CodeCanyon AI Commerce - WordPress Plugin

How to Set Up CodeCanyon AI Commerce for a WooCommerce Store

CodeCanyon AI Commerce is not just a small widget. It is a full set of AI tools for WooCommerce: semantic search, photo search, recommendations, product comparison, description generation, translation, sales analysis, order screening, and support reply drafts. This guide takes a practical approach to the plugin: what to check before installation, which features to enable first, how to avoid over-automating your store, and how to tell whether the outcome actually helps shoppers.

This guide is written for store owners, WordPress administrators, and specialists who already have a working WooCommerce catalog and want to add AI features without rebuilding the site. We are not going to repeat the product sales page. Instead, we will walk through the full path from data preparation to testing search, recommendations, content, forecasting, suspicious order checks, and customer message handling.

The main idea before installation is simple: AI Commerce is useful only when the store has data it can learn from and data you can verify. If products are filled out inconsistently, attributes are disorganized, shipping policies are empty, and there is little order history, some features will behave more like rough assistants than finished decision-making tools.

What This AI Toolset Is Designed to Solve

A typical WooCommerce store often grows in layers. First the owner adds products, then categories, then filters, search, reviews, email, support, analytics, and fraud protection. After a while, a familiar problem appears: shoppers struggle to phrase what they want, admins struggle to maintain product pages, and owners end up making decisions by instinct because the numbers never turn into clear action.

AI Commerce is designed to address several of those weak spots at once. For shoppers, it creates a smarter path to the right product through natural-language search, photo search, recommendations, comparison, and a chat assistant. For catalog managers, it offers description generation, pricing suggestions, automatic categories, tags, translation, and photo processing. For admins, it adds demand forecasting, business insights, new product moderation, order risk scoring, and support reply drafts.

This matters before setup: the plugin does not replace inventory management, a payment gateway, CRM, legal review of content, or a full fraud prevention team. It works as an application layer on top of WooCommerce and related services. The right strategy is to enable features one at a time, test the quality of the outcome on real products, and keep manual control wherever mistakes can cost money or damage trust.

Three Feature Groups

In the AI Commerce documentation, the features are easiest to understand in three layers. The first layer is shopper-facing: search, photo search, recommendations, review summaries, the chat assistant, and product comparison. The second layer supports store operations: descriptions, images, pricing, categories, tags, translation, and forecasting. The third layer is administrative: moderation, order risk, business insights, trending products, and support replies.

That breakdown helps you avoid turning everything on at once. For example, a small clothing store can reasonably start with semantic search, photo search, and recommendations. A store with many suppliers will care more about product moderation and automatic category assignment. A store with a high volume of repetitive inquiries will benefit more from the chat assistant and reply drafts. A store with unstable demand will get more value from sales forecasting and trending products.

Who It Fits and Who It May Not

The plugin works best for WooCommerce stores that already have at least a few dozen products, clear categories, variations, reviews, orders, and published pages for shipping, returns, and FAQs. In that kind of environment, the AI features have enough material to work with for search, recommendations, comparisons, analytics, and responses.

AI Commerce may not be a good fit if the store only has a few simple products and shoppers can already see the entire catalog in two clicks. It also will not fix a poorly structured catalog: if products lack proper names, attributes, images, and descriptions, the first step is to clean up the catalog. For stores with strict legal constraints around products, medical claims, financial services, or complex certification requirements, any AI-generated text should be treated strictly as a draft for human review.

Practical rule: enable AI Commerce only where there is a repeatable task, data you can verify, and a clear success metric. If you cannot define the metric, fix the core WooCommerce workflow first instead of automating it with AI.

What to Check Before Installation and First Launch

Before installation, you need to verify more than just WordPress compatibility. You also need to make sure the store is ready for an AI layer. CodeCanyon lists AI Commerce as compatible with current WordPress branches and positions it as WooCommerce-ready. On a real site, that is not enough: AI features depend on catalog quality, REST API availability, background jobs, external keys, and the way your theme renders search, product pages, cart, and checkout.

Start with a backup of both files and the database. That is standard practice before installing any WooCommerce plugin that can read products, orders, reviews, media files, and user data. Then test the store on a staging copy or development clone. AI Commerce touches too many areas to justify turning on every feature in a live store during peak traffic for the first time.

WordPress and WooCommerce Technical Readiness

Check three groups of conditions. The first is WordPress itself: ZIP upload through Plugins -> Add New -> Upload Plugin, a proper permalink structure instead of Plain, correct permissions for media uploads, and no fatal errors in the log. The second is WooCommerce: products open normally, cart and checkout complete a test flow, emails are sent, and order statuses change as expected. The third is the server: HTTPS is working, outgoing HTTPS requests to external APIs are not blocked, and WordPress cron jobs run consistently.

If the store uses HPOS, check compatibility in the WooCommerce settings. This is especially important for AI Commerce because of its order-related features: the chat assistant may answer order questions, forecasting uses sales history, risk screening analyzes new orders, and support drafts may pull context from the order record. If a third-party plugin is not compatible with HPOS, WooCommerce may disable some modes or require a compatible order storage method.

Catalog Readiness for AI Search

Semantic search and recommendations do not work well with an empty catalog. Before indexing, check that your key products have titles, short descriptions, full descriptions, categories, tags, attributes, variations, pricing, stock status, featured images, and reviews. You do not need perfection, but the core data needs to be coherent. If a product is named "Model 12" and described as "good quality," AI will not reliably understand whether it is a cotton T-shirt, a winter jacket, or an accessory.

For variable products, WooCommerce global attributes are especially important: size, color, material, volume, compatibility. A shopper may search for "medium black leather bag" or upload a photo of something similar. If those details only exist in free text, the outcome will be less predictable. If they are stored as attributes, search, recommendations, and comparison all have a more reliable foundation.

External Services and Keys

The product page and documentation mention OpenAI GPT-4o, GPT-4 Vision, Qdrant and Pinecone vector databases, and the remove.bg API for image background removal. That does not mean you need to connect every service on day one, but it does mean the admin should understand in advance which features rely on outside services, where the keys will be stored, and who is responsible for usage limits and costs.

External service keys should not be shared with copywriters, contractors, or content generation systems. Keep them only in the site settings or in a secure secrets manager. If a feature is not currently needed, do not add the key "just in case." That approach reduces unnecessary cost and simplifies troubleshooting: if only semantic search is enabled, it is much easier to understand why that specific feature failed to return results.

Minimum Rollout Plan

1. Create a backup and test a standard order flow without AI Commerce.
2. Install and activate the plugin on a staging site or during a quiet traffic window.
3. Add only the API keys you actually need and do not enable every module at once.
4. Run product indexing for semantic search if you plan to use search and recommendations.
5. Test public search, the product page, cart, checkout, and the order page in the customer account area.
6. Enable administrative features only after the shopper-facing layer no longer disrupts the buying journey.

Installation and Initial Testing Without Putting the Store at Risk

AI Commerce installs like a regular WordPress ZIP plugin. In the admin panel, open Plugins -> Add New, choose Upload Plugin, upload the archive, install it, and click Activate. After activation, locate the plugin settings section in the admin area. Depending on the build, the menu label may vary slightly, so look for AI Commerce and its related settings tabs.

The purpose of the first check is to answer one question: does the site still behave like a WooCommerce store when the AI layer is mostly empty? Open the store homepage, a category page, a product page, the cart, checkout, and the customer account area. Do not limit testing to the admin panel. AI Commerce inserts shopper-facing features into the public journey, so the frontend outcome matters more than the fact that the settings screen opens.

What to Enable First

If the store is small or you are testing the plugin for the first time, start with features that are easy to verify visually and easy to roll back: semantic search, recommendations via shortcode, a review summary on one product, and the description generator for draft content. These features should not automatically change pricing, hold orders, or hide products without your direct involvement.

Be more careful with features that can alter business operations. Pricing suggestions should never be applied without manual approval. Product moderation, if it moves flagged items to draft status, should first be tested on a few controlled product pages. Order risk checks should start with a threshold that sends suspicious orders to manual review rather than silently removing them from the process.

How to Tell Whether Installation Went Smoothly

After activation, check for four signs. First, the admin panel shows no fatal errors and WooCommerce pages still open normally. Second, products can still be edited and saved. Third, a test product can be added to the cart and checked out using a payment gateway in test mode. Fourth, each enabled AI Commerce feature appears only where you expected it to, for example a recommendations shortcode renders a recommendations block instead of breaking the page template.

Expected state after the first launch: the store works as it did before, and AI features have been added in specific, testable places. If the core purchase flow breaks, disable the last module you turned on and return to setup one step at a time.

Semantic Search and Photo Search: Index Setup That Affects Everything

One of the most product-specific parts of AI Commerce is its ability to search by meaning rather than exact word matches. The documentation explains this through a case where a shopper does not know the exact product name and instead describes intent, such as "something warm for cold evenings." A standard search may fail if those exact words are not in the title. Semantic search compares the query against product page content and returns products that are close in meaning.

This feature depends on indexing. The plugin needs to "read" your existing products and build a representation that future queries can be compared against. The documentation specifically says that after enabling the feature, you need to click Re-Index Products so the AI can process the existing catalog; new products are indexed when they are saved. This is one of the most important actions after installation.

How to Prepare Products for Indexing

Before clicking Re-Index Products, open several typical products and make sure the description actually describes the item. For clothing, you need material, season, fit, color, style, size range, and use case. For electronics, include compatibility, power, package contents, warranty limitations, and model differences. For furniture, specify dimensions, material, room use, color, and assembly details. The better the product page reflects a real buying decision, the more useful semantic search becomes.

You do not need to stuff long keyword lists into the description. Semantic search is meant to work with natural language. But it does not guess hidden traits. If a product is suitable for hiking, that should be stated somewhere in the description, attributes, category, or tags. If a dress is intended for evening wear rather than everyday use, that should be explicitly stated as well.

Post-Indexing Validation

After indexing, create a small set of test queries. It should include exact names, misspelled queries, descriptive queries, intent-based queries, attribute-based queries, and queries that should not have a good answer in your store. That last group matters a lot: if search always shows something even when no suitable products exist, shoppers may get a false impression of the catalog.

Check not only whether results appear, but also how they are ranked. A good outcome for AI Commerce is not just "it found something," but "the first results match the shopper's intent." If the query "light rain jacket" starts with winter puffer coats, the problem may be the product descriptions, the attributes, or an overly broad category.

Photo Search

Photo search solves a different problem. A shopper sees an item in a magazine, on social media, or on another person, but does not know what to call it. They upload an image, the AI identifies the product type, color, style, material, and visual traits, and then shows similar products from the catalog. According to the documentation, this is especially useful for clothing, home goods, jewelry, and furniture.

For that use case, image quality is critical. If your main product photos are dark, blurry, or shot against inconsistent backgrounds, visual matching becomes much harder. At minimum, review your top products: the image should clearly show the item itself, not a collage, banner, or promotional scene where the product only takes up a small part of the frame.

Recommendations, Review Summaries, and Product Comparison on the Frontend

AI Commerce shopper-facing features go beyond search. The plugin also adds recommendations, review summaries, and product comparison. These three mechanisms help at different points in the decision process: recommendations keep the selection moving forward, review summaries help shoppers understand social proof faster, and comparison helps reduce hesitation between similar items.

You should not enable these features as decorative widgets. If recommendations appear in the wrong place, they become distracting. If a review summary appears on a product with too few real reviews, it can feel artificial. If product comparison is available in a category where products are too different from one another, the AI will try to compare items that should not be compared.

Where to Place Recommendations

The documentation mentions the [ai_recommendations] shortcode and placement in a widget area. In practice, start with one page or one area where recommendations genuinely help. The clearest locations are the product page after the description, the cart before checkout, a sidebar on a category page, or a dedicated block on a landing page. Do not place the same block in every empty theme area.

For new visitors, recommendations are based on product similarity: category, tags, price, and other traits. As browsing and purchases accumulate, they become more personalized. That means your first test should happen in two modes: a clean browser with no history and a return visit after viewing several products. If the block looks identical in both cases, check whether personalization is enabled and whether there is enough user behavior data.

How to Review a Review Summary

AI Review Summary turns a set of reviews into a short summary, a list of pros and cons, and an overall takeaway. This is useful on products with many reviews where shoppers would otherwise struggle to spot recurring themes quickly. But if there are very few reviews or they all say roughly the same thing, the summary may feel thin. Do not treat it as a substitute for the reviews themselves. It is better to place the summary next to the standard reviews block so shoppers can move from the short takeaway into the details.

The check is simple: choose a product with real reviews, read 10 to 15 reviews manually, then compare the AI summary with the recurring themes. If customers often praise sizing but complain about delivery, the summary should not ignore the second half. If the AI generalizes too confidently, keep more of the original review context on the page and do not hide the source reviews.

Product Comparison Without False Confidence

AI Product Comparison lets shoppers choose two or more products and receive a written comparison with pros, cons, and a recommendation. This is especially useful in categories where products look similar but differ in material, specs, intended use, or price. In fashion, that may be similar jackets; in electronics, models with different power levels; in cosmetics, different sizes and ingredient profiles.

Good comparison setup starts with the data. Products need comparable attributes. If one product has material and size while another only has promotional copy, the comparison will be skewed. Before enabling it publicly, choose three product pairs and make sure the AI is not inventing features that do not exist on the product pages. If you see unsupported conclusions, improve the product records and limit comparison in categories where the information is too inconsistent.

Description Generation, Photo Processing, Categories, and Translation

The second major AI Commerce group helps store owners maintain the catalog. It includes description generation, image cleanup through background removal, pricing suggestions, automatic categories and tags, and product translation. These features save time, but they require editorial oversight. In a live store, an AI draft should be reviewed the same way you would review copy from a junior content manager.

Product Descriptions as Drafts, Not Final Copy

AI Description Generator works inside the product editor: the user enters a product name, chooses a category, and starts generation. According to the documentation, the plugin creates a short description, a full description, and SEO suggestions. This is useful when the catalog was imported with bare product names and you need to turn them into human-readable copy quickly. But the final version still needs human review.

That editorial review should cover four points. First, facts: material, size, package contents, compatibility, and warranty terms must match the real product. Second, claims: the AI must not add medical, legal, financial, or technical guarantees without confirmation. Third, brand tone: the text should sound like your store, not a generic storefront. Fourth, product uniqueness: if products are similar, the descriptions should not be copies with only the color changed.

Background Removal and Keeping the Original

In the documentation, AI Image Enhancement is described as background removal for product images. This feature requires a remove.bg API key, and the original photo is not deleted: the cleaned version is saved to the media library, and you can then assign it as the featured image. That is a good workflow for products photographed against a warehouse-style background, but it should be used selectively.

Not every image needs to be turned into a studio shot. For furniture, clothing, accessories, and electronics, a clean background can make product comparison easier. For decorative sets, interior products, and lifestyle-driven categories, context may matter more. Check the outcome on the product page, the category grid, and the mobile layout. If the product looks cut off or loses its sense of scale after background removal, keep the original photo or use the cleaned image only as a secondary one.

Automatic Categories and Tags

AI Auto Categorization & Tagging reads the product title and description, suggests categories, and adds tags. The documentation notes that you can run the action on individual products or in bulk through Products -> All Products and Bulk Actions. It also states that if no suitable category exists, the AI may create one.

This is exactly where structural control matters. Automatic category creation is convenient during large catalog imports, but it can quickly create chaos: "T-Shirts," "T Shirt," "Tee," "Men Tops," and similar variants. Before running anything in bulk, create a list of approved categories, test on 10 to 20 products, and make sure the AI is not creating unnecessary branches. Use the same rule for tags: they should help with filtering and internal search, not turn into a random dictionary.

Product Translation and Multilingual Compatibility

AI Auto Translation translates product titles, short descriptions, and full descriptions. The documentation points to integration with WPML and Polylang: if those plugins are in use, the translation can be saved into the corresponding language slot. This is useful for a store that is already building a multilingual structure, but it does not replace proper localization review.

Check more than the text alone. Review SEO URLs, currency, units of measurement, size charts, local return rules, and WooCommerce emails. Automatic product translation may be strong, but the store is still a system: if the product page is translated while shipping, checkout, and emails remain in another language, the shopper still ends up with a mixed experience.

Sales Forecasting, Trending Products, and Business Insights

The administrative side of AI Commerce is interesting because it tries to turn WooCommerce data into clear actions. The documentation includes Sales Forecast, Business Insights, and Predicted Trending Products. These features analyze order history, categories, revenue, average order value, growth or decline, and then explain the findings in plain language.

These features are especially useful for owners who do not want to build reports manually every day. But forecasts should still be treated as a supporting signal. AI can detect a trend in the data, but it does not know every outside factor: an ad campaign, a product delivery, a seasonal promotion, a warehouse delay, a competitor pricing change, or a technical issue in analytics.

When Forecasting Becomes More Useful

According to the documentation, Sales Forecast is most accurate when there are at least several months of order history. A new store will show a not-enough-data message. That is a normal limitation. If orders are sparse, the forecast may show too much uncertainty and the confidence level will be low. In that case, it is better to use forecasting for observation rather than purchasing decisions: which products appear more often in AI recommendations, which categories are growing, and where the data is too noisy.

For a mature store, set a regular review cadence: once a week, look at the revenue forecast, projected order volume, growth trend, and recommendations. Compare them with the actual outcome. If the AI consistently overestimates demand in a specific category, investigate why: there may have been one-time promotions, inaccurate order statuses, returns, or a seasonal spike.

Trending Products as an Early Signal

Predicted Trending Products looks at sales velocity and tries to identify items that may become bestsellers. The documentation uses the term velocity score: how much faster or slower a product is selling compared with the previous period. This is not magic. It is simply a way to notice a change in momentum.

Validate trending products against inventory and margin. If the AI shows growth in a low-margin, inexpensive item, do not rush into a large purchase. If it highlights a seasonal product with limited stock, that is a reason to check inventory, the product page, advertising, and alternatives. A good outcome from this feature is not an automatic decision, but an earlier and better question for the owner to ask of the data.

Business Insights Without Blind Trust

Business Insights generates a written analysis for a selected period: revenue, order count, average order value, top categories, observations, risks, and recommendations. Use it as an agenda for management review. If the AI says a category is growing, open the actual orders. If it flags a risk of decline, check stock availability, pricing, filters, ad traffic, and checkout errors.

Do not let these recommendations automatically change pricing, inventory, or ad budget. The store owner should stay in control. AI Commerce can speed up problem detection, but decisions about purchasing, promotions, and pricing should still be made by a human using multiple sources.

Product Moderation, Order Risk, and Support Replies

The third distinct AI Commerce area is tied to trust and operational safety. The plugin advertises AI Product Moderation, AI Fraud Detection, and Helpdesk Auto Reply Draft. These features affect sensitive areas: product publication, order handling, and customer communication. That means setup should be cautious and easy to verify.

Moderating New Products

AI Product Moderation is useful in stores where products are added by multiple managers, suppliers, or sellers. According to the documentation, the plugin checks for inappropriate content, misleading descriptions, duplicates, and incorrect categories. If it finds a problem, the product can be moved to draft status and the admin receives an explanation.

Start with test products. Create several drafts: one correct product, one product placed in the wrong category, one product with suspiciously exaggerated copy, and one duplicate of an existing page. See which specific signals the AI flags. If it blocks normal products too often, the setup should be softened or limited to notifications only. If it misses obvious issues, improve the rules, categories, and source data.

Order Risk Scoring

AI Fraud Detection scores each order on a risk scale and may move suspicious orders into On Hold status. The documentation lists signals such as address mismatches, unusually high order value, a new customer placing an expensive order, multiple rapid orders from the same IP, and similar indicators. As a starting point, the documentation suggests a threshold of 70, but the real threshold should be tuned for your store.

Do not use the risk score as a final verdict. Expensive gifts, corporate purchases, and shipping to a relative may all have legitimate reasons for mismatched addresses. The best workflow is to hold a suspicious order for manual review rather than automatically cancel it. Review the order record, payment gateway, customer history, address, phone, email, order frequency, and notes.

Support Reply Drafts

Helpdesk Auto Reply Draft helps prepare a response to a customer message. According to the documentation, a manager opens the order, pastes in the message, chooses a tone, and receives a draft. One detail matters: AI writes the draft, but a human still reads, edits, and sends it manually. That is the right model for support because a customer reply may involve promises, timelines, refunds, discounts, or exceptions to policy.

To make drafts more accurate, publish and maintain your shipping, returns, warranty, payment, and FAQ pages. The chat assistant and reply drafts should not invent store policies. If a policy is not clearly described on the site, the AI may produce an overly generic response. It is better to give it a verified source inside the store and still leave the final decision to the manager.

Practical Setup Example: Enabling AI Search, Recommendations, and Safe Order Review

Let us walk through a setup for a WooCommerce clothing store. The goal is to help shoppers find suitable products faster, show personalized recommendations, and route suspicious orders into manual review. This is a concrete example because it connects the customer journey with administrative control: search leads to a product, recommendations support the choice, the order goes through checkout, and risk screening prevents a questionable purchase from being processed automatically.

Goal

The target outcome is this: a shopper enters a descriptive query such as "light rain jacket," sees relevant products, opens a product page, gets a recommendations block, places a test order, and the admin sees a clear risk status. At the same time, the plugin should not change pricing, mass-create categories, or publish AI-written text without review.

Preparation

Before setup, choose 20 to 30 products from one category, such as jackets, hoodies, and accessories. Fill in the titles, short descriptions, full descriptions, categories, tags, global attributes, images, and stock status. Add several real or test reviews to one product if you want to test review summaries. Make sure checkout works in the payment gateway's test mode.

Setup Steps

1. Open the AI Commerce settings and enable the semantic search module.
2. Run Re-Index Products for the existing catalog and wait for completion or the expected processing status.
3. Add [ai_recommendations] to a test page, widget area, or a safe block on the product page.
4. Enable AI Review Summary only for products with enough reviews.
5. In the order risk section, enable scoring but choose manual review through On Hold instead of automatic cancellation.
6. Place one normal-looking test order and a separate test order with unusual conditions, such as a high total or a different shipping address.

Validating the Outcome

Public testing starts with search. Enter an exact query, a misspelled query, and an intent-based query. Make sure the first results match the meaning, not just a single similar word. Then open a product page and check whether the recommendations interfere with the purchase button, variations, price, tabs, or reviews. On mobile, the recommendations block should remain readable and must not break the add-to-cart button.

Administrative testing goes through the order. Open the test order and check whether the risk score, its explanation, and the expected status are present. If a normal order gets a high risk score for no good reason, lower the sensitivity or review which signals the AI is treating as suspicious. If a suspicious order passes as safe, raise the sensitivity or add manual rules in the payment and fraud-prevention layers.

A Common Pitfall

After enabling recommendations and search, do not remove standard category navigation. AI search helps the shopper who can describe what they want, but another shopper will still rely on filters, menus, sorting, and category browsing. AI Commerce should strengthen the core WooCommerce path, not replace it completely.

How to Evaluate Quality After Setup

Once the first setup is in place, run a calm quality review. A common mistake is to treat the job as finished the moment the interface appears on the site. With AI Commerce, the outcome is not measured by the existence of a button, block, or report. It is measured by whether those elements actually help shoppers and admins make better decisions.

Testing the Customer Journey

Create a table of 10 to 15 queries: exact product names, synonyms, misspellings, use-case queries, photo search queries, and searches by size, color, and material. For each query, record the first result, the relevance of the top five products, and whether an empty result appears when no matching product exists. If search returns too much irrelevant content, go back to the product pages and attributes.

Check recommendations for three user types: a new visitor, a visitor after viewing several products, and a shopper after adding a product to the cart. Recommendations should shift with the context. If the cart shows products that do not go well with the selected item, review your categories, tags, related products, and behavior history.

Reviewing Content Quality

Generate descriptions for several products, but do not publish them immediately. Flag unsupported claims, unnecessary promises, unnatural tone, repeated phrasing, and weak SEO suggestions. A good AI draft saves editor time, but it does not remove responsibility for factual accuracy. If a product has legally sensitive limitations, do not trust the generator with the final copy.

Reviewing Administrative Outputs

For forecasts and business insights, compare the recommendations with real orders. For risk scoring, test several orders with different profiles. For moderation, create control product pages. For support, paste in typical customer emails and see whether the AI promises anything that is not actually covered by store policy. Any AI output that affects money, order status, or customer promises must go through human review.

Safe Improvements Without Editing the Plugin Core

For AI Commerce, it is not a good idea to recommend PHP hooks or internal classes unless they are confirmed in the product documentation. Without access to the code of the specific installed version, the safer path is to stay with WooCommerce settings, theme options, shortcodes, CSS, and verifiable operational rules. Below is a small example that does not interfere with the plugin itself and simply helps the recommendations block sit more cleanly inside the theme.

If the [ai_recommendations] block is rendered inside a theme container and looks too cramped, you can add CSS through Appearance -> Customize -> Additional CSS or through a child theme. Before doing that, inspect the real HTML class of the block in your browser developer tools. The example below uses a placeholder class that you need to replace with the actual class used by your block.

.ai-commerce-recommendations {
  margin: 32px 0;
  padding: 20px;
  border: 1px solid #e5e7eb;
  border-radius: 10px;
  background: #ffffff;
}

.ai-commerce-recommendations .products {
  gap: 18px;
}

.ai-commerce-recommendations .price,
.ai-commerce-recommendations .button {
  margin-top: 10px;
}

After adding it, check the product page, category page, and mobile layout. If the theme product grid changes size or breaks the button, remove the CSS and adjust the container through the theme's built-in options. Rolling back is simple: remove the block from Additional CSS or temporarily disable it in the child theme.

For performance, use another safe technique: do not place AI recommendations on every page right away. Start with one location, measure page load time before and after, check caching, and make sure the block does not conflict with lazy-loaded images. If the site uses aggressive caching, exclude only the dynamic fragment if the theme or caching plugin supports that. Do not disable the entire store cache just for one block.

Common AI Commerce Issues and Troubleshooting

Problems with AI Commerce usually do not come from a single button. They come from the chain of "catalog data -> external service -> background jobs -> theme -> WooCommerce." That is why troubleshooting should move from simple to complex: first identify the symptom, then check the data, settings, queue, theme, and only then look for a conflict with another plugin.

Search Does Not Find Obvious Products

Symptom: the shopper enters a clear query, but the results are empty or irrelevant. Possible causes include indexing not being run, products being saved before the feature was enabled and never added to the index, descriptions being too sparse, key traits not being stored as attributes, or cron not running background tasks.

Check whether you ran Re-Index Products, whether products open in the admin panel, whether they contain descriptions and attributes, and whether WordPress background jobs are running. Fix: improve a few control product pages, rerun indexing, and repeat the test queries. If the outcome improves only for the updated products, the problem was in the catalog data.

Photo Search Shows Strange Matches

Symptom: an uploaded bag photo returns shoes, decor, or random products. Possible causes include poor featured images, inconsistent angles, the product taking up only a small portion of the frame, products lacking descriptive traits, or the visual category being too broad.

Check the product images that appear in the results. If they use decorative banners instead of clear product photos, replace the featured image or add a clean item photo as the primary image. For fashion and furniture, use attributes for color, material, and intended use. If the feature is not useful for a specific category, do not display it there just for effect.

Recommendations Look the Same for Every Shopper

Symptom: the recommendations block does not change after viewing different products. Some causes are normal at the beginning: too little behavior data, a new visitor, no accumulated order history, or products described too similarly. Caching can also be the issue if a dynamic block is being cached as static HTML.

Check in a private window and in a regular session, then compare the behavior after viewing several products. If the site uses a caching plugin, verify that the same block HTML is not being stored for all users. Fix: configure the dynamic area carefully without disabling all caching, and improve product relationships through categories, attributes, and tags.

AI Descriptions Sound Good but Add Extra Facts

Symptom: the description generator promises properties the product does not have. The cause is usually insufficient source context or an overly generic product title. The AI fills in the text based on probability, not on your store's internal inventory knowledge.

Fix: before generation, add the key facts to the product page, and after generation, verify the material, dimensions, package contents, limitations, and promises. For risky categories, use generation only as a draft. Rollback: do not save the generated text or restore the previous product revision.

Background Indexing or Forecasts Do Not Update

Symptom: the index never reaches the expected state, forecasts do not refresh, or new products take too long to appear in AI search. Possible causes include WP-Cron not running because the site gets too little traffic, the server blocking outgoing HTTPS requests, an external service being unavailable, or an incorrect key.

Check cron jobs using standard WordPress hosting tools or through your system administrator, without posting keys in tickets or chats. Verify API keys and usage limits in the external services. If the issue started after enabling a specific module, disable it temporarily and test core search and WooCommerce first.

Orders Get High Risk Scores Too Often

Symptom: normal orders keep ending up in On Hold. Possible causes include a threshold that is too low, gift shipping patterns, corporate purchasing, an active promotion, a new sales region, or an unusual payment gateway. Do not automatically cancel these orders.

Fix: review the risk explanation in the order, compare it with the payment gateway, customer history, and addresses. Raise the threshold or leave the feature in manual review mode. If the store often receives legitimate orders with different billing and shipping addresses, do not treat that signal alone as enough reason to block them.

The Chat Assistant or Support Draft Replies Ignore Store Policy

Symptom: the AI promises a refund, delivery timeline, or stock availability that is not confirmed on the site. Causes include empty or outdated policy pages, incomplete order data, or the lack of a clear knowledge base.

Fix: publish current shipping, returns, payment, warranty, and FAQ pages. Make sure the assistant is reading those sources. For disputed or sensitive responses, keep manual editing mandatory. If a manager sees an unsupported promise, it should be removed before sending.

When CodeCanyon AI Commerce Is a Good Choice

CodeCanyon AI Commerce is worth testing if you want to add several AI capabilities to a WooCommerce store at once: semantic search, photo search, recommendations, comparison, help with descriptions, translation, forecasting, moderation, risk scoring, and support drafts. Its strength is the breadth of the feature set. Its risk is the temptation to turn everything on at once and end up with a complex system that no one is evaluating clearly.

A better rollout looks different: you choose two or three features for a specific problem, prepare the catalog, run indexing, test the public journey, measure the quality of the outcome, and only then expand the module set. For a large catalog, that may mean search, recommendations, and comparison. For a store with frequent inquiries, it may mean chat, policy pages, and reply drafts. For a catalog with many suppliers, it may mean moderation, categories, and manual product-page review.

If you are ready to test the plugin on your WooCommerce site, start with a backup, a staging environment, and a control set of products. After that, you can download CodeCanyon AI Commerce, install the ZIP archive, and go through setup step by step using this guide. Do not judge the product by the first polished AI answer. Judge it by the repeatable outcome: shoppers find products faster, managers spend less time on drafts, and admins see risks before they turn into real problems.