Looking to improve your content rules and guidelines for consistent, accurate product content? Read on!
In this article
- Overview
-
What’s the difference between “default” and “custom” content fields?
-
How do I experiment with one prompt across different products?
-
Some outputs contain strange symbols (e.g., asterisks). Why?
-
How do I handle multiple product categories without writing a new prompt each time?
-
What’s the best way to test my prompt before launching a full-catalog generation?
Overview
Describely's content rules empower you to scale high-quality, on-brand content across thousands of product pages.
Keep in mind that getting consistent, accurate results in bulk—especially across different product types—can require a bit of upfront tuning.
Check out the below common questions and scenarios to ramp up quickly, and get your content instructions working for YOU.
What’s the difference between “default” and “custom” content fields?
-
Default fields (like Title, Description, Feature Bullet Points, Meta Title, Meta Description) are system-standard and supported out-of-the-box.
-
Custom fields are user-defined content types (like "Category page description", "Summary", and "Short" or "Long" descriptions) and can be used to tailor prompts to more personalized needs.
Why does this matter? Default and custom fields are controlled a little differently in Describely:
- Default fields can be managed through one "master" content ruleset, with a mix of command-style "instructions" and drop-down selections for content length, language, tone, and more.
- Custom fields (OR descriptions requiring advanced branding or HTML formatting) are managed through their own ruleset to maximize quality. These support a combination of drop-down selections and one long-form "instruction". We often refer to these as "custom prompts".
IMPORTANT: Regardless of the field type, you can easily make any needed changes to further guide the AI on your preferences. If you need help, we're here to support you at support@describely.ai.
Why is my content inconsistent, & how do I fix it?
Inconsistency often comes from unclear or overly broad instructions. To ensure uniformity at scale:
-
Use structured formatting (e.g., HTML headers, bullet points, bold text) to guide output shape.
-
Clarify expectations in the prompt, such as tone of voice, content order, or character count. Don't forget these best practices.
-
Test your prompt with 5-10, then 20-50 varied products before scaling to 1,000+.
The existing (or lack of) product data imported into Describely may also be the source of inconsistency. Consider:
- If you're referencing an imported data point or attribute in your instructions, but it’s missing or misformatted in the source data, your output will reflect that. Always verify that attribute or field names match exactly and that the data within them is consistently formatted (e.g., no missing units like “cm” or “oz”).
- If you're managing multiple categories with unique requirements, one-size-fits-all prompts may fall short. Instead, set up tailored prompts per main product category (e.g., laptops vs. headphones) to take your outputs to the next level.
Read more about how your data import impacts content quality here.
How can I shorten long paragraphs of text?
Add a simple instruction like:
“Limit each paragraph to 2–3 sentences” or
“Keep product descriptions under 300 characters.”
You can also apply formatting commands to create a concise structure:
“Use a bulleted list to highlight key features or technical specifications.”
Tip: This applies to ALL content elements, not just paragraphs. Try it on your titles, headers, bullet points, etc. if you need more control over content length.
How do I experiment with one prompt across different products?
That’s exactly what Describely is designed for—but it depends on data consistency and clear instructions. For best results:
-
Test the prompt on a diverse sample set
-
Look for content patterns that don’t align (e.g., missing features, mismatched tone, style or structure inconsistencies).
-
Adjust your prompt to account for edge cases—e.g., “If the [Battery Life] field of the original data is empty, leave all mentions of [Battery Life] out.”
Once your instructions perform well across a small sample, it should scale reliably to larger batches and ultimately thousands of SKUs.
Some outputs contain strange symbols (e.g., asterisks). Why?
While rare, rogue characters may appear when product data contains odd formatting or markdown.
Describely includes validation that automatically flags outputs containing prohibited keywords and symbols you've identified. If you still see them:
-
Double-check the source data for strange characters.
-
Save the product SKU so the generation can be retried or corrected via editing.
Stray asterisks can also be a sign that the AI needs more guidance on how to handle "list items" (usually tweaking the instruction to specify bullet points as a "bulleted list" works well).
How do I handle multiple product categories without writing a new prompt each time?
If your product categories have different brand tone and description structures (e.g., furniture vs. electronics), we recommend creating category-specific prompts. This allows you to:
-
Address different key specs or selling points per category.
-
Maintain consistency within categories while adapting language appropriately.
Using “if” logic in your prompts (e.g., “If category is ‘furniture’, emphasize dimensions and materials”) can help streamline this without full rewrites. Experiment to find what works best for you!
What should I do when the AI misses key product details?
First, confirm that:
-
The relevant product data exists in your original source data (either your imported file OR connected store) and is properly mapped to your fields.
-
The instruction explicitly calls out that data point (e.g., “Include power source from [Power Type] field of the original source data”).
Then update the prompt to reinforce it, such as:
“Always include the material type in the first sentence, if available in the original data.”
What’s the best way to test my prompt before launching a full-catalog generation?
Use the test batch method:
-
Select 5-10 diverse SKUs.
-
Run your prompt and review the outputs.
-
Check for:
-
Missing or inconsistent info
-
Tone and structure issues
-
Formatting gaps
-
Apply learnings and adjust your prompt accordingly. Once the test batch looks great, scale it up confidently.
Still stuck? Check out our Best Practices to Guide AI or Content Rules Library, or reach out to support@describely.ai with where you're struggling and a few examples—we’re here to help optimize for scale and quality.