Brand salience and the shopping agent that, reportedly, picks famous first

Brand salience meets a shopping agent: what Muse actually does#

The fox has a rule for new gadgets: find out what it does before you find out what it is said to do. Meta announced Muse for small business on 29 September 2026, and its own post stresses control: nothing publishes, sends, or spends without your approval. TechCrunch reported that Muse can also link to Instagram professional account analytics, Facebook pages, and Meta ad accounts. That is the merchant side of the story.

The shopper side is the part brand teams should read twice. A personal agent browses, compares and shortlists for someone who may never see a search results page. In a test by the AI-visibility vendor Profound, 50 prompts from a single consumer retail category went through one Muse instance. Meta's product catalog returned roughly 60 products from 30 merchant sites, and the final answer typically contained 6 product cards: 2 from the catalog and 4 from the browser.

That chart is a derived shape, not a measured survey: the 95 percent comes from PPC Land's report of a median catalog share of candidates, and the 33 and 67 percent are simply two cards and four cards out of six. It still carries the lesson. The place a brand is listed and the place a brand is chosen are different places, and the second one is open to the live web.

What is reported, and what Meta has confirmed#

Here is the sentence that travelled: Muse places famous sellers first. PPC Land reported that Muse places well-known sellers first and places a brand's own store first when a shopper asks for it by name. A second report said Meta's Muse agent must rank well-known merchants first in its top five picks, attributing the rule to an instruction file Profound is said to have extracted on 25 September.

Now the discipline. Meta's own announcement, as retrieved for this piece, says nothing about ranking merchants. Profound's own page describes the test and the catalog split, and does not contain the instruction-file wording that the trade reports attribute to it. So the accurate sentence is that trade reports, citing Profound's work, say Muse has such a rule, and that Meta has not publicly confirmed it. Write attribution only, and keep the denial on the same panel as the claim.

There is a second reason for care. Two regulatory commentators have argued that a biased algorithm, masquerading as neutral optimization, may be much harder to detect and contest than a biased salesperson. That is an opinion, not a finding, and it is the question this whole episode is really about: when an agent shortlists for you, who audits the shortlist?

Does AI really favour famous brands? Two preprints disagree#

The broader question has evidence behind it, and the evidence points in two directions. A June 2026 preprint by Chu and Hou studied skincare recommendations and found that, with identical specifications, well-known brands get recommended 100% of the time. A September 2026 preprint by Malthouse and colleagues, across six models and five product categories, found that category-only queries reveal substantial omission of established brands and limited evidence that recommendation prominence follows conventional brand popularity.

Neither is peer reviewed, and the designs differ: a controlled setup in one category against real-world category prompts across several. The second paper's twist is the useful one for brand teams. It reports that prominence is associated with broader marketplace-visibility signals, particularly search interest. Familiarity is not one number. It is a bundle of signals that a model, or an agent, can see.

Add two older anchors. EMNLP 2024 researchers found that models were disproportionately associating global brands with positive attributes. And Ahrefs, studying 75,000 brands, reported that brand web mentions show the strongest correlation (0.664) with AI Overview brand visibility, before adding its own warning that correlation is not causation.

• 100%

  recommendation rate for the well-known brand when specifications were identical (preprint)

• 0.664

  correlation between brand web mentions and AI Overview brand visibility (vendor study)

• 40%

  visibility boost reported for the best content methods on one benchmark (preprint)

That last figure comes from the original generative-engine-optimisation paper, which reports that GEO can boost visibility by up to 40% in generative engine responses. It measured content tactics on the authors' own benchmark, not brand familiarity, so it answers a different question. It is useful here because it shows that something other than fame moves the needle.

Brand salience vs brand awareness: the line an agent draws#

People who ask what is brand salience usually want a one-line answer, so here it is. Brand awareness asks whether someone knows you exist. Brand salience asks whether you come to mind, or to hand, at the moment a purchase is being decided. Marketing scientists use the phrase mental availability for the second idea, and it is the better test for an agent, because an agent has no feelings about you, only signals.

That is why a brand awareness strategy built on reach alone can look healthy and still lose the shortlist. The IPA's long-running work with Binet and Field leans on a simple line that share of voice matters more than ever, and the same logic applies to the signals an agent can see: how often, and in which places, your name appears next to the category.

Measure it the honest way. The Ehrenberg-Bass researcher Jenni Romaniuk has stressed testing recognition unprompted, because prompted tests flatter. As reported, prompted fame scores can be inflated by up to about 20 percentage points. An agent works unprompted. Your measurement should too.

What a smaller brand can do on Monday#

Be careful here, because the causal evidence is thin. The Advertising Research Foundation says in its early study that models tend to rely on narrow, repetitive sets of familiar brands, but that page states no method. Treat everything below as a hypothesis worth a test, not a guarantee. The same patient prowl runs through folkfox's brand strategy work and its SEO and GEO services.

For related thinking, see how a company can organise a portfolio around one name and why AI visibility is not the same as being remembered. Both are the same argument from different ends: the signal and the memory have to agree.

Think of salience as a scent trail through the undergrowth: every mention is a paw-print, and an agent follows the den that smells most of the category along the hedgerow. Brand salience is the name behind the click, and it is now also the name behind the shortlist. If the agent is going to read your name, make it the clearest word on the page.

Frequently asked questions#