What AI can and cannot do for product descriptions

AI is good at turning facts you already have into readable text. It's poor at supplying facts you don't have. That split explains most of what follows.

What it does well

Give a writing tool a supplier's bullet points and it can turn them into two tidy paragraphs. It can shorten a rambling description, smooth clumsy phrasing and offer a few different openings so you can pick one. Google says generative AI can be useful for researching a topic and for adding structure to original content.¹

For a small shop, wording is often the slow part, and wording is what the tool does.

Where it falls short

A language model doesn't look anything up. Google says generative models don't retrieve facts. They predict a likely run of words.¹ So when your prompt says "stainless steel bottle" and nothing about capacity, the draft may still state a capacity. It will sound sure of itself.

The usual name for this is hallucination. Wikipedia defines it as AI output containing false or misleading information that's presented as fact, and notes that chatbots can slip plausible-sounding falsehoods into their writing.²

In a product description, the usual culprits are:

• sizes and weights the tool never saw
• materials it guessed from the product name
• care instructions copied from a similar item
• features that belong to a rival product
• promises about durability or results

Who answers for the text

You do. The ACCC, Australia's consumer regulator, says information a business gives about its products must be accurate, truthful and based on reasonable grounds. Its list includes images and descriptions of what is offered, and it says intent makes no difference.³

That's an Australian rule, though other countries have their own. Nothing in the ACCC's list depends on who typed the words, so treat an AI draft as your own writing. For wording that needs extra proof, and where to find the official rules, see claims you shouldn't let AI write.

A rough split of jobs

A note on volume

Google also warns that using generative AI tools to generate many pages without adding value for users may break its spam policy.¹ A catalogue of near-identical descriptions is a risk for that reason alone. Add the facts that make each product different.

Selling through Google's Merchant Center? Its policy says AI-generated product data, such as title and description attributes, must be specified separately and labelled. Check the current policy first.¹

Start with the facts, then the prompt. The page on what to gather first is the next step.