Content Generation

Content generation is the core workflow of XC AI Content Automation. Once you have uploaded your product data (see Data Sources), you configure the AI to write optimized product descriptions, run the generation process, review the results, and then export or sync the finalized content to your store. This guide walks through every step in detail.

AI Settings

Before generating content, open the AI Settings panel inside your project. This is where you tell the AI how to write. Every setting here auto-saves as you change it, so there is no separate save button -- just adjust and go.

Provider

Choose the AI provider that will generate your content. The available providers are:

• Anthropic -- Uses Claude models. A strong general-purpose choice with excellent multilingual support.
• DeepSeek -- An alternative provider with competitive performance.
• Google -- Uses Gemini models.

Each provider offers different capabilities, pricing tiers, and speed characteristics. You can switch providers at any time and regenerate content if you want to compare results.

Model

After selecting a provider, pick a specific model from the dropdown. The list of available models changes depending on the provider. For example, choosing Anthropic gives you access to models like Claude Sonnet and Claude Haiku. Larger models tend to produce higher-quality output but may take longer and consume more tokens.

Language

Set the target language for the generated content. The platform supports over 30 languages, including English, German, French, Spanish, Portuguese, Italian, Dutch, Croatian, Serbian, Bosnian, Montenegrin, Slovenian, Macedonian, Bulgarian, Romanian, Hungarian, Czech, Slovak, Polish, Ukrainian, Turkish, Greek, Swedish, Norwegian, Danish, Finnish, Arabic, Chinese (Simplified), Japanese, Korean, and Hindi. The AI will write all descriptions in your chosen language regardless of the language of your source data. This is especially useful when you need to localize product catalogs for international markets.

Tone

Select the tone of voice for your generated descriptions. Options include professional, casual, and others. The tone setting influences the writing style across all generated fields -- short descriptions, long descriptions, and translated product names alike.

Content Type

Specify the type of product content to generate. This helps the AI understand what kind of content structure to produce and what context is relevant.

Short Description

Toggle this on or off to control whether the AI generates short descriptions for your products. When enabled, you can also set a character limit to control the maximum length. The default limit is 200 characters, and you can adjust it between 50 and 500.

Short descriptions are concise summaries -- think of the kind of text that appears in product listing pages or search results. If your platform does not support short descriptions (Shopify, for instance, does not have a separate short description field), this option is automatically hidden.

Long Description

Toggle this on or off to control whether the AI generates long descriptions. When enabled, set a character limit for the maximum length. The default is 1,000 characters, and you can go up to 5,000.

Long descriptions are the full product descriptions that typically appear on a product detail page. These tend to include feature highlights, materials, sizing information, and any other details that help the customer make a purchase decision.

Translate Product Names

Enable this option to have the AI translate product names into the target language. When turned off, the original product name from your source data is kept as-is. This is useful when your source catalog is in one language and you need localized product names for another market.

Use HTML

When enabled, the AI includes HTML formatting in long descriptions -- headings, lists, bold text, and other structural elements that render well on product pages. When disabled, the AI outputs plain text only. This setting only applies to long descriptions.

Custom Instructions

The Custom Instructions field is a free-text area where you can give the AI specific guidance about how to write your content. This is one of the most powerful settings available, because it lets you tailor the output to your exact needs without changing any other configuration.

Examples of effective custom instructions:

• "Focus on premium quality and craftsmanship."
• "Include relevant SEO keywords naturally throughout the description."
• "Mention free shipping on all orders over $50."
• "Keep the language simple and accessible -- our audience is non-technical."
• "Highlight sustainability and eco-friendly materials whenever applicable."
• "Always mention the warranty period."
• "Write in a way that appeals to young professionals aged 25-35."

The more specific your instructions, the better the output. You can combine multiple directives in a single block of text. If you change your custom instructions after an initial generation, use Regenerate All to apply the updated instructions to every item.

Running Generation

Once your AI settings are configured, you are ready to generate content. The project view provides three generation actions, each suited to a different situation.

Generate All Pending

Click Generate to process all items that are currently in Pending status. This is the action you will use most often -- especially the first time you generate content for a project. Items that have already been generated, approved, or rejected are skipped.

Before generation starts, a confirmation dialog shows you the number of pending items and the AI provider and model that will be used. Confirm to proceed.

Regenerate All

Use Regenerate All when you want to start fresh. This resets every item in the project back to Pending and re-generates content for all of them. It is particularly useful after you have changed AI settings -- for example, if you switched to a different model, updated the tone, or revised your custom instructions and want the changes applied uniformly across the entire catalog.

Regenerate Rejected

Use Regenerate Rejected to re-generate content only for items that you previously rejected during review. This is the most efficient option when you have already reviewed a batch and approved most items but need the AI to take another pass at a subset. Rejected items are reset to Pending and then processed; everything else is left untouched.

Progress Tracking

When generation begins, a progress modal appears showing real-time updates. You will see:

• A progress bar with the current item count and percentage complete.
• The name of the item currently being processed.
• A summary when generation is finished, showing how many items were generated successfully and how many were flagged for review.

You do not need to keep the modal open. Click Minimize to close the modal and continue working elsewhere in the application. Generation continues in the background, and you can check progress at any time from the dashboard. When generation completes, the project status moves to Review.

Reviewing Content

After generation is complete, the project enters the Review phase. This is where you inspect what the AI produced and decide what to keep, what to reject, and what to tweak.

Item List and Filtering

The left side of the review screen shows a scrollable list of all content items. Use the controls at the top to narrow down what you see:

• Status filter -- Show items by status: All, Pending, Generated, Flagged, Approved, Rejected, or Synced.
• Data source filter -- If your project has multiple data sources, filter items by source.
• Search -- Type a product name to find specific items instantly.

The list is paginated, showing 20 items per page, with navigation controls at the bottom.

Detail Panel

Click any item in the list to open the detail panel on the right side. The detail panel uses a side-by-side layout so you can compare the original source data with the AI-generated content:

• Left column -- The original data from your product catalog: product name, short description, and long description as they exist in your source file.
• Right column -- The AI-generated versions of each field: translated name, generated short description, and generated long description.

You can edit generated content directly in the detail panel. Click on any generated field to modify the text, then save your changes. Long descriptions that were generated with HTML enabled are displayed with full formatting and can be edited in a rich text view.

Confidence Scores

Each item displays a confidence score -- a percentage that indicates how certain the AI was about the quality of its output. High confidence scores (close to 100%) generally mean the AI had sufficient source data to work with and produced output it considers reliable. Lower scores may indicate that the source data was sparse, ambiguous, or that the AI encountered difficulty fulfilling the instructions.

Use confidence scores as a quick triage tool: items with high scores are more likely to need only a brief glance, while items with lower scores may benefit from closer inspection.

Actions Per Item

The detail panel provides three primary actions for each item:

• Approve -- Accept the generated content as ready for export or sync. The item's status changes to Approved.
• Reject -- Mark the item for regeneration. The item's status changes to Rejected. You can later use Regenerate Rejected to have the AI take another pass at these items.
• Regenerate -- Immediately re-generate content for this specific item using the current AI settings. This is useful when you want a quick retry on a single item without affecting anything else.

Bulk Operations

For efficient review at scale, you can select multiple items and act on them in batch:

1. Use the checkbox next to each item to select it, or use the Select All control at the top of the list.
2. A selection toolbar appears showing the number of selected items.
3. Click Approve Selected to approve all selected items at once, or Reject Selected to reject them.
4. Click Clear Selection to deselect everything.

There is also a Select All Flagged shortcut that selects every item with a Flagged status, making it easy to review all flagged items as a batch.

Flagged Items

During generation, the AI automatically flags items where it had low confidence or detected potential issues with the output. Flagged items appear with a warning badge in the item list and can be filtered using the Flagged status filter.

Pay special attention to flagged items. Common reasons for flagging include:

• Very little source data was available for the product.
• The source data contained conflicting or ambiguous information.
• The generated content did not meet the confidence threshold.

After reviewing a flagged item, you can approve it if the content looks acceptable, reject it for regeneration, or edit the generated text directly to fix any issues.

Keyboard Shortcuts

The item list supports keyboard navigation for faster review workflows. Click on the item list area to activate keyboard controls, then use the following shortcuts:

Shortcut	Action
Arrow Down	Move to the next item in the list
Arrow Up	Move to the previous item in the list
Space	Toggle selection (check/uncheck) on the current item
Enter	Approve the currently selected item

These shortcuts are especially useful when reviewing large batches. You can quickly move through items with the arrow keys, approve good ones with Enter, and flag items for closer review by moving past them.

Exporting

Once you have reviewed and approved your content, you can export it as a file for use outside the platform or as a backup before syncing.

Click Export and choose a status filter to control which items are included in the download:

• Approved -- Exports only items you have approved. This is the default and most common choice.
• Generated -- Exports all items that have generated content, including Generated, Flagged, and Approved items.
• All -- Exports every item in the project regardless of status.

The export is downloaded as a CSV file. For most platforms, the CSV includes the following columns: external ID, original name, generated name, original short description, generated short description, original description, generated description, status, and confidence score.

For Magento 2 projects, the export uses a specialized format with just three columns -- SKU, description, and short description -- optimized for direct import into Magento. If your project uses configurable products with variants, variant rows automatically inherit descriptions from their parent product.

If your project has multiple data sources, you can narrow the export to a specific source. You can also combine the export with the search filter to export only items matching a particular query.