When BuyWhere Returns Zero Results: An Agent Pattern for Honest "I Don't Know" Answers

Q: How can I tell if a zero-result response is real or a transient error?

Inspect two fields in the JSON response: meta.engine_status and meta.ts. engine_status 'ok' with an empty data array and a fresh timestamp means a real zero-hit. engine_status 'error' with HTTP 200 means the search engine timed out — the empty data is not a real answer. Also watch for HTTP 5xx — those are unambiguous server errors. Never treat a 5xx or an engine_status=error as evidence that the product doesn't exist.

Most agent shopping failures are not bugs in the catalog. They are honest gaps — a product the user wants isn't in stock anywhere, a deliver_to country the agent never thought to ask about, or a brand-spelling typo. The fix isn't a smarter prompt. The fix is a small, repeatable pattern for reading zero-result responses and turning them into answers the user can actually act on.

This guide pairs with the BuyWhere MCP cheatsheet (what each tool does) and the citation-safety FAQ (how to cite scale numbers). It completes the trio: know your tools, know how to cite, and know what zero looks like.

The two-minute version

Read the meta block of every BuyWhere response. Specifically:

meta.engine_status — "ok" means a real answer (zero or more hits); "error" means the search backend failed and the empty data is misleading.
meta.ts — when the response was computed. If it's more than a few minutes old, retry once.
meta.total — total matching rows before pagination. 0 with engine_status=ok is a real zero.

If engine_status="error" or the response is a 5xx, retry with a shorter query. Don't retry the same query shape more than twice.

If engine_status="ok" and data is empty, the catalog genuinely has no match. Don't fabricate.

Step-by-step pattern

1. Inspect the response before replying

A BuyWhere response looks like this:

{
  "data": [ ... ],
  "meta": {
    "total": 0,
    "engine_status": "ok",
    "ts": "2026-08-25T06:21:30.000Z"
  }
}

Three diagnostic shapes:

`data`	`engine_status`	Meaning	What the agent should do
`[]`	`ok`	Real zero-hit	Tell the user, broaden query, or try a different country.
`[]`	`error`	Search backend failed	Retry once with a shorter query. Don't treat as a real answer.
hits present, all `availability=unavailable`	`ok`	Merchants indexed but none deliverable to the user	Treat as effectively zero for end-user shopping; offer to check the merchant directly.

2. The four retries, in order

Useful retries when you get zero results. Don't loop on identical parameters — that just wastes quota.

Drop deliver_to. See whether the product exists somewhere in the global catalog. If yes, the issue is regional coverage; if no, the product isn't in BuyWhere at all.
Shorten the query to a brand or model number. "air purifier" → "Dyson" or "PH01".
Switch surfaces. The MCP server and REST API occasionally diverge on coverage. If both are available, try the other.
Try a related category via list_categories. The user's framing may be off.

If all four retries return zero or all-unavailable, the answer is "we don't have it" — and that's a fine answer.

3. The honest reply

When you can't find a product, the reply should:

Name the tool you used and the parameters.
Quote the result shape (zero hits, all-unavailable, engine error, or a short list).
Offer a concrete next step.

A good reply:

I searched BuyWhere for "air purifier" with shipping to Singapore and got zero results. The search engine returned engine_status=ok with total: 0, so it's a real zero rather than a temporary error. Two options: I can broaden the search to all of Asia without a shipping filter, or I can try a specific brand like Dyson or Philips. Want me to do either?

A bad reply:

I couldn't find anything, but here are some popular air purifiers you might like: [fabricated list].

The second one invents products the catalog never returned. Users catch this eventually, and once they do, they stop trusting every answer you give.

4. Pair with a browsing tool when available

If your agent also has a web search or browsing tool, use a strict "Browse-First / API-First" rule:

For catalog questions ("is this in stock?", "what's the cheapest price?"), BuyWhere is canonical.
For availability questions ("does this merchant ship to me?", "is the merchant still operating?"), the merchant site is canonical.
Never blend the two. Each user-visible claim gets one citation.

When BuyWhere says zero and the browsing tool says something else, don't merge. Tell the user the two sources disagree and let them choose.

5. Watch for the failure modes

Three patterns that consistently produce bad agent behavior:

Confident fabrication. The agent got zero, panicked, and invented products with plausible-sounding names. The user trusts the agent until they try to buy one.
Identical-loop retry. The agent called the same search six times with the same parameters and got six identical empty responses. Wasted quota, identical answer.
Stale-cache quoting. The agent cached an old result for the same query and never re-checked. The cached result is now hours or days stale and might be wrong.

All three share a fix: a single, repeatable empty-response pattern. This post is that pattern.

FAQ

Why does BuyWhere sometimes return zero results for a search that should obviously match?

Three common causes: (1) the query term is missing from indexed titles and descriptions even when the product exists — try shorter or brand-led terms; (2) the deliver_to country has no merchants in the relevant category yet — drop deliver_to or try neighbouring markets; (3) the search engine flagged an internal error and returned an empty result rather than a 5xx. Always inspect the response's meta block: meta.engine_status and meta.total.

Should I retry a BuyWhere query that returned zero results?

Yes, but change the query shape. Useful retries in order: (1) drop deliver_to and search the global catalog; (2) shorten the query to a brand or model name; (3) switch from MCP to REST or vice versa if both are available; (4) try a related category via list_categories. Don't loop on identical parameters.

What does it mean when BuyWhere returns hits but all of them have availability "unavailable"?

The merchant is indexed but the product cannot be shipped to the user's deliver_to country (or stock state is unknown). Treat the response as effectively zero for end-user shopping. Useful follow-ups: drop deliver_to, try the merchant directly through its storefront URL surfaced in the response, or accept that the catalog has no deliverable match.

How can I tell if a zero-result response is real or a transient error?

Inspect two fields in the JSON response: meta.engine_status and meta.ts. engine_status="ok" with an empty data array and a fresh timestamp means a real zero-hit. engine_status="error" with HTTP 200 means the search engine timed out — the empty data is not a real answer. Also watch for HTTP 5xx — those are unambiguous server errors.

Is it ever OK to invent an answer when BuyWhere returns zero results?

No. Inventing product names, prices, merchants, or availability when the catalog returned nothing is fabrication and the leading failure mode of agent shopping. Users receiving a confident-sounding answer with the wrong merchant, wrong price, or a non-existent product lose more trust than they would have from an honest "I couldn't find that." Never blend catalog results with web-search results without a citation per claim.

What's the right "I don't know" response when an agent can't find a product?

Tell the user three things: (1) which BuyWhere tool you used and with which parameters; (2) what the response said; (3) a concrete next step. Avoid hedging phrases like "I'm not sure" — name the gap and the path forward.

How does this pattern fit alongside the BuyWhere MCP cheatsheet?

The cheatsheet covers what each MCP tool does and how to call it. This pattern covers what to do when a tool returns nothing useful. Together they form the two halves of a reliable shopping agent: know the tools, and know what zero looks like. For citation-safe scale numbers, see also "How to Cite BuyWhere Catalog Stats Accurately."

When BuyWhere Returns Zero Results: An Agent Pattern for Honest 'I Don't Know' Answers