Does being named first make a brand more likely to survive the switch between memory and web search

One natural question about Claude's brand recommendations is whether the brands it leads with, the ones named first in an answer, are more stable than the brands that show up further down the list. If a brand is genuinely well-known or well-matched to the query, you might expect the model to name it first regardless of whether it searched the web, and to keep naming it first. This study tested that by splitting all brand mentions into two segments: mentions where the brand was named first at least once, and mentions where it was never named first.

The split works like this. Across 33,734 total brand mentions, 4,712 fall into the "named first at least once" segment, and 29,022 fall into "never named first." Within each segment, mentions are again classified into the same three buckets used throughout this study: named only from memory, named only with web search, or named under both conditions.

For brands named first at least once:

• Named only from memory: 28.1% (1,323 of 4,712)
• Named only with web search: 38.9% (1,832 of 4,712)
• Named with and without web search: 33% (1,557 of 4,712)

For brands never named first:

• Named only from memory: 36.2% (10,502 of 29,022)
• Named only with web search: 50.6% (14,693 of 29,022)
• Named with and without web search: 13.2% (3,827 of 29,022)

The overlap share for brands named first at least once (33%) is higher than the overlap share for brands never named first (13.2%), so there is a real difference, in the direction you'd expect: brands that get top billing at least once are somewhat more likely to be among the stable, cross-condition mentions. But the gap is a matter of roughly one segment outperforming the other by around twenty percentage points of overlap share, not a case where position-one brands become reliably consistent. Even among brands named first at least once, two out of three mentions still fall into one of the only-one-condition classes, meaning search-only (38.9%) still substantially outweighs the overlap class.

This matters for how you read "being named first" as a signal. It is tempting to treat a first-position mention as proof that a brand has a durable, search-independent reputation with the model. The data suggests that's only partly true: being named first does correlate with somewhat more cross-condition stability, but most first-position mentions are still condition-specific. A brand can lead the pack in a search-on answer and be absent entirely once search is off, or vice versa. The segment split narrows the gap between the "only with web search" and "only from memory" classes slightly compared to the full dataset, but it does not close it, and it does not change which class is largest: search-only mentions remain the biggest bucket in both segments.

One limitation worth naming: this study did not track rank position beyond first versus not-first, so it cannot say whether second-position or third-position mentions behave differently from mentions further down. A finer-grained position analysis might reveal a smoother gradient, but that would require re-cutting data this study did not capture at that resolution.