AcademicarXiv, 25 Sep 2026

Three larger Qwen3 models leaned toward later-listed hotels, picking the first one only 18.5% to 22.9% of the time

Channel mix & planning

First-shown hotel selection rate by LLM, models below 50%

First-shown hotel selection rate by LLM, models below 50%
Model	First-shown selection rate (%)
GPT-5.4 Nano	46.2%
Gemma3 1B	44%
Claude Haiku 4.5	43.4%
Phi-3 Mini	36.4%
DeepSeek-R1 7B	29.4%
Phi-2 2.7B	28.4%
Qwen3 30B-A3B	22.9%
Qwen3 4B	22.7%
DeepSeek-R1 1.5B	19.8%
Qwen3 8B	18.5%

Reader Note:: The share of tasks where the model picked whichever of the two reference hotels was shown first, pooled across both orderings; 50% means no position bias. Qwen3 30B-A3B: 22.9%, Qwen3 4B: 22.7%, Qwen3 8B: 18.5%. The smallest Qwen3 0.6B did the opposite at 98.5%. Synthetic NYC hotel tasks, not live shopping.
Evidence:: Experimental
Vendor interest:: None
Sample:: 28 LLMs from 8 providers tested on 3,600 hotel choice tasks built from 179 real New York City hotel properties; each task was scored in both original and swapped orderings.

Good for

Backing budget and pricing decisions in similar situations.

Not for

Situations far from the one tested, without checking again.
This was measured in one market. It may not hold where you are.
The source never says when this data was collected. It could be considerably older than the publication date suggests.
The two groups were not split at random — they picked their own side. The difference may have been there before anything was measured.

Paid media overtook martech as the largest slice of the marketing budget between 2018 and 2025, rising from 23.2% to 30.6% while martech fell from 29.2% to 22.4%

Change in marketing budget allocation by resource, 2018 to 2025
Resource	Change 2018 to 2025 (percentage points)
Paid media	7.4
Labor	-2.6
Agencies	-2.3
Martech	-6.8

33.8%

of US Fashion site visits now come from paid channels (Search + Social + Display), overtaking Direct traffic's 27.8% share for the first time in three years (Jul 2023–Jul 2026).

Go deeper on this topic

Our Substack turns the data behind these cards into full breakdowns: what changed, why, and what to do about it.

Fixing your influencer marketing campaign: How to get into the top 10% of advertisers

53.9% of US marketing leaders say marketing keeps influencing a customer for a year or longer, though the most common single answer is several months at 24.4% and the report's median is six months

Marketing Performance: Perceived marketing influence duration on customers, as reported by US CMOs in 2026
Duration	Percent (%)
An hour	0.6%
12 hours	0.6%
A day	1.3%
A week	2.6%
A month	6.4%
Several months	24.4%
6 months	10.3%
1 year	21.8%
2 years	18.6%
5 years	9%
7 years	0%
10 years or more	4.5%

Open full graph

9.6%

of overall budget and company revenues go to marketing on average (2026 survey of US companies)

SEO, social, email, retail media and influencer all rank higher on impact than on the budget they receive

Marketing budgets ranged from 5.8% to 9.7% of company revenue across industries in 2025, with consumer products highest and IT and business services lowest

Marketing budget as a percent of total revenue by industry, 2025
Industry	2025 (%)
Consumer products	9.7%
Manufacturing	9.5%
Pharma	9%
Media	8%
Insurance	7.5%
Financial services	7.2%
Retail	7.1%
Tech services and products companies with over $250M revenue	7.1%
Travel and hospitality	6.7%
Healthcare	5.9%
IT and business services	5.8%

Open full graph

Three larger Qwen3 models leaned toward later-listed hotels, picking the first one only 18.5% to 22.9% of the time

Good for

Not for

Related cards

Go deeper on this topic