How assistants choose products
When someone asks “best vegan leather tote under $100”, an assistant searches the web, reads a handful of pages and writes an answer. It favours pages where it can find the facts the question needs (material, price, size, who it's for) stated plainly and consistently, and where other sources agree.
That means the work is mostly about your product pages, not about the assistant. Three things matter most:
- Specific facts in the text: “100% apple leather, fits a 15 inch laptop, 680 g”, not “premium materials, roomy design”.
- Answers to the questions people ask: is it waterproof, does it run small, how long does it last.
- Machine-readable data: Product structured data with price, availability and properties, so nothing has to be guessed.
Step 1: find out what assistants say now
Ask the questions your customers would ask, not your brand name. Write down whether you're mentioned, who is mentioned instead, and whether the facts about you are right. Ask more than once, because answers vary from run to run.
AI Mirror does this for you across ChatGPT, Gemini and Perplexity and keeps the history, so you can see change over time.
Step 2: fix wrong and missing facts
Most lost answers come down to a missing fact. If an assistant says your candle “doesn't say how long it burns”, the fix is one line on the page: “Burns for 40 hours.” Start with the facts behind the most common questions in your category.
| Category | Facts assistants look for |
|---|---|
| Bags | Material, laptop size, weight, waterproofing, strap length |
| Clothing | Fabric, fit (runs small or large), care, length |
| Candles and home | Burn time, wax type, scent notes, size |
| Shoes | Width, drop, weight, terrain, sizing advice |
| Beauty | Key ingredients, skin type, size, cruelty-free status |
Only state what's true. An assistant that finds a claim contradicted elsewhere trusts the whole page less. StoreAi's fact lock exists for exactly this reason.
Step 3: make the facts machine-readable
Put the same facts in structured form: Shopify metafields for product attributes, and schema.org Product markup on the page. Consistency matters: the price and availability in your markup must match the page.
The structured data guide walks through it, or use AI-ready product data to do it with an app embed.
Step 4: learn from who wins
When a rival keeps getting picked, read their page the way an assistant would. Which questions does it answer that yours doesn't? Which facts does it state? Competitor X-ray lays the two pages side by side.
Step 5: measure, and repeat
Assistants pick up changes at different speeds, from days to weeks. Rescan the same questions after each fix and track your share of answers per assistant. Do your best sellers first: they have the most to gain.
Questions
- Can I pay to appear in ChatGPT answers?
- Not in the organic answers this guide is about. They are chosen from what the assistant reads, which is why clear, factual product pages matter.
- Does an llms.txt file help?
- It can help AI tools find your important pages, but it can't make up for thin product pages. Fix the pages first. Shopify can generate llms.txt from your store, and it's only as good as the titles and descriptions it reads.
- Is this the same as SEO?
- It overlaps: clear, specific pages help both. The difference is that assistants quote facts and answers directly, so missing or vague facts cost you more.