A supplier sends you a product catalog. Some items have thoughtful descriptions, useful specifications, and detailed application information. Others arrive with a title that looks like an internal inventory label and a description that says little more than what the product is.
Both create work for your store. Thin descriptions leave customers with unanswered questions. More complete supplier copy may already appear on dozens of competing websites. Writing every listing yourself becomes difficult as your catalog grows.
In our October 21 webinar, Spark Shipping founder Charles Palleschi and Danny DiGiovanni explored how AI can help turn that source material into clearer product titles and descriptions. The session included a live product catalog walkthrough, three very different prompt styles, and questions from retailers about existing products, custom feeds, and accuracy.
Watch the recording below, then use this guide and the example prompts to try the workflow with your own products.
Want to jump to a specific part?
Open all three example prompt sets →
One of the webinar’s simplest examples was a brake rotor with a supplier description consisting of just two words: “brake rotor.” That identifies the general product, but gives a shopper very little help deciding whether this particular item belongs in their cart.
A useful listing makes the available facts easier to understand. What is the product? Which identifying details distinguish it? What specifications or documented applications should the buyer check? What does the package contain, if that information is supplied?
Those questions also apply when a manufacturer provides a long description. The information may be good, while its organization or tone may not suit your customers. A specialist parts buyer may want identifying specifications near the top. A general consumer may benefit from plain-language explanations of the same details.
The webinar showed both situations: rewriting an existing floor-mat description and developing a more useful starting point from sparse brake-rotor data. The purpose was to make the supplied information easier to use, with a presentation suited to the store.
During the session, Charles explained that Spark Shipping can pass product descriptions and available attributes into the content-generation process. Where a store product is linked to multiple suppliers, the available mapped product information can provide additional context.
This matters because useful details are often stored outside the main description. A supplier may put material, dimensions, color, or application data in separate fields. Those fields can help shape a more informative listing when they are available to the generation workflow.
Before experimenting with wording, review the source data. Bring in the details you actually have, including accurate identifiers and relevant attributes. Check conflicting values before asking a model to turn them into customer-facing copy.
The example prompts make accuracy an explicit requirement. They tell the model to preserve identifiers and measurements, avoid inventing compatibility or included accessories, and leave out sections the source cannot support. A more detailed input gives the model more to work with; a longer requested output does not create missing facts.
The shared document contains a title prompt and a description prompt for each approach. Together, the examples show how the same underlying product information can support different audiences, tones, and languages.
The first approach uses a knowledgeable product specialist’s voice. It prioritizes the information a serious buyer needs to identify the exact item and evaluate it against their requirements.
The sample title prompt is written for eBay. It starts with the product type, follows with important verified specifications or compatibility, and places the brand and manufacturer part number toward the end when supplied. It asks for a compact title within its specified character budget and protects the part number when shortening is necessary.
The description prompt gives the output a useful structure: an overview, feature explanations, technical specifications, documented compatibility, package contents, buyer questions, and a closing selection summary. Each section is conditional on the source supporting it.
Use this approach as a starting point when buyers compare similar products and need precise distinctions. Adapt the structure to your category and destination channel. A specification-heavy replacement part and a simple household accessory will not need the same depth in every section.
The sample descriptions aim for 800–1,200 words only when the source supports that depth. Treat that as a demonstration setting, not a requirement for every listing. A shorter, complete description can serve a buyer better than a long page of repeated claims.
For the second example, Charles applied a theatrical pirate persona to the product description. The prompt asks a retired pirate captain and a skeptical parrot to sell ordinary merchandise with exaggerated enthusiasm. The live demo used the floor-mat product to make the change in voice obvious.
The practical lesson is how much direction a prompt can provide. You can ask for a professional, technical, conversational, or playful tone, then describe what that should mean for your customers. A store serving experienced buyers might emphasize precise terminology. Another might prefer friendly explanations for first-time shoppers.
Even the pirate prompt keeps the real product facts separate from the performance. It preserves identifiers and specifications and explicitly avoids turning nautical jokes into claims about marine suitability. That is a useful pattern for any brand voice: creative language should remain consistent with what the product actually is.
When adapting this example, write down the tone you want, the audience you serve, and the language you want to avoid. Then review whether the resulting copy is still easy to scan and understand. Personality should help customers engage with the product information.
The third prompt set asks for natural Mexican Spanish, including a product title and a substantial description. It emphasizes terminology that local shoppers would recognize and a helpful product specialist’s tone.
The title instructions preserve brand names, model names, part numbers, measurements, and units. The description uses Spanish headings to organize the overview, features, technical details, compatibility, included items, purchasing considerations, and questions.
Charles demonstrated switching the prompt and generating Spanish-language copy during the session. For a retailer, this illustrates how language instructions can be part of the same product-content workflow.
Have a fluent reviewer check the result before using it with customers, especially technical terminology and selection details. Translating a listing does not establish local availability, shipping arrangements, certifications, or compliance. The example prompt deliberately avoids making those assumptions.
The live walkthrough began in the Vendor Product Catalog. Charles showed separate controls for AI-generated titles and descriptions, along with fields for custom prompts. He first demonstrated the default output, then changed the instructions to show how the presentation could shift.
Charles also opened an existing product and regenerated its content individually. That is a different workflow from adding new catalog products. During the Q&A, batch regeneration across an existing catalog was discussed as planned work, rather than a capability demonstrated in that session.
One detail worth checking before you experiment: the custom prompts shown in the demo were shared across the site, including other vendors. Start with instructions you would be comfortable applying beyond one test item, and review the scope before saving changes.
Start with a small, varied group of products. Include one with a strong supplier description, one with sparse text but useful attributes, and one with important compatibility or sizing details. That gives you a better view of how your instructions handle different inputs.
For each product, save the original copy and note what you want to improve. Perhaps the title buries the identifying specification, the description reads like an internal catalog entry, or the structure makes important selection details hard to find.
Then build your instructions around five decisions:
Change one instruction at a time when evaluating the results. If you change the tone, title order, length, and structure together, it becomes harder to tell which change improved the listing. Keep the adjustments that make the content clearer and remove instructions that encourage repetition.
Once a prompt works well on your test group, try it on a wider selection before using it broadly. Include unusual products and incomplete records. Those are often the cases that reveal whether your instructions are precise enough.
A polished description still needs a factual check. Compare the generated text with the supplier data, especially where a sentence sounds persuasive but adds a detail that was never supplied.
For search visibility, focus on the usefulness of the finished page. Google’s guidance on generative AI content emphasizes accuracy, quality, and relevance. Generating large volumes of pages without adding value can create problems, so changing the wording alone should not be the goal.
Clearer product information can support your broader SEO and conversion work, but using AI does not guarantee rankings or sales. Evaluate the content against customer needs and measure the results alongside your other store improvements.
Yes. In the Q&A, Charles explained that product data from a custom CSV can be used in the AI workflow. He encouraged retailers to bring in descriptions and available attributes so the generation process has more context. The same source-data and review principles apply.
Charles explained that Spark Shipping generates fresh output rather than distributing one shared cached description to every retailer. Custom prompts add further direction. Review the actual result for the voice and information your store needs, rather than treating different wording as the only measure of quality.
The webinar demonstrated regeneration from an individual existing product page. For a large catalog refresh, confirm the available workflow with the Spark Shipping team before planning a bulk change. The session distinguished that request from generating content while adding new products.
No. Charles showed default generation before adding the demonstration prompts. Start by reviewing that output. Add instructions when you have a specific requirement, such as title ordering, a clearer brand voice, a particular structure, or a language choice.
The next step is a small, practical experiment. Choose a few products, pick the prompt approach closest to your audience, and compare the generated content with the source. Keep the factual checks consistent as you adjust the presentation.
The shared document includes all three title-and-description prompt pairs: the marketplace specialist, the pirate shopkeeper, and the international storefront. Copy the parts that fit your store and adapt them to your products.
Explore Spark Shipping’s AI Product Titles & Descriptions to learn more, or contact hello@sparkshipping.com with questions about your product data and workflow.