Adoption: who actually uses AI

McKinsey's global survey work is the most-cited adoption yardstick, and the trend line is unambiguous: 78% of organizations used AI in at least one business function in early 2025 (up from 55% in 2023, with 71% regularly using generative AI) [1], and the latest survey — published as "The state of AI in 2026" — puts adoption at 88%, with 44% reporting enterprise-wide deployment. Yet most organizations remain stuck in pilot mode, and only a minority can point to clear bottom-line impact [2][3]. Money is moving in the same direction: US private AI investment reached $109.1 billion in 2024 — nearly 12x China's $9.3B and 24x the UK's $4.5B — out of a global total of $252.3B (+26% year over year), with generative AI capturing $33.9B [4]. Forecasters expect continued growth: IDC projects worldwide AI spending of $632 billion by 2028 [5], and Gartner put 2025's total at $1.5 trillion, including $644B in generative AI [6][7].

88%of organizations use AI in ≥1 function (McKinsey, latest survey) [2]
$632Bprojected worldwide AI spending by 2028 (IDC) [5]
$109.1BUS private AI investment in 2024 (Stanford AI Index) [4]

Productivity: what the controlled studies say

The headline claims survive contact with peer review, with important caveats:

The pattern across all four studies: AI compresses the skill gap — it helps novices the most, and it can hurt experts who over-trust it. That's the honest answer to "does AI actually improve productivity": yes, measurably, inside its competence frontier.

"Artificial intelligence benchmark" explained

A benchmark is a standardized test set plus a scoring procedure used to compare AI systems. The two you'll meet constantly: MMLU ("Measuring Massive Multitask Language Understanding"), which tests language models across 57 tasks from elementary math to US history, computer science and law [13], and MLPerf, the MLCommons industry-standard suite for measuring training and inference speed on real workloads — what NVIDIA calls "the leading industry benchmark for AI performance" [14]. Leaderboards built on these tests move fast: the Stanford AI Index recorded one-year gains of +18.8 points on MMMU, +48.9 on GPQA and +67.3 on SWE-bench [4]. Treat benchmark scores as directional, not definitive: models are routinely fine-tuned against public test sets, so a "record score" does not always equal better real-world behavior.

AI in healthcare & medical imaging

The "artificial intelligence x ray app" search usually means one of two very different things: consumer apps that chat about an image, or FDA-authorized medical devices. Only the second category is medicine. The FDA had authorized nearly 1,000 AI-enabled medical devices by October 2024 [15], and a longitudinal analysis of 1,430 AI/ML authorizations between September 1995 and December 2025 found radiology dominant: 1,094 authorizations (76.5%), with radiology, cardiovascular and neurology together covering 90.6% — and 331 authorizations in 2025 alone [16]. The FDA maintains the official public list of AI/ML-enabled devices [17]. The practical rule: a chatbot that "reviews X-rays" is not a medical device and must not be used for diagnosis — the devices that are approved come through hospital imaging workflows, not app stores.

AI in education: the split-screen moment

Students have adopted AI faster than institutions have policies for it. Pew Research found about a quarter of US teens used ChatGPT for schoolwork in 2024 — double the 2023 share [18] — and by early 2026, surveys reported more than half of teens using chatbots for schoolwork [19]. In higher education, 90% of college students report using AI [20]. Educators are more conflicted: a quarter of US teachers say AI tools do more harm than good in K-12 [21], even as RAND found 25% of ELA/math/science teachers and nearly 60% of principals used AI tools during the 2023–24 school year [22]. Our Writing & Study guide covers the honest-use playbook — disclosure, verification, and why detector scores are not proof.

AI in hiring: volume is the problem

The "best ai job application tool" question now cuts both ways. Job seekers use AI at scale — LinkedIn reported applicants submitting roughly 11,000 applications a minute, with AI tools a major driver [23], and recruiters describe "drinking through a fire hose" of AI-assisted applications [24]; Kickresume's data counts over 1.2 million job seekers using AI in 2025, mostly to beat applicant-tracking systems [25]. The honest advice: use AI to tailor a truthful application — never to fabricate qualifications — because AI screening on the employer side is getting better at spotting inflated resumes. For "review ai and get paid" searches: treat any tool promising guaranteed income as a red flag; real AI income comes from work output, not participation schemes.

Everyday AI tools worth knowing

Beyond the big assistants, the "best ai websites" lists converge on a few useful categories: AI directories that catalog tools (useful, but check the date — the space churns monthly), workflow tools like Ziflow for creative review automation, music-analysis tools for "ai review your song" feedback (fine for drafts; they don't replace a producer), and "best ai life coach" chatbots, which are best treated as structured journaling prompts, not therapy. Our rule across all of these: free tier first, verify any paid claim, and never share sensitive personal data with an unreviewed tool.

Regulation: the EU AI Act, explained in one screen

The EU AI Act is the world's most consequential AI law. It entered into force on 1 August 2024 and becomes generally applicable on 2 August 2026 [26]. The timeline that matters for businesses:

In the United States, Executive Order 14110 on safe and trustworthy AI was signed on 30 October 2023, directing safety testing, watermarking and the creation of the US AI Safety Institute [29]; it was revoked by Executive Order 14179 on 20 January 2025 [30]. The UK AI Safety Institute, founded after the November 2023 Bletchley Park summit, reports publicly on progress [31], and the International Network of AI Safety Institutes held its first meeting in San Francisco in November 2024 [32].

Using AI in China

"Best ai to use in china" has a clear practical answer, because most Western services are restricted or unavailable in mainland China. The main domestic options: DeepSeek — free chat app with about 22.2 million daily users, having overtaken ByteDance's Doubao [33], though it faces government scrutiny abroad over data security [34]; Qwen, Alibaba's open-weight model family (the subject of most "review qwen ai" searches) [35]; and Doubao, ByteDance's assistant. If data jurisdiction matters to you, the same caution we gave in the chatbot guide applies to every one of these apps.

Frequently asked questions

What is AI in simple words?+
AI is the field of making computer systems perform tasks that normally require human intelligence — understanding language, recognizing images, making decisions. Most modern AI is machine learning: systems learn patterns from data instead of following hand-written rules. See our home primer for the 60-second version.
How many companies use AI in 2026?+
McKinsey's latest global survey reports 88% of organizations use AI in at least one business function — up from 78% the prior year and 55% in 2023 — with 44% at enterprise-wide deployment [2][3].
Does AI actually improve worker productivity?+
In controlled studies, yes — with limits. Writing with ChatGPT finished tasks 40% faster (MIT/Science) [8]; BCG consultants completed 12% more tasks, 25% faster, at 40% higher quality inside AI's capabilities — but performed worse on harder tasks outside them [10].
How many FDA-approved AI medical devices are there?+
Nearly 1,000 AI-enabled devices by October 2024 [15]; a longitudinal analysis counts 1,430 authorizations between 1995 and 2025, with radiology at 76.5% [16]. The FDA publishes the official list [17].
When does the EU AI Act apply?+
It entered into force on 1 August 2024 and becomes generally applicable on 2 August 2026. Prohibited practices applied from 2 February 2025, GPAI obligations from 2 August 2025, and high-risk Annex I rules extend to 2 August 2028 [26].
What is an AI benchmark?+
A standardized test set plus scoring procedure for comparing AI systems. MMLU tests language models across 57 tasks [13]; MLPerf measures training and inference speed on real workloads [14].
What is the best AI to use in China?+
The main domestic options are DeepSeek (free chat app, ~22.2M daily users [33]), Alibaba's open-weight Qwen models [35], and ByteDance's Doubao. Western services like ChatGPT and Gemini are restricted or unavailable in mainland China.

Sources

  1. McKinsey, "The State of AI" (early 2025 survey PDF) — mckinsey.com
  2. McKinsey, "The state of AI" (latest, "The state of AI in 2026: On the road to ROI") — mckinsey.com
  3. FeeOnlyNews, "McKinsey's 2025 global AI survey: 88% of organizations now use AI" — feeonlynews.com
  4. Stanford HAI, "AI Index Report 2025" — hai.stanford.edu
  5. IDC, "Worldwide spending on AI forecast to reach $632 billion in 2028" — idc.com
  6. Gartner, "Worldwide AI spending will total $1.5 trillion in 2025" — gartner.com
  7. Nasdaq/Gartner, "Worldwide GenAI spending to reach $644 billion in 2025" — nasdaq.com
  8. MIT News, "Study finds ChatGPT boosts worker productivity for writing" — news.mit.edu
  9. Noy & Zhang, Science, "Experimental evidence on the productivity effects of generative artificial intelligence" — science.org
  10. Harvard Business School, "Humans vs. machines: untangling the tasks AI can and can't handle" (Dell'Acqua et al.) — library.hbs.edu
  11. NBER Working Paper 31161 (Brynjolfsson, Li & Raymond) — nber.org
  12. Peng et al., "The Impact of AI on Developer Productivity" (GitHub Copilot study) — arxiv.org
  13. Hendrycks et al., "Measuring Massive Multitask Language Understanding" (MMLU) — arxiv.org
  14. NVIDIA, MLPerf benchmarks — nvidia.com
  15. Healthcare Dive, "FDA AI medical devices growth", Oct 2024 — healthcaredive.com
  16. AuntMinnie, "Radiology dominates thirty years of FDA AI device approvals" — auntminnie.com
  17. FDA, "Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices" — fda.gov
  18. Pew Research Center, "About a quarter of US teens have used ChatGPT for schoolwork" — pewresearch.org
  19. Boston Globe, "Teens, chatbots, schoolwork", Feb 2026 — bostonglobe.com
  20. Forbes, "90% of college students use AI" — forbes.com
  21. Pew Research Center, "A quarter of US teachers say AI tools do more harm than good" — pewresearch.org
  22. RAND, "Teacher and principal AI use" (RR-A134-25) — rand.org
  23. eWeek, "AI job applications: LinkedIn" — eweek.com
  24. CNBC, "Recruiters are drinking through a fire hose of job applications" — cnbc.com
  25. Kickresume, "AI job search data" — kickresume.com
  26. European Commission, "AI Act: regulatory framework" — digital-strategy.ec.europa.eu
  27. DLA Piper, "European Union published Code of Practice" — dlapiper.com
  28. European Commission, "AI Omnibus enters force", Jul 2026 — digital-strategy.ec.europa.eu
  29. GovInfo, Executive Order 14110 — govinfo.gov
  30. OECD.AI, Executive Order 14179 — oecd.ai
  31. UK Government, "AI Safety Institute third progress report" — gov.uk
  32. European Commission, International Network of AI Safety Institutes — digital-strategy.ec.europa.eu
  33. Quid.ma, "DeepSeek dépasse ByteDance avec 22,2 millions d'utilisateurs quotidiens" — quid.ma
  34. Reuters, "South Korea's industry ministry temporarily bans DeepSeek", Feb 2025 — reuters.com
  35. Hugging Face, Qwen model family — huggingface.co