When shoppers in Auckland, NZ ask for coffee beans, they're sent to roaster-a.example 67% of the time (8 of 12); you were named 0 of 12 times — because of delivery time (confirmed).CONFIRMED

You: 0% (0 of 12).OUTCOME

Protocol v4 · 05c5939c…c083Run coffee-pilot-v4-24 assistants

Who consumer AI named

“I need coffee beans delivered quickly in Auckland. Which online store should I order from and why?” · Auckland, NZOUTCOME

Store	Share	Answers	Interval	ChatGPT	Claude	Gemini	Perplexity
You — Your Storeyour-store.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-a.example	67%	8 of 12	95% interval 39–86%	2/3	3/3	3/3	0/3
roaster-f.example	58%	7 of 12	95% interval 32–81%	3/3	2/3	2/3	0/3
roaster-j.example	50%	6 of 12	95% interval 25–75%	1/3	3/3	0/3	2/3
roaster-q.example	33%	4 of 12	95% interval 14–61%	1/3	3/3	0/3	0/3
roaster-b.example	25%	3 of 12	95% interval 9–53%	0/3	1/3	0/3	2/3
roaster-aj.example	17%	2 of 12	95% interval 5–45%	0/3	0/3	2/3	0/3
roaster-d.example	17%	2 of 12	95% interval 5–45%	0/3	1/3	0/3	1/3
Roaster Eroaster-e.example	17%	2 of 12	95% interval 5–45%	0/3	1/3	0/3	1/3
roaster-ae.example	8%	1 of 12	95% interval 1–35%	0/3	1/3	0/3	0/3
Roaster Akroaster-ak.example	8%	1 of 12	95% interval 1–35%	1/3	0/3	0/3	0/3
roaster-am.example	8%	1 of 12	95% interval 1–35%	1/3	0/3	0/3	0/3
roaster-c.example	8%	1 of 12	95% interval 1–35%	1/3	0/3	0/3	0/3
roaster-g.example	8%	1 of 12	95% interval 1–35%	0/3	0/3	0/3	1/3
roaster-m.example	8%	1 of 12	95% interval 1–35%	0/3	0/3	1/3	0/3
roaster-o.example	8%	1 of 12	95% interval 1–35%	0/3	1/3	0/3	0/3
roaster-r.example	8%	1 of 12	95% interval 1–35%	1/3	0/3	0/3	0/3
roaster-x.example	8%	1 of 12	95% interval 1–35%	0/3	0/3	1/3	0/3
roaster-y.example	8%	1 of 12	95% interval 1–35%	1/3	0/3	0/3	0/3
roaster-aa.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-ab.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-ac.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-ad.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-af.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-ag.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-ah.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-ai.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-al.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-an.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-h.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-i.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-k.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-l.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-n.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-p.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-s.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-t.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-u.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-v.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-w.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3
roaster-z.example	0%	0 of 12	95% interval 0–24%	0/3	0/3	0/3	0/3

Share pools every assistant equally — the protocol asks each the same number of times. It is not weighted by how many shoppers use each one, so it moves if an assistant is added or dropped. The per-assistant columns carry no weighting.

CONFIRMEDOwner: Marketing

31 answers sent shoppers elsewhere on delivery time. Confirmed: Gemini tells shoppers your delivery estimate is 2 days, flat; your site states 1 day for urban delivery, 2 days for rural delivery — that is only true for rural delivery (1 of 2 answers). Named ahead of you: Roaster E.

Proof: Flattening diff — Gemini (Google grounding), delivery estimate: 2 days flat vs site 1 day for urban delivery, 2 days for rural delivery (holds for rural delivery), 1 of 2 answers, over 3 observations.

Proof: Flattening diff — ChatGPT (OpenAI web search), delivery estimate: 1 day flat vs site 1 day for urban delivery, 2 days for rural delivery (holds for urban delivery), 1 of 2 answers, over 3 observations.

How sure we are

Based on 12 answers from ChatGPT, Claude, Gemini and Perplexity; proportions carry 95% intervals. Two shares whose intervals overlap are not separated. Every assistant counts the same in the pooled Share whatever its reach; read the per-assistant columns for the unweighted picture.

How to read the labels

OUTCOMEa count. Ask the same assistants the same question and you get a number inside the stated range.
INFERENCEa reason consumer AI gave in its answers. It stays marked “consumer AI said” until a check backs it up.
CONFIRMEDa “consumer AI said” backed by a check you can re-run: the assistant’s claim against your live site, a price compared, a page read. The proof prints with it.

What this run pinned

Field	Value
Protocol	`v4 · sha256 05c5939c61b0c89a345604f70b18305dd845f3628149699a8782cb29d6b6c083`
Run	coffee-pilot-v4-2 · 2026-08-25T05:30:52.492Z → 2026-08-25T05:45:02.960Z
Panel	anonymised-sample-panel (paid)
Engines	ChatGPT (OpenAI web search), Claude (Anthropic web search), Gemini (Google grounding), Perplexity (Perplexity sonar)
Geos	Auckland, NZ
Queries	3
Repeats	3 per query, geo and engine
Observations	164 committed · 0 rejected · 0 protocol issues

Re-run: ask the same assistants the same query under protocol v4 (sha256 05c5939c61b0c89a345604f70b18305dd845f3628149699a8782cb29d6b6c083); the counts land inside the stated interval.