LLM Product Discovery Benchmarks
Introduction
Across 2025, LLMs moved from early experimentation to a durable source of e-commerce discovery. Attribution depth has begun to compound meaningfully among brands already seeing LLM influence, and Q4 confirms that this shift is structural rather than transient.
By year-end, the data reflects a channel maturing along predictable lines. Brand-level adoption continues to expand gradually, while increased attribution share concentrates among a growing subset of brands. Usage intensity has stabilized at a higher baseline while platform and category patterns have sharpened. At the same time, traditional last-click measurement continues to miss the majority of LLM-driven influence.
The Q4 benchmarks clarify how LLM discovery is evolving and why accurate measurement increasingly depends on self-reported attribution, not just where the customer clicked last.
Key Findings
LLM adoption continues to expand, with more brands seeing mentions in Q4, though growth remains steady rather than explosive.
Attribution depth is concentrating, as some brands move into higher LLM response share bands.
The LLM ecosystem is increasingly multi-platform, with ChatGPT still leading but gradually ceding share to other models.
Category-specific behavior persists, reinforcing that LLM discovery does not function as a single uniform channel.
Last-click attribution still misses most LLM influence, leaving a significant measurement gap.
Q4 extends Q3’s steady, not explosive story on breadth, but with a meaningfully higher ceiling. In Q3, the share of brands seeing at least one LLM mention climbed from 15% to 18%. In Q4, that upward trend continued, with the share peaking around 21.2% in December.
The key nuance is that while adoption still appears gradual, the ceiling keeps lifting. LLM-driven discovery is spreading across more brands and becoming a more durable part of the acquisition mix rather than a short-lived experiment.
For marketers, this has a clear implication. If you’re already seeing LLM mentions, you’re not part of a temporary cohort. And if you’re not seeing them yet, LLM discovery continues to expand quarter over quarter across Fairing customers.
The quarter-over-quarter shift from Q3 to Q4 is subtle, but meaningful. In Q3, 64% of brands saw at least 0.5% of response share attributed to LLMs. In Q4, that figure declined slightly to 61%.
Rather than a lower LLM response share, some brands moved up the curve. In Q4:
17.3% of brands (up 0.9% from Q3) exceeded 1.5% LLM response share
1.9% of brands (up 0.5% from Q3) exceeded 5.5% LLM response share
1.5% of brands (up 0.8% from Q3) exceeded 6.5% LLM response share
In other words, the small decline at the 0.5% response share threshold reflects progression, not churn. Some brands that previously sat just above the minimum threshold have continued compounding into materially higher levels of LLM attribution.
The distribution still exhibits a long tail, but the center of mass continues to shift right. At this stage, the relevant question is no longer whether LLMs appear in attribution data, it’s how quickly their influence compounds once they do.
Over the year, normalized LLM growth increased materially, ending Q4 around 0.40, nearly double the 0.21 level seen at the start of 2025.
Quarter over quarter, this is one of the few metrics where Q4 looks more like consolidation than acceleration. Growth peaked at 0.46 in Q3, reached 0.44 in Q4, and settled at 0.40 by year end.
This is less a slowdown than a signal of maturation. Q4 maintains an elevated level of intensity rather than continuing the steep ascent seen earlier in the year, suggesting LLM discovery has reached a new, higher baseline of usage.
Q4 largely confirms and sharpens the category-level patterns that emerged in Q3. Automotive remains overwhelmingly Perplexity-led (93%), unchanged from the prior quarter. Consumer Electronics continues to be the most competitive category, with ChatGPT (37%), Claude (33%), and Gemini (21%) all playing meaningful roles.
Lifestyle-oriented categories remain ChatGPT-forward, but with increasingly visible secondary platforms. Apparel & Accessories shows a notable Grok share (15.8%), Food & Drug also sees Grok (13.5%), and Health & Beauty has Perplexity (19.7%) as a clear second player.
The practical takeaway is straightforward: LLMs do not behave as a single channel. In some categories, discovery is concentrated around one platform; in others, it resembles a diversified portfolio. Measurement and optimization strategies need to reflect that reality.
While ChatGPT still dominates LLM attribution in Q4, its share continues to decline quarter over quarter. Platforms like Perplexity and Gemini are steadily capturing more of that share, reinforcing the view that ecommerce discovery via LLMs is structurally multi-platform.
For marketers, this strengthens the case against treating “LLMs” as a monolithic channel. Testing, optimization, and content strategy increasingly need to account for differences across platforms rather than assuming one default experience.
When it comes to last-click attribution, Q4 looks almost identical to Q3. The majority of LLM-attributed transactions still arrive without identifiable last click UTM parameters, and when a last click UTM is present, it is attributed to Google roughly 20% of the time.
This reinforces a familiar pattern: Google often receives last-click credit for demand generated elsewhere, and LLMs are no exception. Without an alternative measurement approach, brands risk over-investing in lower-funnel channels that appear to convert, while underestimating the upstream influence of LLM-driven discovery.
For performance teams, the implication is not to abandon search, but to ensure keyword spend is truly incremental rather than capturing demand that would have materialized regardless.
LLM Discovery Is Compounding While Attribution Lags
Q4 confirms that the patterns observed in Q3 were not temporary fluctuations, but signs of a channel entering a more mature phase. LLM discovery continues to spread across brands, but more importantly, it is concentrating more deeply among those already seeing impact. LLMs are becoming structurally embedded in how consumers research and decide where to purchase, but the measurement challenge still remains.
Most LLM-driven influence sits outside of last-click attribution, meaning its contribution is systematically understated in traditional performance reporting. As LLM usage continues to compound, so does the risk of misattributing demand to downstream channels.
For high-GMV e-commerce brands, the takeaway is clear: LLMs now meaningfully shape consideration and purchase behavior, and evaluating their impact requires measurement approaches that capture influence even when click data breaks down.
