How visible is a retailer in AI-generated answers? A single score makes the question look straightforward. The underlying category results tell a more complicated (and more useful) story.
Table of Contents
In Target’s public RankLens report, displayed discovery ranges from 50% for beauty and personal care to 100% in several other tested contexts. Columbia Sportswear, by contrast, records appearances in all four reported samples for each of its five tested contexts. These are observations within particular reports, not verdicts about either company’s overall marketing performance. But they illustrate why measuring the brand alone can conceal the categories that need attention. [2] [3]
SV examined all 83 profile records in RankLens’ Retailing table. The unweighted mean of their displayed discovery scores is 45.9%, and 45 profiles—54.2% of the roster—display a score of 50% or less. Those figures describe the dashboard’s profile scores. They do not mean that AI omitted these retailers from half of all shopping conversations. Some underlying rows have unresolved sample denominators, making that stronger interpretation unjustified. [1] [4]
Scope and disclosure: RankLens is an SV product, and SV has a commercial interest in AI visibility measurement and marketing services. This is a descriptive analysis of public dashboard records, not an independent audit, consumer survey or controlled experiment. October 1, 2026 is the review date, not a verified date for the underlying model runs.
The retail distribution: useful context, not a market-share estimate
The sector’s median displayed discovery score is 45%. Fourteen profiles fall above 75%, while 12 display zero. Reporting the full distribution is more informative than selecting only a dramatic winner or a familiar name near the bottom: it shows how scores are spread across the particular roster being examined. [1]
| Displayed discovery band | Profile count | Share of roster |
|---|---|---|
| 0% | 12 | 14.5% |
| Above 0%, up to 25% | 13 | 15.7% |
| Above 25%, up to 50% | 20 | 24.1% |
| Above 50%, up to 75% | 24 | 28.9% |
| Above 75%, up to 100% | 14 | 16.9% |
| Total | 83 | 100% |
A retail team should resist treating this distribution as a league table of commercial success. The records are not weighted by sales, customer numbers or search demand. A company’s tested category phrases can also differ from another company’s. A higher score under one set of contexts does not establish superior performance under a common shopping task.
The better use is diagnostic. A dashboard can identify where to ask more specific questions: Which merchandise categories are represented? Which relevant contexts have completed observations? Where does the company appear consistently, and where is the evidence incomplete? These questions turn a broad visibility metric into a research agenda rather than a reputational judgment.
Target’s 50-point gap makes the category issue concrete
Target provides a useful within-profile example because each of its five displayed category rows reports four samples. The company appears in all four for three contexts, in three of four for home goods, and in two of four for beauty and personal care. Its displayed profile discovery score is 85%, matching 17 appearances across the 20 recorded sample slots. [2]
| Category phrase in report | Recorded appearances | Displayed discovery |
|---|---|---|
| Discount retail shopping experience | 4/4 | 100% |
| Everyday essentials and groceries | 4/4 | 100% |
| Home goods and decor | 3/4 | 75% |
| Beauty and personal care products | 2/4 | 50% |
| Clothing for all ages | 4/4 | 100% |
The gap between 50% and 100% is 50 percentage points. It is not evidence that the retailer’s beauty business performs worse commercially, nor does it identify a defect in its website. It is a reason to investigate whether the tested beauty context represents an important customer task and whether the observed difference persists under a larger, documented test.
This distinction changes who should participate in an AI discovery review. A brand team may reasonably care about overall recognition. A category team needs to know whether the retailer is considered for the specific products and needs it is responsible for. An aggregate result can be satisfactory for the first purpose and insufficient for the second.
The practical hypothesis is that category-level reporting will reveal opportunities hidden by company averages. This dataset illustrates that possibility; it does not prove that publishing additional category content will improve a score. Testing that intervention would require baseline observations, a defined change and a comparable follow-up period.
Columbia shows what complete observed coverage looks like
Columbia Sportswear’s five category rows each show four appearances out of four. Across the report, that is 20 recorded appearances in 20 sample slots and a displayed discovery score of 100%. The contexts cover outdoor clothing, performance gear and related apparel needs. Its displayed average rank is 5.8, demonstrating that complete recorded appearance and the highest list position are different outcomes. [3]
This is a useful counterexample to the idea that only a first-place result matters. A company can be consistently present in the tested answers without always occupying the earliest position. For a marketer, presence and prominence answer separate questions: was the company included, and how prominently was it presented?
The limits matter just as much. Twenty sample slots are not twenty verified independent shoppers, and five contexts are not the entire universe of purchase intentions. A 100% result in this report should be described as complete coverage of the recorded sample—not as a guarantee that an assistant will always mention Columbia.
It would also be premature to attribute that outcome to any particular optimization technique. The public table does not isolate the effects of content, external coverage, brand history, retrieval sources or model configuration. Those remain possible subjects for further investigation, not explanations established by this analysis.
A missing denominator is not a failed appearance
Amazon’s report illustrates a different issue. Its displayed profile discovery score is 50%, but two of its five entity rows show 0/0 rather than a completed set of non-mentions. The three other rows show 3/4, 4/4 and 3/4. Treating the two unresolved rows as eight additional failed responses would invent observations that the public table does not provide. [4]
This article therefore preserves the displayed score when describing the roster, while treating 0/0 as unresolved in response-level interpretation. It does not silently replace the score or reclassify the company as universally absent. A reader needs to know whether a zero describes a completed test, a reporting convention or unavailable evidence.
The broader reporting lesson is simple: completeness belongs beside performance. A dashboard that shows a percentage without the number of usable observations makes it too easy to confuse an apparent weakness with an incomplete measurement. Editors citing a visibility finding should request both the numerator and the denominator, alongside the exact scope of the test.
What retail teams should change in their reporting
Make the customer task the unit of investigation
A useful next-stage audit would begin with a deliberately chosen set of purchase tasks, not only a list of brand names. For a retailer with several merchandise businesses, the plan might separate replenishment, product comparison, occasion-based shopping and specialist requirements. These are proposed research categories, not additional observations from the RankLens data.
Each category should have an owner who can judge whether the questions are commercially relevant and whether an answer represents the retailer accurately. Otherwise, a team can optimize reporting around easily measured contexts while overlooking the tasks that customers actually care about.
Separate inclusion, prominence and factual representation
Report whether a brand appears before discussing where it ranks. Then review what the answer says. A mention is not necessarily a recommendation, and neither is automatically a source citation or a visit to the retailer’s website. Combining those events into a single success label obscures what was actually observed.
For retail teams, a proposed factual review should examine the attributes that could affect a purchase: the right company, the right category and any claims about product availability or service. This study did not audit those claims. Adding that review would make a future visibility report more useful than a larger collection of scores alone.
Fix the test conditions before comparing periods
A meaningful follow-up should preserve the question set, geography, language, model identification and retrieval settings where possible, and log any changes. Results collected under changed conditions should be labeled as such. A higher score after a content update is not evidence that the update caused the improvement when the measurement system also changed.
The IAB’s August 2026 AI visibility guidance distinguishes directional measurement from measurement suitable for consequential decisions and emphasizes disclosure and reproducibility. That is an appropriate lens for this snapshot: it can guide investigation, but it does not establish a business outcome or demonstrate compliance with the IAB framework. [5]
The opportunity is a better question, not a louder claim
The most useful retail finding is not that one company “wins AI.” It is that a company’s visibility can vary materially across its own reported contexts. Target’s complete-denominator rows make that variation visible. Columbia’s report distinguishes broad observed inclusion from first-place prominence. Amazon’s unresolved rows show why sample completeness must remain part of the story.
For a retail executive, the resulting question is more actionable than “What is our AI score?” It is: “For which relevant shopping tasks do we have reliable evidence of inclusion, and what should we investigate next?” A research program built around that question is less dramatic than a universal ranking—and more useful for making defensible decisions.
Methodology and limitations
SV transcribed the selected displayed fields for all 83 records in the supplied Retailing table and calculated an unweighted profile mean, median, threshold counts and score bands. The mean is the sum of displayed discovery percentages divided by 83; it is not a pooled percentage of valid model responses. Profiles and sector membership were retained as published, without independently auditing corporate status or imposing a new industry classification.
Company examples were selected to illustrate within-profile variation, complete recorded coverage and unresolved denominators, not through random sampling. Public labels identify ChatGPT responses, but complete prompt texts, raw outputs, model versions, run timestamps and sampling independence were not independently verified. Retrieved pages showed inconsistent month labels; the review date must not be substituted for the fieldwork date. With four recorded samples in a row, one changed appearance moves its rate by 25 percentage points.
RankLens’ Visibility Index is a provider-defined composite, not a financial or retail performance metric. This analysis does not reconstruct its weighting, interpret inconsistent rank-range fields, estimate statistical significance, evaluate customer behavior, or measure sales, traffic or revenue effects. Some zero-score records may reflect unresolved observations. The public pages can change after review; any later comparison requires a separately documented snapshot.
Sources
- RankLens: Retailing company-profile table. Selected displayed fields reviewed October 1, 2026.
- RankLens: Target brand-visibility report. Company and entity-level data.
- RankLens: Columbia Sportswear brand-visibility report. Company and entity-level data.
- RankLens: Amazon brand-visibility report. Includes unresolved 0/0 entity rows.
- IAB: Measuring Visibility in the AI Era. Published August 3, 2026; framework overview.
