The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI that assists with drafting documents and improving writing quality through editing, style, and tone suggestions. Includes document drafting from outlines and readability improvement; distinct from content generation in Marketing which targets published content rather than personal professional writing.
AI writing assistance is a proven productivity category with mature tooling, validated ROI, and mainstream adoption — but its real-world impact remains tightly constrained and adoption paradoxes suggest the category has reached a stable equilibrium rather than further growth. The ecosystem spans dedicated vendors like Grammarly (40M daily users, $13B+ valuation, $200M+ ARR), platform-native offerings like Microsoft 365 Copilot (90%+ Fortune 500 adoption), and general-purpose LLM interfaces. Randomised controlled trials and field studies confirm bounded gains: Microsoft's RCT of 7,137 workers across 66 firms found 3.6 hours saved weekly on email alone, and practitioner analysis shows the category succeeds precisely because writing tasks have clear inputs and measurable outputs, unlike broader productivity claims. Consumer uptake is broad (63% of U.S. adults use AI tools monthly, writing assistance is the top use case) and sector-specific adoption is strong (73% college students, 89% marketing teams, 94% content agencies). Yet UC Berkeley peer-reviewed research (April 2026) reveals a critical paradox: heavy AI writing assistance use reduces argument coherence by 70% and erodes writer voice, while users report equal or higher satisfaction — suggesting a systematic misalignment between perceived writing quality and actual quality outcomes.
The structural barriers preventing tier advancement are both technical and behavioural. Hallucination rates in professional writing are escalating (18% in news-related prompts in Aug 2024 rose to 35% by Aug 2025, resulting in 12,842 articles pulled from circulation in Q1 2025), with legal domain showing 69-88% failure rates and 486+ court cases involving AI hallucinations. Enterprise adoption remains stuck in pilot limbo: only 16% of pilots reach production, and 73% of regulated-sector organisations have paused rollouts over security, data-privacy, and measurement gaps. Specialist vendors face intensifying pressure: Jasper AI collapsed with 40% layoffs as users migrated to free platforms; Grammarly's Expert Review feature (Aug 2025–Mar 2026) was disabled after 7 months following a class action lawsuit for using journalists' identities without consent; Writer.com abandoned SMB for enterprise-only positioning. Market consolidation data (22.5% CAGR, $4.8B→$20B 2025–2034) masks a permanent bifurcation: Fortune 500 firms and content professionals extract value in bounded tasks (drafting, summarizing, light editing), while mid-market and regulated organisations remain structurally blocked by accuracy, compliance, and measurement immaturity.
The vendor landscape is consolidating around two poles while specialist competitors exit or retreat. Microsoft continues expanding Copilot with Agent Mode for Word/Excel/PowerPoint and new draft-generation agents, reporting 90%+ Fortune 500 adoption. Grammarly ($700M+ revenue, 50,000 enterprise customers, 40M daily active users) is specialising vertically — shipping dedicated products for HR and customer-support teams with features like style-guide enforcement, term banks, and multilingual support. Both platforms are maturing: Grammarly reports 283% return for enterprise customers; Writer.com released 200+ Skills, observability/Datadog integration, and NVIDIA NIM support (Mar-Apr 2026). Specialist vendors face structural pressure: Jasper AI (once valued at $1.5B) collapsed with 40% layoffs in early 2026 as users migrated to free ChatGPT/Claude; Grammarly's Expert Review feature was disabled in March 2026 following a class action lawsuit for using journalists' identities without consent; Writer.com abandoned SMB markets and now targets enterprise exclusively at $18,000/year minimums. Market consolidation signals that writing assistance is moving toward general-purpose platforms and vertical specialization.
Adoption breadth masks deployment stall. Writing and communications account for 80% of all workplace generative AI use — the single largest application category. Sector-specific metrics show strength: 73% of college students use AI for writing, 89% of marketing teams use AI (up from 67% in 2024), 94% of content agencies embed it in workflows, and 74% of content marketers use AI tools. In professional services (legal, tax), 40% have deployed generative AI with 30% drafting time reductions. Yet enterprise conversion remains the critical bottleneck: only 19% of content marketing teams track AI-specific KPIs despite reporting 3.4x velocity and 67% cost reduction, indicating measurement immaturity; Copilot rollouts routinely stall at 20% adoption within organisations due to absent manager modelling, poor workflow integration, and unclear permissions. EU enterprises face additional compliance barriers: Grammarly Enterprise requires Data Processing Agreement (Standard Contractual Clauses only, no EU residency), DPA unavailable for Free/Premium tiers, and Works Council participation rights under German labor law, raising deployment friction for regulated sectors.
Accuracy and reliability remain limiting factors showing escalation. Hallucination rates in professional writing are worsening: news-related prompts doubled from 18% (Aug 2024) to 35% (Aug 2025), with 12,842 AI-generated articles pulled from circulation in Q1 2025. Enterprise testing of 6 AI content tools (March 2026) documented 15-25% hallucination rates. Legal domain shows systematic failure: 69-88% hallucination rate on specific legal queries, with 486+ court cases now involving AI hallucinations and 128 lawyers sanctioned (including Mata v. Avianca where ChatGPT fabricated 6 cases). Medical domain (Mount Sinai 2025): 64% hallucination rate on long clinical cases without mitigation. The paradox: UC Berkeley research shows heavy AI use reduces argument coherence by 70% and erodes writer voice, yet users report equal satisfaction—suggesting confidence masking actual quality loss. The market's next phase hinges less on feature expansion than on solving reliability (acceptable hallucination rates for high-stakes contexts), measurement (demonstrating ROI beyond early-adopter cohorts), and user-preference alignment (actual vs. perceived writing quality).
May 2026 platform developments reveal the "verification tax" shaping deployment reality. Google shipped tone/style personalization in Gmail (May 7, 2026), and Microsoft deployed Writing Tools natively in Notepad, confirming platform convergence on integrated drafting—yet field research shows this integration paradox: MIT's 1,258-person experiment validated 137% message volume and 20% editing time reduction, but independent studies document that 80% of recovered time is reabsorbed into review and quality control. Leaders report spending more time validating polished AI output than writing from scratch, and researchers identify that improved fluency masks errors and hallucinations rather than reducing them. In regulated domains (legal, healthcare, finance), verification burden blocks adoption entirely: 58% hallucination on legal queries and institutional shifts to sampling-based triage confirm that polished AI writing requires deeper human scrutiny, not less. The category has completed the "generation" phase and entered the "verification" phase—adoption now depends on whether organizations can afford the human overhead of validation rather than whether tools generate plausible text.
By June 2026, the tension between deployment breadth and reliability has crystallized into structural market bifurcation. Named deployments at Amazon, Google, and Apple show concrete productivity: product document drafting compressed from 3-4 hours to 30 minutes, meeting notes from 1-2 hours to 5-10 minutes, corporate emails from 1+ hour to 15-20 minutes (Business Insider, June 6). Enterprise adoption metrics reveal sustained real-world use (Redress Compliance analysis of 25-35 estates: 30-55% weekly active use, though effective cost-per-user runs 1.8-3x list price due to low-utilization seats; task-specific deployments drive 2-3x higher adoption than broad rollouts). Yet reliability crisis deepens: Stanford AI Index documents that 74% of surveyed companies now cite inaccuracy as their #1 AI risk (overtaking cybersecurity for first time); legal-specific tools hallucinate 17-33% on domain queries (Ninth Circuit ruling, June 3, formalizing professional liability and sanctioning lawyers for fabricated case citations). Institutional failures accelerate in June 2026: KPMG's October 2025 agentic AI report, published widely and cited in industry publications, contained 40 of 45 fabricated citations including false org deployments and contradicted KPMG's own 2025 CEO data; external forensic review (GPTZero) discovered the hallucinations after publication and retraction. Legal sector sees explosive adoption growth alongside intensifying failure signals: Association of Corporate Counsel survey (June 2026) documents 52% of US in-house counsel now use GenAI (up from 23% prior year—126% YoY growth), with 73% deploying for drafting and 82% expecting contract-drafting cost savings; simultaneously, approximately 900 AI hallucination incidents documented in US court filings since 2023, with 25+ federal courts now requiring standing orders for AI-use certification and citation verification before filing, and lawyers facing six-figure sanctions. Multi-model verification offers partial mitigation (61% hallucination reduction when combining Claude + Gemini independently), but production deployment in high-stakes contexts now requires systematic validation layers—shifting the discussion from capability to operational feasibility. SMB economics break down for small teams: independent testing shows free Grammarly + ChatGPT ($20/month total) outperforms paid Grammarly Business ($180/year minimum, justified only for 3+ writers with brand consistency requirements) for solo/2-person shops. The trajectory is clear: writing assistance has achieved production maturity and mainstream adoption in bounded, low-risk tasks (drafting, rewriting, summarization), but faces permanent structural barriers in regulated sectors, high-stakes contexts, and mid-market deployment economics—not through further capability advance but through fundamental verification and compliance friction.
— Association of Corporate Counsel survey of 657 in-house counsel: US adoption reached 52% (up from 23% prior year—126% YoY growth); 73% use AI for drafting, 53% for research; 82% expect contract-drafting savings and cost reduction.
— Named LegalTech deployment across five document types (NDAs, contracts, agreements) reduced per-document drafting from 2-4 hours to 30-60 minutes; 80%+ generated drafts validated legally usable by professional review panel across jurisdictions.
— Cites independent adoption surveys showing corporate legal acceleration: ACC/Everlaw 23%→52% (123% YoY growth), FTI/Relativity 44%→87% (98% YoY growth); in-house adoption outpacing law firms; usage concentrated in drafting, review, redlining with hallucination risks on citations.
— Comprehensive analysis documenting approximately 900 AI hallucination incidents in US court filings since 2023 with escalating sanctions (now six figures); 25+ federal courts adopted standing orders requiring AI-use certification and citation verification.
— Major enterprise case: KPMG's AI-drafted agentic AI report contained 40/45 hallucinated citations with fabricated org deployments and contradicted KPMG's own prior data; report retracted and misinformation propagated to other AI systems.
— High-credibility fact-checking rejecting unverifiable '$67.4B' cost claims; verified benchmarks from Stanford/Vectara: legal AI tools hallucinate 17-33%, general LLMs 1-30% on summarization, 34% more confident when false; Air Canada tribunal established corporate liability for bot-generated statements.
— Named employees at Amazon, Google, Apple report concrete writing task time savings: document drafting 15-20 min vs 1+ hour, meeting summarization 5-10 min vs 1-2 hours, product documentation 30 min vs 3-4 hours—validating real deployment and productivity gains.
— Large-scale study (480M outputs) shows multi-model verification reduces hallucination from 8.3% to 3.2%; Claude + Gemini combo lowest error rates; demonstrates viable production mitigation though not complete elimination.