The BaseResearcharXiv
How Web Research Agents Interpret User Priorities
According to its abstract, a preprint by Robin Sharma and Parisa Suchdev examines how two web research agents, built on OpenAI and Anthropic models, interpret a stated priority such as growth or resilience. In 36 runs comparing Microsoft and Alphabet, the authors report that the agents chose differing criteria and sometimes favoured different companies.
Why it matters
The same thing, for different readers.
What it means for you
A study reports that AI research assistants told to prioritize 'growth' can mean different things and recommend different companies, so users may not see the criteria applied.
For researchers
Sharma and Suchdev report that in 36 runs, 21 of 24 prioritized reports focused on the priority, but criteria varied; they argue evaluations should examine the criteria behind judgements.
How the finding was reached
In our words, from the source. Weigh the finding by how it was made.
- What was done
- 36 runs of two web research agents, built on OpenAI and Anthropic models, comparing Microsoft and Alphabet under a growth priority, a resilience priority, or none; reports, searches and sources were analyzed.
- What it found
- The authors report that 21 of the 24 reports given a priority focused on it, but criteria varied; under growth, the OpenAI reports that ranked the companies favoured Microsoft while the Anthropic ones favoured Alphabet.
- What it does not show
- One comparison task and two agents; of the search and source measures tested, only the Anthropic agent's search wording differed detectably by condition.
- Status
- A preprint: not yet peer reviewed
The record
- Questions
- How is AI changing childhood, health and daily life? · What happens to work, income and the economy?
- People
- Robin Sharma (Wrote it) · Parisa Suchdev (Wrote it)
- What we read
- We read the abstract.
- Found by
- The Radar, run of November 18, 2026, via
arxiv-ai-safety-query(tier 1: the source’s own channel or an official record) - Cite this
- arXiv, “How Web Research Agents Interpret User Priorities”, November 17, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/arxiv-2026-11-17-web-research-agents-priorities