The right price can still be an incomplete AI answer

A six-question product pilot shows why facts, missing information and source relevance need separate checks. An AI shopping answer can give you the correct price and still leave a basic buying question unanswered. That happened in a small pilot collected on 2 October 2026: Google's AI Overview gave the UK manufacturer price for a Logitech mouse, but did not say whether the manufacturer had it in stock. A second answer in the same pilot gave the maximum DPI shown on the preferred current product page. Its source panel also included a document for an older model. These observations expose two things a single correct-or-incorrect score can hide: whether the answer covers the question, and whether its sources fit the claim.

The price matched and the stock answer was missing

The question specified Logitech's own UK store, the MX Master 3S for Mac in Space Gray, and part number 910-006571. It asked for both price and stock status, explicitly separating the manufacturer's offer from other retailers. The captured answer gave £119.99. Logitech's UK product page showed the same price immediately before and after the answer was collected. Those two manufacturer snapshots also showed the item as temporarily sold out. The answer left stock unspecified and directed the reader back to the store. [1] The fair judgment is precise: the price was consistent with the snapshots, and the requested stock fact was not answered. The answer did not claim the mouse was in stock. Calling this a false availability claim would overstate the evidence. The timing matters. The manufacturer snapshots were taken at 07:10:59.554 and 07:11:50.659 UTC, with the answer retained between them. Matching observations do not prove the stock state was continuously unchanged in that interval. They also say nothing about today's price, another colour or another retailer. For a buyer, the practical consequence is an extra step: the answer has not established whether that exact offer can be purchased. For an evaluator, the consequence is a separate completeness finding. A price check alone would miss it.

The answer used the current value and also showed an older source

Another question specified the Logitech MX Master 3S for Business, UK part 910-006582. It asked for maximum tracking DPI, whether a Logi Bolt receiver was included, and the published limited hardware warranty. The answer gave 8,000 DPI, agreeing with the preferred exact-model manufacturer page. Current 3S material appeared among its sources. The expanded source panel also contained an older MX Master 3 for Business PDF. That document names the model without the S, lists 4,000 DPI, and gives EMEA part 910-006199. [2, 3]

The older PDF was a source card, rather than the named marker attached to the 8,000-DPI statement. Its presence does not show that the answer used it for that value. The supported finding is a model mismatch within the source set. There is no basis here for claiming that the old document caused a wrong DPI answer. This case also had a reference uncertainty. Before collecting the answer, an official support search extract showed a 200-4,000 DPI range, while the exact Business 3S product page and current datasheet said 8,000. The discrepancy was recorded in advance. A later direct read of the support hub did not expose that specification, so the pilot kept the check reference-ambiguous rather than declaring a current manufacturer defect or an AI error. An official domain is a useful starting point. The next check is the document itself: model name, part number, market, variant and the particular statement it supports. A document for the previous model can sit on the same trusted domain as the current one.

Give each answer three separate judgments

Start with the facts the question actually requested. In the price-and-stock case, there are two requested facts. A complete review gives each one its own record.

Factual agreement

Does the statement agree with a suitable reference for this exact product and context? Keep maximum battery life separate from typical runtime, and a published hardware warranty separate from statutory rights. If authoritative references conflict, preserve the conflict instead of selecting whichever source makes the answer look right or wrong.

Completeness

Did the answer address every requested fact? Missing stock status deserves an explicit unanswered label. A suggestion to check the shop may be sensible caution, but it does not supply the stock answer. Likewise, earbuds' water resistance does not answer whether their charging case has the same protection.

Source relevance

Does an inspected source support this particular statement for the right model, variant, country and time? Keep a visible inline marker, a source card and a shopping link distinct. If the interface does not establish which claim a card supports, record that uncertainty. A relevant source in the panel does not validate every sentence in the answer.

A review record a buyer can use

For each requested fact, keep the question, answer paraphrase, reference URL, reference time and one of these judgments: supported, contradicted, partially qualified, unanswered, reference-ambiguous or unassessable. Use a separate field for source support and explain any mismatch. For changing prices or stock, add the exact seller, market, currency, variant and part number. Compare observations close to the answer time. If you cannot align the times, say so. “Snapshot-consistent” is a more defensible label than treating a changing offer as a permanent truth.

Retain the original answer and source evidence privately so someone else can review the judgment. A later run of the same question is a new observation; it may return a different answer. Repeating a query should not replace an inconvenient first result.

What the six question pilot covered

The private questions and reference checks were fixed before the first consumer query. The purposive sample covered four named products or variants across Sony and Logitech:

Sony WH-1000XM5: maximum battery life with noise cancelling on and off, plus Bluetooth codecs including aptX
Sony WF-1000XM5: earbud-only battery life, keeping charging-case top-ups separate
Sony WH-1000XM5 versus WF-1000XM5: water resistance, including the earbud charging case
Logitech MX Master 3S for Mac, UK part 910-006571: receiver, cable and hardware warranty
That exact UK Mac variant: the manufacturer's own price and stock
Logitech MX Master 3S for Business, UK part 910-006582: maximum DPI, receiver and hardware warranty All six questions produced real Google AI Overviews on one attempt each. Of 17 prespecified fact checks, 13 were supported by the scoped references, one was partial, one was price snapshot-consistent, one was unanswered and one retained the reference ambiguity described above. Some checks overlap and are not independent trials. No unambiguous stable-fact contradiction was established. These categories should not be collapsed into a headline error rate. Collection used logged-out cloud Chrome, English questions and a United Kingdom country footer. Tabs shared a profile and cookies, with manufacturer-page browsing before collection. Personalization and the hidden model version were unknown. Nine selected answer-linked destinations were inspected, at most two per question; the other sources were not fully assessed. Collection and review were performed by an automated assistant, without independent human calibration. This was a small diagnostic pilot, not a representative benchmark, an evaluation of every claim in every answer, or a test of whether changing a page improves citations. It did not measure unbranded discovery of Beacon or demonstrate a Beacon software correction workflow. No product fix or before-and-after intervention was performed. The useful lesson is the review habit: check the value, account for what is missing, and inspect the source at the level of the claim. In this pilot, those separate judgments revealed information that a single accuracy score would have lost.

Sources for the two worked examples

Logitech UK MX Master 3S for Mac in Space Gray, part 910-006571
Logitech UK MX Master 3S for Business
Older MX Master 3 for Business datasheet The findings refer to retained observations from 2 October 2026. These live pages may change. Product names identify the study cases; no affiliation with Google, Sony or Logitech is implied.