Do not start by changing the product page
First preserve what the shopper saw. Save the prompt, answer text, engine or surface, time, cited URL, market, currency, and the product name as written. Then open the source. It can be old, describe another variant, or be a retailer rather than the manufacturer. An answer can also join facts from more than one page.
Do not call the answer wrong simply because it disagrees with the page you opened today. You need an exact product identity and two timestamps before you know what changed.
A real study limitation, not proof of an old price
In Cuebase’s 14–15 October 2026 pour-over study, two saved answers described a Hario ceramic kit at about $25.50. The official Hario page checked on 15 October showed product code XVDD-3012W at $44.50. This is an unresolved mismatch. The answers did not establish the seller, product code, or capture-time listing behind $25.50. The official page does not prove that no other kit or retailer used that price. We therefore did not label either price old.
That is the right standard for your catalog too: record the mismatch first, then find evidence that can explain it.
Compare four records in order
| Record | Question | Save |
|---|---|---|
| AI answer | What exact price, currency, product and source did it state? | Prompt, answer, time, engine, URL |
| Source page | Does it show the same variant and price now? Is there a visible date or sale condition? | Capture, URL, time, selected options |
| Your product record | Which SKU, variant and market are you comparing? | SKU, options, price, compare-at price, availability |
| Downstream offer data | Do feed and structured data show the same offer? | Item ID, price, availability, fetch time |
Classify the mismatch before acting
Different variant: stop comparing until you have matched SKU or option values. Different market or currency: record country, tax treatment, and currency; do not convert prices and call them equal. Sale or promotion: preserve the condition and date. Stale controlled data: correct the first Shopify, feed, or markup record you own, then recheck. Unresolved: keep evidence, schedule another check, and do not invent a cause.
Use Cuebase for repeatable observation
Cuebase can record whether a fixed AI-shopping prompt mentions or cites a tracked brand. It does not automatically confirm a price, identify the intended variant, or repair a feed. In Share of Voice, keep your prompt and tracked set stable, select Run probe now, and use Export CSV for saved metrics. Keep the answer and price investigation in a separate dated log.
Close the loop carefully
- Make the controlled correction, if you found one.
- Check the visible product page with the same variant and market.
- Wait for the relevant feed or markup update, then capture its next state.
- Repeat the prompt only when justified; a changed answer does not prove your edit caused it.
For record matching, read how to keep prices and availability consistent. For answer evidence, read what to count in an AI answer and AI share of voice.
Download the log
The blank log keeps source and product facts separate. It does not diagnose the cause.
Download the price-mismatch log (CSV). When your offer records are ready to inspect, start your free catalog scan.