Evidence · strength 0 of 5 · searched to 2 Nov 2026

Does AI help people find information for their work?

Not enough evidence yet to say whether AI helps people find information for their work.

What randomised trials found when people looked things up with an AI tool instead of a search engine, and why so few of them tested real work.

The studies How sure are we?

VerdictNot enough evidence yet

No good evidence. Studies so far point to a gain, but there are too few or too weak to rate it.

ImpactNot shown

We show no impact figure until at least two studies qualify.

Evidence strength0 of 5

There is not enough evidence to rate this yet.

Cost per user a year$$

These studies mostly used general AI assistants; business plans list about $20–$30 a user a month (checked 2 Nov 2026).

Pilot rating under our v0.3 rules. The rules are published on our method page but not yet independently reviewed, so ratings may change as the method is checked and new studies arrive. Our method.

What the studies found

No impact figure yet. Fewer than two studies qualify, so under our rules we do not show how big the effect is. The studies we found are listed below, with what each measured.

How sure are we?

Starting level0 of 5, from 1 qualifying study: fewer than 2.
Result0 of 5. There is not enough evidence to rate this yet.

Behind the average

By specialty (Castano-Villegas et al. (2026)): smallest in psychiatry (1.75% in Results; 2.7% in Discussion)
By question type (Castano-Villegas et al. (2026)): Research questions gained most, general knowledge least; the percentages differ between the Results (9.47% and 2.95%) and Discussion (8.65% and 2.87%) sections

How, where and why it works

Why it works

1 of 4. AI gives one written answer in place of a list of links, which saves reading and hides where the answer came from.

For whom and where

1 of 4. Lab studies found people did worse on tasks where the AI was wrong, and few checked.

What it takes

0 of 4. No study followed people finding information in their real jobs.

The full cost

0 of 4. No study counts cost.

Four separate scores from 0 to 4, never added up: how much the studies tell us about why it works, for whom, what it takes and what it really costs (the EMMIE method).

What to do next

Measure your pilot →

If you are trying it yourself

Ask for the sources, then open them.
Check names, dates and numbers against the original.

If you are rolling it out

Test the tool on questions with known answers from your own work.
Tell people which questions must be checked at the source.

What it takes

Open the source behind any answer you plan to rely on.
Keep a search engine in the routine for facts that must be right.

What can go wrong

In lab studies, people using an AI tool were faster but much less accurate when the AI was wrong, and few checked.
The only counted study was run by the company that makes the tool.

The studies

Study	Grade	Design	People	AI model (released)	Company role
Clinical Med students' validation of Arkangel AI: Are their responses any better when supported by the AI? Natalia Castano-Villegas, Isabella Llano, Maria Camila Villa, Jose Zea, Laura Velasquez, 2026. medRxiv, posted 9 January 2026 (preprint)	C	RCT, lab or benchmark	83 clinical-years medical students randomised in Colombia: 43 to Arkangel AI (group A), 40 to traditional search (group B). Excluded: 2 in A (platform error, answered nothing), 1 in B (used ChatGPT), 1 incomplete quiz in A for validity and acceptability. 1,328 question-answer pairs from 4 fictional cases	Arkangel AI (retrieval through Tavily with several LLM workflows; model names are in Table 4, an image, and could not be read)	vendor-run: All authors are employees or founders of Arkangel AI (one author is its CEO, another its president)

Grades: A, a well-run randomised trial in real work; B, a strong natural experiment in real work or a well-run trial on realistic tasks; C, a trial with clear limits or a matched comparison; D, before-and-after, surveys or case studies. How studies are graded.

Studies we found but do not count

Early LLM-based Tools for Enterprise Information Workers Likely Provide Meaningful Boosts to Productivity (Copilot Information Retrieval Study only) (Cambon et al. (2023), grade C). The report gives no effect size for this task, and the full paper could not be opened.
Effects of LLM-based Search on Decision Making: Speed, Accuracy, and Overreliance (arXiv title: Comparing Traditional and LLM-based Search for Consumer Choice: A Randomized Experiment) (Spatharioti et al. (2025), grade C). Not work: paid online workers choosing between cars, with one task built so the AI errs. Shown as context, not counted (section 4, rule 1).
Experimental evidence of the effects of large language models versus web search on depth of learning (Melumad and Yun (2025), grade C). Not work: members of the public and students learning about everyday topics. Shown as context, not counted (section 4, rule 1).
ChatGPT vs. Google: A Comparative Study of Search Performance and User Experience (Xu et al. (2023), grade C). Not work: members of the public answering trivia, finding flights and fact-checking a claim. Shown as context, not counted (section 4, rule 1).
Google it or ask ChatGPT? Impact of software developers' information search practices on organisational information security (Paz-Chen et al. (2026), grade D). Only the abstract could be read, and the outcome is what developers shared with the tool, not whether they found information.

These studies are real and often cited, but they fall outside our rules for this rating. What counts as a study.

Does the rating hold up?

Run without company-run studies0 studies: strength 0, verdict no good evidence.
Run without working papers and preprints0 studies: strength 0, verdict no good evidence.

How up to date is this evidence?

Method →

Evidence searched to 2 Nov 2026. The studies do not all state model release dates. Rating computed 5 Nov 2026 under Module 15 v0.3 (pilot).

How this page was made: AI research agents searched for studies and read each paper; a second pass checked every number on this page against the paper itself, with the quote and where it appears. The rating is computed by fixed rules, not chosen. No person outside AI Adoption has reviewed it yet. Editorial owner: AI Adoption editors. Spot a mistake? Corrections.

Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. No company pays for or has any say in a rating.