What should an audit example include before you pay?
A useful AI shopping audit example shows the exact buyer question, the dated assistant answer, the source behind it, the specific problem, the proposed change and a repeatable check. Arbling's methodology, reviewed on 24 October 2026, retains those questions, answers, dates and sources. An ecommerce owner deciding whether an audit is worth buying can treat that list as an acceptance test. A score alone does not say what was asked, when the answer was captured or what to do next, so you cannot inspect it or argue with it. The report should let you distinguish what the assistant said from what the auditor inferred. Keep the original answer beside the proposed change so that a later review can challenge either one. If the sample deliverable cannot show each item, the audit example is a pitch rather than evidence, and you can ask for the missing pieces before any money changes hands.
Use the checklist below when you read a sample. Each line is a question for the provider, and a vague reply tells you something too.
- The question. Is the exact buyer question written out, and does the report say whether it named your brand? A question that names the brand tests how it is described. A question asking for suitable products without supplying your name tests whether the assistant brings you up.
- The dated answer. Is the assistant's reply shown with the day it was captured? Answers change, so an undated answer cannot be compared with anything later.
- The supporting source. Does the report say which source sat behind the answer? Arbling's methodology retains sources next to answers for exactly this reason.
- The specific problem. Is the problem tied to a named page, product or variant? "Improve your content" is not a finding.
- The proposed change. Can you preview the change before it goes anywhere? Arbling's product page describes previewing catalog changes for approval.
- The repeatable check. Will the same question be asked again after the change? The same page describes rechecking the same buyer questions.
Merchants are already asking for this kind of proof. A merchant discussion from 28 September 2026 asks whether AI shopping creates real orders and questions the value of being sold an audit without convincing evidence. The checklist is a practical answer to that doubt. It moves the conversation from a promise to something you can read, question and test yourself.
A worked audit example from Arbling's own self-audit
This audit example comes from Arbling's retained self-audit, collected on 23 October 2026. It is not customer proof and not an independent recommendation. The original answer archive is not publicly linked, and the methodology page explains the method, not a public transcript, so read the excerpt as a model of how a record is structured. The question named the brand: What is Arbling (arbling.com), and who is it for? It checks whether the answer describes Arbling accurately; it does not test unaided discovery. Gemini's reply included one short phrase, quoted exactly in the record. The excerpt matters because it compresses a workflow with an approval step into a promise of instant publication. The value of the example is its shape: question, answer, problem, fix and re-check, each one separable and checkable against the original retained answer.
The record, in order:
- Question naming the brand. "What is Arbling (arbling.com), and who is it for?" Source: Arbling's retained self-audit, collected 23 October 2026. The methodology page describes the method; the underlying answer archive is not publicly linked. It tests how Arbling is described, not how buyers find it.
- Gemini's exact excerpt. "It allows merchants to fix and publish missing product facts instantly." That is an assistant's wording, not a verified product claim.
- Why "instantly" is the problem. Arbling's product page says catalog changes are previewed for approval. The word "instantly" leaves out that approval step, so a reader could conclude that changes go live with no review.
- Proposed clarification. Say plainly, on Arbling's own pages, that catalog changes are previewed and then approved before they are published.
- Repeat check. Once the clarification is live, ask Gemini the same question again and compare the new answer with the retained one.
- Result claimed here. None. No correction and no successful remeasurement is claimed in this article.
The table separates what was found, what is suggested and what has actually been shown. The second row is illustrative, a store scenario and not a record.
| Finding | Suggested fix | Demonstrated result |
|---|---|---|
| Gemini's wording said "instantly", which omits the approval step on the product page. | Clarify that changes are previewed, then approved. Ask the same question again afterwards. | None. No correction and no remeasurement is claimed. |
| Illustrative store scenario: an assistant's answer leaves out one of your products. Hypothesis: product facts are missing from the catalog. | Inspect the catalog first. If the facts are confirmed missing, preview and approve the fixes, then recheck the same question. | None. An omission in an answer does not prove its cause. |
Treat missing product facts as a hypothesis until someone has inspected the catalog. An assistant can leave a product out for reasons the answer does not reveal, so an omission is a prompt to look, not a diagnosis. The product page describes recording products, variants and sellers in AI answers, which gives that inspection a starting point.
How to judge a sample deliverable and ask for the next step
Audit scope is the first thing to check in an audit example. Ask for the scope first: which questions were asked, which assistants answered and which pages and catalog items were inspected. Ask for the question versions next, because Arbling's methodology, reviewed on 24 October 2026, freezes and versions question sets and says different sets must not be presented as the same measurement. If the questions change between measurements, you cannot treat the difference as a response to your fix. Then ask for the limitations in writing. The methodology describes a snapshot, not a promise that an assistant will recommend a company or increase its revenue, and a provider who promises either has gone beyond the evidence. Finish with one concrete next action that you can approve, decline or test. Ask the provider to separate the findings from the commercial proposal: which changes are included, who approves them, what the work costs and what will be measured afterwards.
Four requests turn a sample into something you can evaluate:
- Scope. A list of the questions, the assistants and the pages covered, so you know what the report does not cover.
- Question versions. The version of the question set used, so a later check compares like with like. Ask whether the same questions will be repeated after any change.
- Limitations. A plain statement of what the snapshot cannot tell you, including that no result is guaranteed.
- A concrete next action. One change you can preview and approve, tied to a named page, product or variant.
Orders deserve their own question. Arbling's product page describes connecting recorded AI visits with carts, checkouts and orders, naming the linking method and showing missing evidence as missing. Ask any provider the same: how is a visit linked to an order, and what does the report show where the link is absent? A report that never shows a gap is harder to trust than one that does.
Also ask which catalog source the audit inspected. If it covers your product feed, the report should say which fields were checked and what was found. For wider context, the guide on AI visibility audits for online stores covers checks, costs and providers, and the guide on AI traffic analytics versus attribution explains what counters cannot tell you. If a provider will not show a sample at all, how to spot an ecommerce agency scam before you pay lists the warning signs.
Where Arbling fits. Arbling maps the checklist above to what it does. Retained buyer questions, assistant answers, dates and sources form the visibility record, with question sets frozen and versioned so the same question can be asked again. Verified catalog facts lead to approval-based fixes, because Arbling previews catalog changes for approval. Beacon visits and order links feed follow-up measurement, naming the linking method and showing missing evidence as missing. The full sequence runs in order: the buyer questions, what assistants show about products, variants and sellers, the fixes you approve, the AI visits and orders that follow, candidate patterns worth testing, comparable products where those verified changes are appropriate, and finally the same questions asked again to measure the change. It is a snapshot, not a promise of recommendations, citations or revenue.
Run a free scan to see what AI assistants show about your store today.
Sources checked 24 October 2026.