Perplexity commonly displays source links beside answers, which makes domain-level evidence inspectable. That does not make every mention a storefront citation. A merchant must separate crawler access, answer text, and the URL actually shown as a source.
The response tests below were performed on July 27, 2026. They do not claim a Perplexity recommendation.
PerplexityBot access record
We requested five public commerce pages with a normal client and a PerplexityBot user-agent string:
| Store and page | Browser response | PerplexityBot-string response | Product evidence issue |
|---|---|---|---|
| The Woobles, Pierre kit | 200 | 200 | visible sold out vs structured InStock |
| Rhode, Glazing Milk | 200 | 200 | two sizes shared an observed SKU |
| Ring Australia, Battery Doorbell | 200 | 200 | compatibility facts were detailed |
| HexClad UK, pot set | 200 | 200 | Product name leaked Default Title |
| MyFonts, Axiforma | 200 | 200 | no Product entity detected initially |
A matching status proves only that our spoofable request received a response. Perplexity’s official crawler documentation recommends using published IP information with user-agent matching when configuring a WAF.
Robots case: Ring Australia
Ring’s robots file contained a dedicated group including PerplexityBot. It allowed / and disallowed private or transactional paths such as /account, /admin, /checkout, /cart, and /signin.
This demonstrates the distinction between public product discovery and private commerce actions. A store does not need to expose customer or checkout pages to make product facts retrievable.
Test the public response
url='https://example.com/products/example-product'
curl -sSIL "$url"
curl -sSIL \
-A 'PerplexityBot/1.0; +https://perplexity.ai/perplexitybot' \
"$url"
curl -sS https://example.com/robots.txt
Capture:
- initial and final status
- redirect chain
- effective market URL
- content type
- CDN or challenge headers
- canonical link
- Product and Offer facts
Do not whitelist arbitrary traffic based only on the user-agent text used above.
Source-level answer test
Use three prompt classes:
Discovery:
Which beginner crochet kits include video help and left-handed instructions?
Constraint:
Which Australian battery video doorbells support 2.4 GHz Wi-Fi and do not require existing wiring?
Verification:
What sizes and prices are currently listed for Rhode Glazing Milk? Cite the official product page.
For every completed answer, retain:
| Field | Example value |
|---|---|
| query ID | doorbell-au-compat-01 |
| prompt | exact submitted text |
| timestamp | ISO date and time |
| locale | country and language |
| answer state | success, timeout, refusal, provider error |
| brand mention | exact matched text |
| visible source | complete URL |
| domain class | official storefront, retailer, editorial, unknown |
| fact accuracy | expected versus observed |
Citation classification
Domain verified: a known canonical storefront domain appears in a visible source.
Brand only: the answer names the brand but visible sources belong to other domains.
Third-party product evidence: the product is supported by a retailer or editorial page, not the merchant’s canonical page.
Completed non-mention: a usable answer contains neither the approved brand string nor storefront domain.
Provider failed: no usable answer exists. Exclude it from the mention-rate denominator.
Case interpretation
If Perplexity cites The Woobles product page but repeats InStock, the crawl and citation stages succeeded while product evidence accuracy failed. The repair belongs in Shopify offer data, not in prompt wording.
If the answer names HexClad but cites only a review publisher, improve canonical bundle evidence and internal links; do not claim the storefront was cited.
These are classification rules, not claims about a run we did not perform.
Acceptance criteria
- legitimate PerplexityBot traffic is handled by an intentional robots and WAF policy
- public product pages return usable content
- visible and structured product facts agree
- every reported appearance includes the exact visible source URL
- brand mentions are not upgraded to domain citations
- provider errors remain separate from completed non-mentions
- prompt, market, timestamp, and limitations accompany every rate