A program of EffectiveSC
How We Use AISources checked October 24, 2026 to November 12, 2026
Where things stand / What people use AI for / Use cases by job function
Which business functions use AI most, and what software, support, marketing, creative, operations, finance, hiring and research teams do with it, with named companies' own results.
Official surveys in the United States, the European Union and Japan rank functions differently, but sales and marketing sit at or near the top of every list. Most US firms that use AI use it in only a handful of functions.
52%
Among US firms that used AI in at least one business function (November 2025 to January 2026), the most common functions were sales and marketing (52%), strategy and business development (45%) and IT (41%).
Who was asked: US employer firms in the Business Trends and Outlook Survey's 2026 AI supplement
Counts firms, not workers; weighted by employment, IT comes first.
34.70%
Among EU enterprises (10+ employees) that used AI in 2025, 34.70% used it for marketing or sales and 31.05% for business administration or management; logistics was the least common purpose, at 6.08%.
Who was asked: EU enterprises with 10 or more employees
around 70%
In MIC's survey of managers at Japanese firms with 10 or more employees that had begun digitalizing (January to February 2026), around 70% said their firm uses generative AI to help prepare minutes and emails, the most common kind of work. The share who felt an effect was also highest there.
Who was asked: Managers at firms with 10 or more employees that began digitalization by 2025
From a survey company's online panel, not a random sample.
57%
Most US firms that use AI use it in few functions: surveyed from November 2025 to January 2026, 57% of them (firm-weighted) used it in only one to three of the 15 business functions asked about.
Who was asked: US employer firms in the Business Trends and Outlook Survey's 2026 AI supplement
Counts each firm equally; weighting by employment gives somewhat higher shares for many functions.
Developers use AI most to write and debug code in areas they already know. Companies report large shares of AI-written code by their own definitions, while controlled trials range from a solid gain to a slowdown.
67%
In Stack Overflow's 2026 developer survey (June to August 2026), the most common uses of AI were writing code in a familiar area (67% currently use AI for it) and debugging or refactoring (61%).
Who was asked: 30,903 valid responses from 169 countries
Opt-in survey of people who use Stack Overflow.
Disclosure: Stack Overflow sells developer products and data access, including to AI companies.
75%
Google says 75% of all its new code is AI-generated and approved by engineers, up from 50% in fall 2025 (April 2026).
Who was asked: Google's own teams, as the company states it
Google's own measure, with no method given; not comparable with other companies' code shares.
Disclosure: Google sells the Gemini models and the cloud AI products the post promotes.
62.1%
GitHub reported in July 2026 that 62.1% of engaged Copilot Business and Enterprise users were still mainly using code completions rather than its agent features.
Who was asked: Engaged Copilot Business and Enterprise users across GitHub's enterprise customers
GitHub's own telemetry, with no count; its 'code first' group also counts agent mode in the editor.
Disclosure: GitHub, which Microsoft owns, sells Copilot.
26.08%
Across three randomized field experiments in 2022 and 2023 at Microsoft, Accenture and a Fortune 100 company, with 4,867 software developers, using the GitHub Copilot coding assistant raised the number of tasks completed each week by 26.08%. The estimate is imprecise (standard error 10.3%).
Who was asked: 4,867 software developers at Microsoft, Accenture and an anonymous Fortune 100 electronics maker
Tested early code completion, not today's coding agents.
Disclosure: Two of the authors work at Microsoft, which owns GitHub Copilot.
19%
In METR's early-2025 randomized trial, 16 experienced open-source developers took 19% longer with AI tools, yet afterward believed AI had saved them 20%.
METR's later update suggests newer tools speed up returning developers, but METR calls that evidence weak.
See also: What agents have delivered
Customer contact is the most common use of AI in German firms. In large field studies, assistants helped newer and weaker agents most and did little or harm for the best.
Agents two months into the job with the AI assistant performed as well as or better than agents with more than six months' tenure without it.
0.283 points
In Alibaba's randomized field experiment with new Taobao after-sales chat agents in early 2024, an AI assistant raised customer ratings most for the weakest agents. For the top fifth it backfired: their ratings fell by 0.283 points on a five-point scale, and more of their customers were dissatisfied.
Who was asked: 5,940 new Taobao after-sales chat agents (tenure under one year)
Disclosure: Alibaba built the assistant, and Alibaba employees are among the authors.
nearly 45%
Airbnb says nearly 45% of support issues that begin with its AI assistant are resolved without a human agent (Q2 2026).
Who was asked: Airbnb guests whose support issues begin with its AI assistant
Airbnb's own figure, reported to shareholders, with no definition of resolved.
72%
Customer contact, such as handling inquiries, is the most common use of AI among German firms (20+ employees) that use it: 72% in 2026.
Who was asked: 603 German firms with 20 or more employees
Disclosure: Bitkom is Germany's digital industry association; its members sell IT and AI.
Marketing is among the most common business uses of AI in every survey. The results here come from companies measuring their own teams, not from controlled trials.
39%
Stripe reported in July 2026 that in weeks its account executives used Kai, an internal AI agent launched in April, they created 17% more opportunities, generated 26% more revenue opportunities and closed 39% more deals than the same sellers in weeks they didn't.
Who was asked: Stripe account executives, comparing each seller's weeks with and without Kai use
Not randomized: sellers may use the agent more in busier weeks.
70%
Google said in April 2026 that its marketers used its models to make thousands of creative variants for the Gemini in Chrome launch, with 70% faster turnaround and a 20% increase in conversions.
Who was asked: Google's own teams, as the company states it
One campaign, Google's own account, with no comparison method given.
Disclosure: Google sells the Gemini models its marketers used.
54%
54% of German firms that use AI use it in marketing and communication (2026), second only to customer contact.
Who was asked: 603 German firms with 20 or more employees
Disclosure: Bitkom is Germany's digital industry association; its members sell IT and AI.
Making images and video is now one of the most common business uses of AI in the UK. In a writing experiment, AI ideas made each story more novel but made the stories more alike.
16%
In June 2026, 16% of UK businesses with 10 or more employees used AI for visual content creation, second only to large language models (18%).
Who was asked: UK businesses in the Business Insights and Conditions Survey
Some industries, such as finance and public administration, aren't surveyed.
8.1%
In a preregistered writing experiment published in 2024, short stories written with access to up to five GPT-4 story ideas were rated 8.1% more novel than stories written without AI; access to one idea raised novelty by 5.4%.
Who was asked: 293 UK writers on Prolific writing an eight-sentence story
A short online writing task, not professional creative work.
In a writing experiment published in 2024, stories written with GPT-4 story ideas were more similar to one another than stories written without AI. The rise in similarity equaled 10.7% (one idea) and 8.9% (up to five ideas) of the range seen among stories written without AI.
Who was asked: 293 UK writers on Prolific writing an eight-sentence story
Similarity was measured by comparing the stories' text embeddings.
80%
In Adobe's September 2025 press release, IBM's senior vice president of marketing and communications said Adobe Firefly cut content costs by 80% in a previous IBM campaign and shortened ideation from weeks to two days.
Who was asked: One IBM marketing campaign
One campaign, described by a customer in the vendor's press release, with no baseline given.
Disclosure: Adobe sells Firefly and issued the release.
Improving day-to-day operations is the most common business use of AI among UK firms that use it. In the EU, large firms use it for security far more than small ones, and a US utility credits its monitoring, which uses machine learning, with avoided fires and outages.
close to 60%
In June 2026, close to 60% of UK businesses that use AI said they use it to improve business operations, the most common purpose in every size band.
Who was asked: UK businesses in the Business Insights and Conditions Survey
Businesses could name more than one purpose; some industries, such as finance, aren't surveyed.
28
PG&E says its continuous grid monitoring avoided 28 potential ignitions in high fire-risk areas and 19.6 million outage minutes from January 2025 to June 2026, and saved $11.2 million through lower-cost repairs. The monitoring is now run from a center that combines people, data and machine learning.
Who was asked: PG&E's electric distribution grid
PG&E's own estimate of what was avoided, credited to its monitoring as a whole, not to machine learning alone.
47.51%
In 2025, 47.51% of large EU enterprises using AI used it for ICT security, against 14.51% of small ones, the widest size gap among AI purposes.
Who was asked: EU enterprises with 10 or more employees
Finance and accounting are a common use of AI in European firms, while legal departments rarely use it. Few companies have published results for these back-office jobs, and almost none for HR.
23.11%
23.11% of EU enterprises (10+ employees) that used AI in 2025 used it for accounting, controlling or finance management.
Who was asked: EU27 enterprises with 10 or more persons employed
30 minutes
In an undated news card on its engineering site (seen October 2026), Uber says its Agentic Pods, which embed engineers alongside operators to find uses for AI, cut the time for capital allocation from 15 hours to 30 minutes.
Who was asked: Uber finance and operations workflows
Uber's own one-line summary, with no method or detail given.
8%
8% of German firms that use AI use it in their legal or tax department, and 7% in management (2026).
Who was asked: 603 German firms with 20 or more employees
Disclosure: Bitkom is Germany's digital industry association; its members sell IT and AI.
See also: Law: faster research, invented citations
Candidates use AI to write applications, and employers use it to write ads and screen them. Sounding polished is a small part of what both sides already get wrong: invented experience on one side, and jobs that are not as advertised on the other.
55%
55% of Kickresume users said they use AI to make their CV sound more confident and professional, and 24% use it mainly to save time. 4% said they had made up or inflated job titles, dates or qualifications.
Who was asked: 1,004 respondents worldwide, reached through Kickresume's user database in December 2025
Disclosure: Kickresume sells an AI résumé builder and surveyed its own users.
67 to 82 points
With only the résumé summary rewritten and the facts held fixed on 2,245 résumés, larger models favored their own summary over the human original by a gap of 67 to 82 percentage points.
Who was asked: 2,245 pre-LLM LiveCareer résumés cleaned from 2,484, covering 24 occupations
23% to 60%
In simulated shortlists across 24 occupations, candidates using the evaluator model were 23% to 60% more likely to be shortlisted than equally qualified candidates with human-written summaries, with the largest gaps in sales and accounting.
Who was asked: 24 occupations times 30 runs
1.8 of 25
In evaluations of cover letters, the worst writer-evaluator pairing displaced about 1.8 of 25 high-fit candidates through the writer model alone, even though writers were told not to invent credentials.
Who was asked: 11 writer models and 9 evaluators, 10 jobs and 50 candidates; 5,500 cover letters and 414,000 evaluations
A preprint experiment about cover letters, not résumé summaries.
17%
In HireRight's 2026 survey, 17% of North American respondents used AI for candidate selection or résumé filtering, the most common HR use, and 52% used no AI in HR at all.
Who was asked: The 17% and 52% are North American respondents
Disclosure: HireRight sells employment screening.
64%
64% of the US hiring managers Resume Genius surveyed said their company had misrepresented some aspect of a role or the hiring process, most often by understating how hard the job or the team would be, or by giving an inaccurate timeline.
Who was asked: 1,500 US hiring managers, surveyed with Pollfish from June 4 to June 6, 2026
Disclosure: Resume Genius sells résumé software.
90%
90% of the US hiring managers Resume Genius surveyed believe companies intentionally post job listings they do not plan to fill immediately, and 46% said their company uses AI to write job descriptions.
Who was asked: Same Resume Genius / Pollfish survey of 1,500 US hiring managers, June 4 to June 6, 2026
This is what managers believe companies do, not a count of listings.
Disclosure: Resume Genius sells résumé software.
10%
10% of job seekers in an AIResumeBuilder poll of 7,884 US adults admitted lying on a résumé, and 31% of those who lied said they used an AI tool to help write it.
Who was asked: 7,884 US adults, online via Pollfish, May 2025
Disclosure: AIResumeBuilder sells an AI résumé tool.
81%
In a June 2025 Kelly survey of more than 1,000 hiring managers, 81% had posted a job, run a full interview process, and then decided not to fill it, and 54% admitted misrepresenting remote work availability to secure hires.
Who was asked: More than 1,000 hiring managers and more than 1,000 job seekers in the United States
Disclosure: Kelly is a staffing company.
71%
71% of HR decision-makers in a Stepstone and HTW Berlin study say job ads often ask for more qualifications than the role actually needs.
Who was asked: HR decision-makers in a Stepstone and HTW Berlin study published October 28, 2026
Disclosure: Stepstone operates a job board.
See also: AI keeping us honest
Text edited by AI now shows up in a measurable share of biomedical research abstracts, and many firms use AI in research and development. Labs report models that run scientific simulations vastly faster.
at least 13.5%
An analysis of excess style words in PubMed estimates that at least 13.5% of biomedical abstracts published in 2024 were processed with large language models. The lower bound reached 40% for some groups of papers by discipline, country or journal.
Who was asked: PubMed biomedical abstracts, 2010 to 2024
A lower bound estimated from word frequencies, not a check of individual papers.
40%
40% of US firms using AI in a business function use it in R&D, fourth after sales and marketing, strategy and IT (November 2025 to January 2026).
Who was asked: US employer firms in the Business Trends and Outlook Survey's 2026 AI supplement
Counts firms, not workers.
six orders of magnitude
In a 2024 paper, researchers at the UK Atomic Energy Authority reported that a neural-operator surrogate model predicted simulated tokamak plasma dynamics six orders of magnitude (about a million times) faster than traditional solvers, with high accuracy.
Who was asked: Simulated (JOREK magnetohydrodynamic) and experimental (MAST) tokamak plasma data
A research result on simulated plasma, not a deployed system.
The AI at Work Benchmark
Score your organization's practices, measure your people with an anonymous pulse, and see both against the published evidence.