01
Two absences that look the same
Applies to marketplace onlyown storefronthybrid
A shopper asks an assistant for something an item matches, and the item is not in the answer. From the seller’s position that is one observation. Underneath it are two separate events.
Either the item was never in the candidate set the system chose from, or it was in that set and something else was chosen. The first is a retrieval outcome and the second is a comprehension or selection outcome. They produce the same result on screen and are addressed by different work, so a response chosen before the two are distinguished applies to one of them by chance.
02
Four gates, in order
Applies to marketplace onlyown storefronthybrid
Four gates describe the sequence more usefully than two outcomes, because the first two do not involve the text of the listing.
| Gate | The question it asks | Addressed by page-level work? |
|---|---|---|
| Eligibility | Is the item permitted here at all — category, account, venue policy? | No |
| Access | Can the system reach the page, and is it permitted to? | No |
| Retrieval | Given a query, does the item enter the candidate set? | Marginally |
| Selection | Among candidates, is it used and described accurately? | Yes |
The gates are worked in order and the check stops at the first closed one. Downstream of a closed gate, a change produces no observation that distinguishes it from no change, so the round yields nothing to read either way.
03
The retrieval gate
Applies to marketplace onlyown storefronthybrid
The third gate is separated out because it depends on inputs held elsewhere. Whether a specific item enters a candidate set for a specific query is decided by an index the seller does not operate, built at a time the seller does not observe, by a process no operator documents at the level a merchant could act on.
The published visibility research states results for the setting it constructed, and does not describe a merchant-side method producing rank, traffic, or sales.1 That is the current state of the cited evidence rather than a qualification added to it.
04
What is different for a single object
Applies to marketplace onlyown storefronthybrid
Four properties of unique stock apply specifically at this gate, and they are properties of the stock rather than of the seller’s work.
- There is no shared identifier joining the item to anything else a system already holds.
- The item exists once, so nothing repeats it across sources.
- It is described in the seller’s words alone, with no corroborating page elsewhere.
- It is removed permanently on sale, where an index holds records that persist.
The catalog-improvement literature is conducted at large-retailer scale with brand knowledge bases available, and the authors state that the setting is their own.2 Those conditions describe a shelf of repeated products rather than a shelf of single objects, so the reported results are about a different input.
05
What is available regardless
Applies to marketplace onlyown storefronthybrid
The following applies at gates one, two, and four, and each item also serves a person reading the page.
- Establish gates one and two and record the date, since both change without notice.
- Make the item unambiguous to a reader who has not seen it: one location, one identity, one condition account.
- Make the facts internally consistent, so a system reading the page is not selecting between two values.
- Keep availability accurate, since an item retrieved and then found to be sold produces a different outcome than one never retrieved.
Structured markup sits at the fourth gate rather than the third. Google documents it as representation that can make a page eligible for particular search features, which is a statement about those features and not a route into another answer system.3 Both halves of that hold at the same time.
06
Recording the diagnosis
Applies to marketplace onlyown storefronthybrid
The output is one sentence: which gate stops the item, and the date that was established. That sentence can be re-checked, and a rewrite performed without it cannot be compared against anything.
| Gate | Established for MER-1978-041 | How | Date checked |
|---|---|---|---|
| Eligibility | Open — used goods accepted, category permitted | Read the venue’s own policy page | Checked this week |
| Access | Closed — a disallow line covered the product path | Read robots.txt line by line | Checked this week |
| Retrieval | Not assessable while gate two is closed | Not attempted | — |
| Selection | Not assessable for the same reason | Not attempted | — |
The two rows carrying a dash record the state of the audit rather than an omission from it. Gates three and four cannot be assessed while gate two is shut, and “not attempted, with the reason” is a different entry from a blank.
The record also fixes the reference point for the next round. Without one, absence prompts a rewrite, the rewrite changes nothing observable, and the silence afterwards has the same two possible causes it had before. With one, the gate is a closed question re-checked on a cadence.
07
The gates and what each answers
Applies to marketplace onlyown storefronthybrid
The cited visibility research reports what it observed in its own setting.1 The catalog-improvement work is set in a large retailer with brand knowledge bases available.2 Google’s documentation states requirements for particular search features.3
None of them states why a given item is absent from a given answer, which is why the diagnosis is first-party and per surface. An item can pass every gate on one surface and stop at the first gate on another, so one verdict covering all of them describes none of them.
08
Practice
Exercise
Diagnose one item
- Pick one live item and work the four gates in order, stopping at the first closed one.
- Write the gate and the date in the same place you keep the listing.
- Do no rewriting this round, and note what you would have rewritten had you not checked.
Check yourself
An item never appears. What is established first?
Whether it is eligible and reachable. Those are the first two gates, neither is addressed by page-level work, and a change made past a closed gate produces nothing to compare.
Why is the retrieval gate the least tractable?
Because entry into a candidate set depends on an index the seller does not operate and no operator documents at an actionable level. The cited research reports its own setting rather than a merchant-side method.
Does structured markup address a retrieval outcome?
No. It is representation documented as making a page eligible for particular search features, which sits at the selection gate. Adding it and its not being a retrieval remedy are both true.
Progress is saved in this browser only. No account, nothing sent anywhere.
09
Common questions
How are retrieval and selection told apart in practice?
Query for the item by its most unambiguous identifying facts. Absence under those conditions places the outcome upstream of selection, where the wording of the description is not the variable.
Does this differ across surfaces?
Yes, and a diagnosis is recorded per surface. An item can pass every gate on one and stop at the first on another.
Does this mean page-level work should stop?
The parts that serve a human buyer hold regardless. What changes is the outcome they are recorded against, since they operate after retrieval rather than on it.
Does more content affect retrieval?
The cited sources do not state that it does. More text is also more values to keep accurate, and an inaccurate statement on a page is read as a statement.
10
Research and sources
Rules and platform policies change. These primary sources were reviewed on ; confirm the current position for your jurisdiction and account before acting.
Claim evidence
- Published generative-engine visibility research reports measures inside its own tested setting and does not establish a merchant-side method that produces ranking, organic traffic, or sales.
- observed outcome. Supported by GEO: Generative Engine Optimization .
- Catalog-improvement research conducted at large-retailer scale with brand knowledge bases available is explicitly scoped to the authors’ own setting and affiliation, so it does not transfer to a shelf of single objects.
- observed outcome. Supported by Using brand knowledge bases and LLM agents to enhance e-commerce retailers’ catalog quality .
- Product structured data is documented as representation that can make a page eligible for particular Google Search features, and that documentation does not govern behavior in other answer or shopping systems.
- technical requirement. Supported by Product structured data .
- GEO: Generative Engine Optimization Association for Computing Machinery · Tier B · peer reviewed · KDD 2024 proceedings
Visibility measures in the paper’s tested generative-engine setting. Limit: The study does not establish a universal merchant ranking method, organic traffic effect, or sales outcome.
- Using brand knowledge bases and LLM agents to enhance e-commerce retailers’ catalog quality Amazon Science · Tier B · peer reviewed; publisher-affiliated research · WSDM 2026 proceedings, pages 1343–1344
Catalog repair and entity matching using brand knowledge bases in the authors’ large-retailer setting. Limit: A two-page proceedings contribution rather than a full paper, so method detail is limited. Amazon-authored, large-catalog research does not establish transfer to one-of-one resale stock; affiliation and setting must remain visible.
- Product structured data Google · Tier A · current documentation · live documentation
Google Search product structured-data eligibility and representation. Limit: Google documentation governs Google Search features and does not establish behavior in other answer or shopping systems.