Zum Inhalt springen
Copywriter working with an AI writing assistant on a laptopAI-generated

Product Information Management

Create product descriptions with AI: from PIM data to reviewed copy

Creating product descriptions with AI requires reliable product data, clear writing instructions and a review of the facts. For manufacturers and retailers managing large ranges, variants and sales channels, the work starts with the data: which attributes belong to each product, and which does a particular description need?

A PIM system keeps this information in a structured form. It can provide the basis for drafts for an online shop, a catalogue or another channel. The example below shows the steps, including an inaccurate draft and a corrected version.

What product data does the AI need?

Start with a set of data for the specific product. Include the item number, product type and the attributes a buyer needs to make a decision. Depending on the range, these might include material, dimensions, capacity, colour, connection specifications or documented suitability for a particular use.

Check three things:

  • Correct assignment: Does the information describe the product family or the exact variant on offer? An attribute from a different version must not find its way into the description.
  • Complete values: A number needs a unit and a clear meaning. For dimensions, specify the order of length, width and height.
  • Supported claims: Properties, compatibility and usage instructions need a reliable source. Resolve conflicts between supplier copy and a data sheet before generating the text.

If an essential attribute is missing, send the record back for clarification. An optional attribute can be left out of the description. Define this distinction in advance so editors do not have to decide from scratch for every item. Learn more about the shared data source in OMN Product Information Management.

An example: turn six attributes into a product description

This is a fictional, simplified example created for this article. It uses no customer data and is not an actual output from OMN.

Field Reviewed input for the example
Item LB-24 storage box
Material Polypropylene
Capacity 24 litres
External dimensions 400 × 300 × 270 mm, length × width × height
Colour Blue
Contents supplied One box; lid available separately

The task is to write a factual English shop description of no more than 70 words, starting with the product name and explicitly stating dimensions and contents supplied. There is no evidence for claims about stackability, food suitability or weather resistance.

What an inaccurate draft might look like

The stackable LB-24 storage box is made from durable polypropylene and offers 24 litres of storage. Its matching lid keeps food and accessories reliably protected outdoors.

This illustrative draft adds several properties that are absent from the data. It claims stackability, suitability for food and outdoor use. The reference to a lid could also suggest that it is included. External dimensions are missing entirely. Fluent writing alone is therefore insufficient for approval.

A version based on the available facts

The blue LB-24 storage box is made from polypropylene and has a capacity of 24 litres. Its external dimensions are 400 × 300 × 270 mm (length × width × height). One box is supplied. A lid is available separately.

Each statement can be traced to a field. A more detailed description would require additional reliable information, such as the intended use. Adding adjectives does not supply those missing facts.

Write a useful generation brief

A reusable brief combines approved facts with editorial requirements. It specifies the audience, channel, language, tone and text format. Also define which details must appear and how missing information should be handled.

For the storage box, the brief could read:

Write a factual English shop description for the LB-24 storage box in no more than 70 words. Use only these facts: polypropylene, 24 litres, external dimensions of 400 × 300 × 270 mm in the order length × width × height, blue, one box supplied, lid available separately. Start with the product name. Explicitly state dimensions and contents supplied. Add no properties or applications. List any open questions separately from the product description.

Adapt this editorial working template to the text service you use.

Test the brief with different cases: a complete record, a variant, an item with optional accessories and a product with conflicting data. This reveals whether the rules work beyond a straightforward example.

Handle channel requirements and languages separately

A shop description, a short catalogue entry and a technical description can share the same source data. They differ in the information they emphasise and the space available. Define requirements for each output: mandatory attributes, length, units, terminology and tone of voice.

Keep essential details when shortening copy. A short version for the storage box could be:

LB-24 storage box, blue, polypropylene, 24 l. External dimensions (L × W × H): 400 × 300 × 270 mm. One box supplied; lid available separately.

Approve the facts in the source text before creating additional language versions. During translation review, check numbers, units, variant names and technical terms again. AI Translate in OMN supports the translation of existing product copy. Suitable languages, services and workflows are agreed for the use case.

Using an external writing assistant or a PIM integration

You can also supply an external writing assistant with product attributes and writing rules. This can be a starting point for drafting and testing a brief. Plan how to pass in the correct data version, associate the draft with its item and return the reviewed result to product maintenance.

Across larger ranges, these steps repeat for many products. An integrated text generation service can use existing structured data as its source. The actual handovers matter: which fields are processed, where does the draft go, who reviews it and how is the approved version used?

AI Text in apollon OMN generates product copy drafts from attributes in OMN PIM. Channel and tone of voice are considered during configuration. The product page describes data preparation, configuration and editorial review. Services, requirements and scope are agreed during the project.

Before approval: check the facts and the process

Compare the draft with its product record. A practical approval checklist covers:

  1. Product and variant: Does every detail belong to the correct item?
  2. Numbers and units: Are dimensions, quantities, sequence and notation correct?
  3. Contents supplied: Is it clear what is included and what is available separately?
  4. Properties and applications: Is every statement supported by the approved data?
  5. Channel and language: Does the copy meet the agreed requirements and terminology?
  6. Further use: Are responsibilities, the data version and the approved text version documented clearly?

Also decide how changes will be handled. If the contents supplied change in the example, the affected descriptions and language versions need another review. Automatic detection or blocking of such cases is part of the process only if it has actually been configured and tested.

For a pilot, choose a manageable product group that includes typical variants and missing data. Record which drafts pass without factual corrections, which errors occur and how much time data preparation and review take. This provides a firmer basis for expanding the process than the speed of a single generation request.

Common questions about AI product descriptions

Can AI fill in missing product attributes?

A generated assumption is not a reliable product property. Obtain missing facts from an appropriate source and resolve them in product data maintenance. Until then, leave the claim out of the description or hold the item for clarification.

Do I need a PIM system for a first trial?

For a single trial, you can provide a reviewed data record manually. With many products, variants and languages, maintaining and transferring data traceably becomes more important. That is where the decision about a PIM integration should start.

Does AI copy automatically improve search rankings?

Generating text does not establish that a particular ranking will follow. The first practical measure is whether the description answers relevant product questions accurately and clearly. Search queries, usage and actual enquiries need to be assessed separately after publication.

Discuss your product copy process with apollon

To explore AI text generation for your range, tell us about a typical product group, the channels and languages you need, and your current review process. Together, we can establish which data is available and what a suitable workflow with AI Text in OMN could look like.

Request a demo