Using AI for ecommerce without losing the facts

Choose a task before choosing a tool

Start with one repetitive task that has a clear input and a result someone can review. Examples include reorganising verified specifications into a description, drafting FAQs from approved information or suggesting category labels for a small catalogue. Treat these as tasks to evaluate, not guaranteed capabilities of every AI product.

Write down how long the task takes today and what a good result looks like. Tool features, pricing and terms change, so compare current official documentation against your actual workflow. This guide offers an assessment method rather than a paid-tool ranking or a claim of tested performance.

Give the draft a reliable source of facts

Prepare a short product fact sheet with the exact model, dimensions, materials, included items and known limitations. Mark missing details as unknown. Keep a reference to the manufacturer document, approved supplier specification or other evidence supporting each fact.

Ask the tool to use only the supplied facts and flag gaps instead of filling them. For example: draft a clear Australian English description, preserve all measurements, add no certifications or performance claims, and list unanswered questions separately. Good instructions help structure the task, but they do not eliminate the need to check the result.

Review every claim a customer could rely on

Compare the draft with the fact sheet line by line. Pay particular attention to compatibility, water resistance, safety, environmental claims and what is included. The ACCC explains that advertising claims must not be false or misleading. AI-generated wording still needs the same evidence as copy written by a person.

Images need a similar check. An altered product photo might change colour, texture, shape or accessories in ways that affect a buying decision. Use accurate product photography for the item itself. Keep conceptual illustrations clearly identified and separate from evidence about a product's appearance or performance.

• Remove unsupported superlatives, invented reviews and unverified certifications.
• Check measurements, units, model names and any numerical comparisons.
• Read the final page with its images, title and delivery information together.

Keep customer information out of experiments

Use fictional enquiries when testing a tool. Avoid pasting customer names, addresses, order histories or private supplier documents into a public chatbot. Check the provider's data handling, retention, training settings and access controls before deciding what information is appropriate to use.

The OAIC's commercial AI guidance recommends careful product assessment and human oversight, with particular caution around personal information. If you are considering a customer-facing assistant, assess the privacy obligations that apply to your business, explain the use of AI clearly and provide a practical route to a person.

Measure editing effort as well as drafting speed

Evaluate a small batch of representative items, including one with missing specifications and one with several variants. Record drafting time, checking time, factual errors and how much rewriting was needed. A fast first draft is not a saving if the review takes longer than the original task.

Keep approved text separate from drafts and assign a person to publish it. Repeat the check when the tool or workflow changes. Continue only where the process produces accurate, useful content at a reasonable total effort; leave judgement and accountability with the business.