Recommendation Intelligence Research™ · Study #44

Match the sponsored listing to the search term, and the penalty overshoots by 21 points.

Direct follow-up to the Sponsored Tag Penalty (Study #43, -18.4pp). One outside paper claims sellers can partially defend against the penalty with query-conditional descriptions, text that adapts to the exact search term instead of staying static. We tested it on the same 4 brands, changing one sentence. It didn't just recover the penalty. It blew past the organic baseline entirely: static sponsored 55.8%, organic reference 74.2%, query-matched sponsored 95.2%.

Atom Foundry · September 2026 · Tests Question 2 of the "Sponsored tag penalty" candidate, direct follow-up to Sponsored Tag Penalty (Study #43), part of the Recommendation Intelligence Research™ series

Cite this study: 10.5281/zenodo.23350889

+39.4ppRecovery vs. static sponsored

95.2%Query-conditional win rate

214.1%Of the organic-static gap closed

4Brands, all 4 same direction

Where this comes from

Question 1 confirmed the penalty. This is Question 2

Sponsored Tag Penalty (Study #43) confirmed that marking a listing "Sponsored" instead of "a standard search result" costs a brand 18.4 points in a forced two-way comparison, cluster mean across 4 brands. Allouah, Besbes, Figueroa, Kanoria, and Kumar (ACM Web Conference 2026, arXiv:2508.02630), the same paper behind that finding, go one step further: they suggest sellers can partially defend against the penalty by making the listing's description query-conditional, adapting the text to the specific search term instead of running the same static copy for every query.

That's a testable, narrower claim than "sponsored listings lose." Does tailoring the description to the exact thing someone searched for actually move the needle, and if so, by how much?

The test

One sentence, same sponsored tag

Same 4 brands as Study #43, byte-identical: Colored Organics, BodyArtForms, Barbaro Mojo, Hearthloom, each against its existing competitor across 20 real buyer intents. The static-sponsored condition is reused directly from Study #43's judged data, zero new calls, since it's definitionally identical: same target, same generic claim, same "This listing is marked as a Sponsored result" line. The new condition adds exactly one sentence to that same sponsored listing, echoing the exact search term back into the description: "Built specifically for shoppers searching for {the intent}." Nothing else changes. The sponsored label stays in place in both conditions.

Model: gpt-4o, temperature 1.0

Brands: 4, reused from Study #43

Conditions: static sponsored (reused) vs. query-conditional sponsored (new)

Comparison: forced two-way vs. named competitor, LLM judge

New calls: 400 chat, 400 judge (static reused, 0 new)

Judge prompt: unchanged from Study #43 / Winner vs Loser

Spot-checked before publication. 5 random colored-organics responses (the brand with the largest swing) were read against the raw response text. All 5 check out, the model states its reasoning as something like "because they specifically highlight being built for shoppers seeking X," confirming it's responding to the added sentence, not an artifact. The sponsored-tag sentence was also confirmed present, unedited, in every system message.

The result

Not a recovery. An overshoot

We expected query-matching to close some or all of the 18.4-point gap between organic and sponsored. It closed all of it, plus 21 more points on top of the organic ceiling. A single sentence matching the exact search term made a sponsored listing win more often than a plain organic one that didn't match the query at all.

Selection/winner rate by condition

Cluster mean across 4 brands · n=400 per condition

Static sponsored

55.8%

Organic (reference)

74.2%

Query-conditional

95.2%

+39.4pp

Per-brand breakdown

Same direction on every brand. One jump is dramatic

All 4 brands move the same direction, and two land right at the ceiling. The most striking move is the brand that had the most room to gain.

Brand	Static	Query-conditional	Recovery
Barbaro Mojo	62.0%	100.0%	+38.0ppat ceiling
BodyArtForms	97.0%	100.0%	+3.0ppalready near ceiling
Colored Organics	4.0%	83.0%	+79.0pp
Hearthloom	60.0%	98.0%	+38.0ppat ceiling

Colored Organics had the worst static score (4.0%) and the biggest jump (+79pp), but it's still the only brand not at or near 100% under query-conditional. Three of the four brands land at 98-100% once the description matches the query, meaning the effect is closer to a strong binary switch (the sentence is present or it isn't) than a smoothly scaling dial. Position bias check: target named first in the question won 98.4% (n=183) vs. 92.6% named second (n=217), a real but small gap next to the size of the main effect.

Why it matters

Not a defense against the penalty. A stronger signal entirely

Calling this a "defense" undersells what happened. A defense would neutralize the penalty and land back near 74.2%. Instead the sponsored, query-matched listing beat the plain organic listing by 21 points. That's the same shape as two earlier findings in this series: Authority Signal and Fact Injection both showed that a single confident, relevant claim can dominate a comparison regardless of other factors in play, and the Fragile vs. Resilient Candidacy follow-up found that any confident contextual claim lifted survival roughly equally, independent of its specific direction. Query-matching looks like another member of that family, a strong enough signal that whether the listing is sponsored or organic stops being the thing that decides the outcome.

Limitations

What this does not establish

4 brands

Same small cohort as the rest of this series, not a new or broader sample.

Template echo not real dynamic copy

The query-conditional line is a simple template that echoes the search term back, not the kind of dynamically generated description a real seller or ad platform would produce.

Reused data for the static arm

The static-sponsored condition comes from Study #43's separate run, not the same session as the new query-conditional data. Same model, prompt, and seed convention, but not a single paired run.

No organic + QC arm open follow-up

We didn't test query-matching on an organic (non-sponsored) listing. Without that arm we can't separate "this defends against the sponsored penalty" from "query-matching is just a strong signal on its own, independent of the sponsored tag."