Is this product in stock at this store, and can I collect it today?
What we checked
4products
Ridge Shell Jacket · size M
Merino Base Layer · size L
Trail Hiking Boot · size 43
Frost Beanie
×
5stores
Oslo
Bergen
Trondheim
Stavanger
Tromsø
=
20missions in each mode
Agent mode
Fetch mode
MISSION
one product at one store, asked one question
AGENT MODE
the AI clicks through the store's site, like a shopper
FETCH MODE
the AI takes one quick look at the page to answer in a chat, and clicks nothing
THE QUESTION, FILLED IN FOR ONE MISSION
Is Ridge Shell Jacket, size M in stock at Fjellsport Demo Bergen, and can I collect it today?
01ASK
The question goes to two models from different vendors, Anthropic and OpenAI, with the product and the store filled in.
02ANSWER
In agent mode the AI clicks through the store's site, like a shopper. In fetch mode it takes one quick look at the page, the way a chat answer does.
03CHECK
A fixed recipe reads the store's own page, with no model, at the start and at the end of each check.
04GRADE
Each answer gets a verdict and an owner: the store's site, the access layer or the agent. A finding needs agents from two vendors to agree, or an access check that saw the same refusal.
Results at a glance
20 MISSIONS IN EACH MODE
Agent mode
The AI clicks through the store's site, like a shopper.
14 of 18missions passed
13 PASS
1 PASS (stock only)
3 WRONG ANSWER
1 PARTIAL
1 AGENT ERROR
1 NOT GRADED
4 had a store-facing failure.
1 could not be graded.
1 ended in an agent error, which is not counted against the store.
Fetch mode
The AI takes one quick look at the page to answer in a chat, and clicks nothing.
5 of 19missions passed
5 PASS
12 FAIL
2 BLOCKED
1 NOT GRADED
14 had a store-facing failure.
1 could not be graded.
Who the failures belong to
BOTH MODES
THE STORE'S SITE
16
failures in 3 findings
THE ACCESS LAYER
2
failures in 1 finding
THE AGENT
1
agent error, not counted against the store
Results by product and store
ONE CELL PER MISSION
AGENT MODE · 14 OF 18
The AI clicks through the store's site, like a shopper. 14 of 18 graded missions passed. 1 could not be graded.
The stock answer matched. The collect-today claim could not be checked.
WRONG ANSWER
Agents from two vendors gave the same answer, and the store's own page shows something different.
FAIL
Agents from two vendors could not reach an answer on this site.
PARTIAL
The page did not hold all the data needed to answer.
BLOCKED
The site refused or challenged the request, or a robots.txt rule stopped the agent.
AGENT ERROR
The agent failed in a way that is not the store's fault. It is not a finding.
NOT GRADED
Could not be compared with the store's page. The reason is shown in the cell.
Findings
4 · MOST MISSIONS AFFECTED FIRST
01FINDINGFAILFetch modeAttributed to THE STORE'S SITE
AFFECTS12 of 19Fetch-mode missions
Store stock only shows after someone picks a store
AI assistants from Anthropic and OpenAI could not reach an answer on this site.
Same failure on both first attempts in 12 missions.
AFFECTED PRODUCTS AND STORES
Ridge Shell Jacket size M
Oslo
Bergen
Trondheim
Stavanger
Merino Base Layer size L
Oslo
Bergen
Trondheim
Trail Hiking Boot size 43
Oslo
Bergen
Trondheim
Stavanger
Tromsø
WHAT HAPPENED
The product page shows no store stock until someone picks a store. An AI that takes one quick look at the page clicks nothing, so it never sees a stock answer.
WHY IT MATTERS TO A SHOPPER
When a shopper asks an AI assistant in a chat, it cannot say whether the store has the product, so it sends the shopper elsewhere or guesses.
WHAT TO CHECK
Put each store's stock line in the page HTML itself, or publish store stock in the page data, so it is there before any script runs.
Open fjellsport-demo.test/product/merino-base-layer?size=L.
Without choosing a store, read the stock line. It says "Choose a store to see stock".
Choose "Fjellsport Demo Oslo" under "Choose a store".
The stock line now reads "In stock at this store".
02FINDINGWRONG ANSWERAgent modeAttributed to THE STORE'S SITE
AFFECTS3 of 18Agent-mode missions
Page data says in stock for every store
AI assistants from Anthropic and OpenAI gave the same wrong answer.
Same failure on both first attempts in 2 missions; 2 of 3 after a third attempt in 1 mission.
AFFECTED PRODUCTS AND STORES
Ridge Shell Jacket size M
Bergen
Trondheim
Trail Hiking Boot size 43
Trondheim
WHAT HAPPENED
The page data marks the product InStock with no store attached. Agents from both vendors read it as stock at the store they were asked about. The store's own page says the size is not in stock there.
WHY IT MATTERS TO A SHOPPER
A shopper who asks an AI assistant is told to come in and collect a size that is not on the shelf.
WHAT TO CHECK
Give each store its own offer in the page data, with that store’s availability, and keep the online offer separate.
ANSWER VS PAGE
ANOP
WHAT THE AGENT ANSWEREDAgents from Anthropic and OpenAI
STOCK
In stock
COLLECT TODAY
Yes
GT
WHAT THE STORE'S OWN PAGE SHOWSReference reading
STOCK
Out of stock
COLLECT TODAY
No
What the agent answeredYes, the Ridge Shell Jacket in size M is in stock at Fjellsport Demo Bergen and you can collect it today.
Request /product/merino-base-layer?size=L without a browser.
The answer is HTTP 403 with a security check page.
Request /product/merino-base-layer. The answer is 200.
04FINDINGPARTIALAgent modeAttributed to THE STORE'S SITE
AFFECTS1 of 18Agent-mode missions
Pickup time is missing for the boots at Stavanger
AI assistants from Anthropic and OpenAI found the same data missing.
Same failure on both first attempts in 1 mission.
AFFECTED PRODUCTS AND STORES
Trail Hiking Boot size 43
Stavanger
WHAT HAPPENED
The page shows the boots in stock at Stavanger but gives no pickup time for that store, so the question about collecting today has no answer on the page.
WHY IT MATTERS TO A SHOPPER
A shopper is told the boots are there but not whether they can collect them today.
WHAT TO CHECK
Show the pickup time for every store that has the product in stock, as the other stores already do.
ANSWER VS PAGE
ANOP
WHAT THE AGENT ANSWEREDAgents from Anthropic and OpenAI
STOCK
In stock
COLLECT TODAY
Not stated
GT
WHAT THE STORE'S OWN PAGE SHOWSReference reading
STOCK
In stock
COLLECT TODAY
not on the page
What the agent answeredThe Trail Hiking Boot in size 43 is in stock at Fjellsport Demo Stavanger.
Third attempts, agent-style (Anthropic, claude-opus-5-5, example-id-a3)
Third attempts, agent-style (OpenAI, gpt-5.6-sol, example-id-b3)
MARKETS
Norway
REFERENCE READING
Store page read by recipe “Local stock reading”, version 3, at the start and end of each check.
ENGINE
Rules v1.8, stored cause key ck1, cross-page grouping fk1, wording templates version 2.
HOW FINDINGS ARE COUNTED
Affects is the number of distinct missions (a product at a store, or one task) with this failure, out of the missions in that mode that had at least one graded attempt. Agreement shows how many of those failed the same way on both first attempts, how many were settled 2 of 3 after a third attempt, and how many were confirmed by one attempt and the run's access check. Findings are ordered by affected missions, then agreement. Nothing is combined into a single number.
AGENT ERRORS
1 of 38 graded attempts ended in an agent error. These are the agent's own failures. They are not findings and are not counted against the store.
NOT GRADED
Stock changed during the check: 2.
BLOCK RATE
0 of 18 graded agent missions ended BLOCKED.
2 of 19 graded fetch missions ended BLOCKED.
COLLECT-TODAY CLAIMS
1 PASS (stock only) result: the stock answer matched, and the collect-today claim could not be checked.
LIMITS
These results describe what the checked pages and agents did on the dates shown. Stores change, so a later check can differ.
Findings are confirmed by agents from two different vendors, or by an access check that observed a challenge or a refusal. An agent's own mistake is counted as an agent error and is never a finding.
The reference for each answer is the store's own page, read by a fixed recipe at the start and end of the check. When stock changed between the two readings, the result is marked not graded.
Checks ran as SendTillyBot, a declared automated visitor. A store may treat a declared bot differently from an ordinary shopper.
Agent-style models are not the consumer assistant apps, and these results do not predict what those apps would do.
Payment is not tested. Agents stop before any payment step and use a synthetic identity. Nothing is bought.