on the free plan, the AI will recognize the intent, name the right discount type and explain why it fits, and then build a percentage-discount version you can save today. The reasoning text explains what was substituted and why. This is correct behavior, not a failure. The substitute campaign is usable — it is just not the optimal discount structure for your goal.

Requests for settings the AI cannot emit

Some merchants ask Cycle AI to set usage limits, configure discount stacking, or style the promotional badge. The AI will explain that these settings are not something it can emit in the campaign draft, and direct you to the relevant wizard step. It does not silently ignore the request. The full list of manual-configuration settings is covered below.

The quota: free vs. Pro

Cycle AI is available on both the free and Pro plans of Smart Cycle Discounts. The quota controls how many campaign drafts you can generate per billing period. The quota is enforced server-side by the Webstepper proxy — it is not a soft limit that can be exceeded by timing requests differently. After each successful AI response, the remaining count is displayed in the panel.

For most store owners on the free plan, 10 drafts a month is enough. A typical workflow involves one or two prompts to get a campaign you are happy with, then refinements within the same wizard session. If you are managing several campaigns concurrently or running a high volume of seasonal promotions, Pro’s 50-per-day limit gives considerably more room.

Refinements within an existing wizard session do count against the quota. Each time you send a refinement instruction in the panel, it uses one request. If you are close to your free-plan limit, direct edits to the wizard steps are quota-free and often just as fast for single-field changes.

When the quota is exhausted, the panel will display the reset time. The proxy enforces per-site quotas server-side and the plugin displays whatever the server returns.

What the AI fills in for you

When Cycle AI produces a full campaign draft, it populates all five wizard steps with values it has chosen based on your prompt and store data. Here is a complete accounting of what it fills, with the logic it uses for each field.

Basic information

• Campaign name: Human-readable and merchant-friendly. Names like “Winter Jacket Clearance” or “Weekend Accessories Flash” rather than coded identifiers. Edit freely if it does not match your naming conventions.
• Description: A one-to-two sentence merchant-facing description of the promotion.
• Priority: Defaults to a mid-range value. The AI does not analyze your existing campaigns to pick an optimal priority. It sets a reasonable default. Review this if you are running concurrent campaigns on overlapping products.

Product scope

• Selection mode: All products (storewide or filtered by category and tag), specific product IDs, smart selection (best sellers, featured, low stock, new arrivals), or random products. The mode is chosen based on how you described the scope in your prompt.
• Category and tag targeting: Uses real IDs from your catalog. If you name a category, it targets that category. If you name a tag, it filters by that tag. Both can be combined.
• Specific products: When you name individual products, the AI resolves product IDs from your catalog. Only products that exist in your store will be targeted.
• Smart selection criteria: When you say “bestsellers,” “featured,” “low stock,” or “new arrivals,” the AI sets the smart selection mode with the matching criterion.
• User role targeting: Defaults to all users. Only applies targeting when you explicitly ask for it (“wholesale customers only,” “exclude subscribers”). Uses the actual role slugs registered in your store.
• Location targeting: Defaults to all countries. Only applies when you ask (“only in Germany,” “only in my country”). Uses ISO country codes; “my country” resolves to your store’s base country.

Discount configuration

• Discount type: Chosen based on your prompt and plan. Percentage, fixed, BOGO (same-product), and buy-X-get-Y are available on free. Tiered, spend threshold, and bundle require Pro. On the free plan, Pro-type requests produce a percentage-discount substitute with an explanation in the reasoning text.
• Discount value: A round number within the valid range for the type: 5–60% for percentage, $1–$500 for fixed. Based on promotion logic, not your margins. Always verify against your actual costs before launching.
• BOGO and buy-X-get-Y configuration: Buy quantity, get quantity, discount percentage on the “get” side, and (for cross-product campaigns) which products are on the buy and get sides.
• Tiered tiers (Pro): Quantity breakpoints and discount values for each tier, based on your request.
• Spend thresholds (Pro): Spend amounts and corresponding discount values.
• Free shipping: Off by default. Enabled only when you explicitly ask for it alongside the discount.

Schedule

• Start date and end date: Calculated from your prompt. “This Friday for 10 days” produces the correct start and end dates in your store’s timezone.
• Start time and end time: Defaults to 00:00–23:59 (all day). Set explicitly when you name a time (“6pm to midnight,” “starts at noon”).
• Recurring schedule: Disabled by default. Enabled only when you explicitly ask for it. “Every Friday,” “repeat weekly,” “monthly clearance” all trigger recurrence. The pattern, specific days, and end condition (never, after N cycles, on a date) are filled based on your description. For more on the recurrence options available, see the recurring campaigns documentation.
• Timezone: Uses your store’s configured timezone. Emits a different timezone only when you explicitly name one (“in Europe/Berlin time”).

What you still configure manually

Cycle AI fills in the core campaign fields, but there is a defined set of settings it cannot emit. The AI will tell you about these when you ask for them — it will not silently skip the request. After the campaign loads in the wizard, you configure these in the relevant steps before saving.

Usage limits

Per-customer usage caps and lifetime campaign caps are not emitted by Cycle AI. These are Pro settings configured in the Discounts step after the draft loads. If you ask Cycle AI to “limit to 100 uses” or “one use per customer,” it will acknowledge the request, explain that it is a Pro setting it cannot emit directly, and tell you where to find it in the wizard.

Combination policies

Whether the campaign stacks with other active campaigns, blocks WooCommerce coupons, or excludes products already on sale — these are also Pro settings in the Discounts step. The AI sets sensible defaults (stacking off, coupons allowed, sale items included), but it cannot accept instructions to change them in the draft. Configure these in the wizard after generation.

Promotional visuals

Badges, threshold bars, BOGO panels, and other promotional display elements are assigned after generation via the Visuals picker in the Discounts step. Cycle AI does not style campaigns visually. If you ask for “a badge that says 20% off” or “show a progress bar,” the AI will explain that this is handled in the wizard post-generation.

Specific shipping methods for free shipping

When the AI enables free shipping on a campaign because you asked for it, it applies the free shipping to all shipping methods by default. If you want to limit it to specific methods — flat rate only, excluding express — configure that in the Discounts step after the draft loads.

Subscription discount controls

Advanced subscription settings — targeting the recurring price vs. sign-up fee, limiting discounts to the first N renewals — are Pro settings and not emitted by Cycle AI. On the free plan, subscription products receive the campaign discount automatically. Pro subscription controls are found in the Discounts step.

How to iterate without starting over

When the first draft is not quite right, you have two options: refine it using the Cycle AI panel, or edit the wizard steps directly. Both are valid. The choice depends on how different the corrected campaign is from the draft.

Refinement through the panel

Refinement works differently from re-prompting. When you type a follow-up instruction in the panel after a campaign has loaded in the wizard, Cycle AI reads the current wizard state — all five steps as they are configured right now — and applies only the change you described. It does not go back to your original prompt. It sees the actual fields as they stand and modifies the relevant ones.

This makes refinement most useful for targeted, specific changes:

• “Lower the discount to 15%”
• “Extend the end date by three days”
• “Add Footwear to the product scope”
• “Make this recurring every Friday for six weeks”
• “Include free shipping with this one”
• “Target wholesale customers only”

Each refinement uses one quota request. If you are on the free plan and close to your 10-per-month limit, direct edits in the wizard steps are quota-free and often just as fast for single-field changes.

When to edit directly vs. refine through the panel

Direct wizard edits are faster for single-field changes you know how to make: adjusting a discount value, changing a date, switching a product selection mode. Refinement through the panel is better when multiple related fields need to change together. Converting a one-time campaign to a recurring weekly one, for example, changes the schedule pattern, the recurrence settings, and potentially the end date all at once. The AI handles those interdependencies correctly in a single instruction. You might not catch all of them if you edit manually.

When multiple refinements pull in different directions

If you have made several refinements that changed the core concept of the campaign significantly — different product scope, different discount type, completely different timing — the resulting configuration can be technically valid but logically inconsistent. When this happens, it is usually cleaner to discard the draft and start a new prompt with the updated concept rather than continuing to layer changes on a draft that no longer reflects a coherent idea.

When to ignore the draft entirely

Cycle AI is most useful when you have a direction but do not want to configure every field from scratch. There are situations where skipping the AI is the more efficient choice.

Duplicating an existing campaign

If you have run a campaign before and want to run it again with minor date changes, use the Duplicate action on the Campaigns list. This copies the exact configuration of the original. It is faster and more accurate than re-prompting, and it does not use quota. The campaign duplication workflow takes about 30 seconds.

Complex targeting logic

If your campaign involves combinations of custom user roles, specific product attribute conditions, or subscription renewal targeting, the wizard will be more precise than the AI. Cycle AI handles the common targeting patterns well — category, tag, bestsellers, role — but the more unusual your setup, the more the wizard’s explicit fields will serve you better.

When you already know exactly what you want

The wizard is a well-designed five-step flow. If you already know the discount type, the products, the value, and the schedule, going straight to the wizard is faster than prompting the AI and reviewing a draft. Cycle AI adds most of its value when you are working from a general intention rather than a fully-formed specification.

Frequently asked questions

How do I get Cycle AI to target a specific category?

Name the category in your prompt using the same name it has in your WooCommerce category settings. “20% off Jackets this weekend” will target the Jackets category if it exists in your store. If the name in your prompt does not match any category, Cycle AI will say so in the reasoning text before the campaign loads. It will not silently fall back to all products without telling you.

Do refinements use up my quota?

Yes. Each refinement instruction sent through the Cycle AI panel uses one request from your quota. The panel displays the remaining count after each successful request, including refinements. If you are close to your free-plan limit of 10 per month, direct edits in the wizard steps are quota-free and often just as fast for single-field changes.

Can I use Cycle AI to set a per-customer usage limit?

No. Usage limits are a Pro setting that Cycle AI cannot emit in the campaign draft. If you ask for one, the AI will acknowledge the request, explain that it is a Pro setting it cannot emit directly, and tell you to configure it in the Discounts step after the draft loads in the wizard. The base campaign JSON will be generated without the limit, ready for you to add it manually.

The AI generated a storewide sale when I wanted something more targeted. What happened?

When a prompt does not specify which products or categories to target, Cycle AI defaults to all products storewide. This is by design — it is the safest interpretation of an undirected request. To get a more targeted draft, add a category name, a product trait, or a store intelligence signal (“bestsellers,” “new arrivals,” “slow movers”) to your prompt. Even a single word of context changes what the AI targets.

Can I ask for tiered pricing on the free plan?

You can ask for it. Cycle AI will recognize that tiered pricing is the right fit for your goal, explain that it requires the Pro plan, and then generate a percentage-discount campaign you can save today. The reasoning text will name the Pro type explicitly and include an upgrade link. This is consistent behavior — the AI never silently outputs a campaign type your plan cannot save.

Does Cycle AI know about my existing campaigns?

Yes. The context sent to the AI service includes a summary of your active campaigns. Cycle AI uses this to avoid generating campaigns that obviously duplicate or conflict with what you already have running. That said, the conflict awareness is advisory — it works from a text summary, not a live scope analysis. Campaign Health in the wizard review step does the authoritative conflict check before you save. It is worth reading those results before launching.

What is the difference between the standard model (free) and the smarter model (Pro)?

Both models receive the same store context and campaign schema through Webstepper’s proxy. The Pro tier uses a more capable model that tends to produce tighter reasoning and handles nuanced or multi-part requests more reliably. For straightforward campaigns — a percentage discount on a category for a week — the standard model works well. For complex requests or campaigns that push the limits of what the schema supports, the smarter model generally produces a more accurate draft.