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 for finding, synthesising, verifying, and preserving organisational knowledge. Mostly leading-edge: literature review, competitive intelligence, and knowledge management tools are maturing quickly with five practices actively advancing. The main constraint is hallucination risk — fact-checking and source verification still require human oversight in high-stakes contexts.
The headline: AI research tools are now everywhere, and they still cannot reliably tell you where they got a fact. The fix that works is a design change, not a better model: make the tool cite as it writes.
Most organizations now have AI research tooling switched on by default — summarization in email, meeting notes, document search over internal files. Adoption is no longer the question. A small group is pulling ahead by spending on verification rather than capability: constrained source pools, checks against original documents, and a named human signing off. The rest are exposed, because courts, academic journals, and regulators have started treating unverified AI output as organizational negligence rather than an individual mistake. If you cannot say today who checks the sources in an AI-assisted document before it leaves your building, that is the gap worth closing this quarter.
Fabricated references in medical literature reached 1 in every 277 papers. A Lancet audit of 2.5 million papers found a twelvefold rise since 2023, with almost no publisher acting on the papers flagged. Any document your organization publishes that cites sources now needs those sources checked against the originals, not against another AI tool.
The tools sold to catch fake citations failed their own test. An independent evaluation of five leading citation-checking products found all of them unsuitable for use without a human reviewing the output — those tuned to avoid false alarms miss real fabrications, and those tuned to catch everything bury reviewers in noise. Buying a detector does not discharge the duty; budget for the review time it creates.
One design change produced genuinely clean numbers. Google's Science One approach retrieves each citation at the moment the text is written, rather than writing first and attaching references afterward: zero fabricated references across 337 citations, against 21% for the standard approach. Source-grounded tools show the same pattern, hallucinating (confidently making things up) far less than general chatbots on identical material. Ask any vendor pitching research AI which of these two architectures they use — the answer predicts your error rate.
Plain keyword search beat expensive AI retrieval at scale. A controlled study across corpus sizes up to 600 million words found conventional keyword search overtaking AI-agent-driven search past a certain volume and holding a roughly twenty-point accuracy lead. If a vector-database migration is on your roadmap, this is the moment to re-examine the business case.
Buyers are pulling back on cost, not capability. A survey of 2,145 leaders found 49% had narrowed, delayed, or paused rollouts of agentic AI (software that acts on its own without being prompted) because costs outran value. The teams still funded are the ones that can show a measured baseline; the ones that cannot are being cut.
Verification is becoming a filing requirement, not a best practice. Florida's courts now require signers to certify that cited authorities exist; a major academic conference desk-rejected over a hundred already-accepted papers for invented references; documented court cases involving AI-fabricated citations have passed 1,700 across more than forty countries. Put a named-signatory sign-off standard in place for every externally filed AI-assisted document.
EU anti-money-laundering rules take effect across all 27 member states in July 2027. They mandate continuous transaction monitoring and explicitly permit AI tools with effective human oversight. Financial services firms should start the model-validation and audit-trail work now — an eleven-month build against a fixed date is not a plan.
Patent authorities are tightening the human-contribution test. The five largest patent offices have formed a joint AI working group, and Japan has raised its inventive-step standard to require evidence of genuine human-directed technical choices beyond what an AI-assisted specialist could reach. R&D and IP teams should begin documenting human contribution to inventions now, not at filing.
Checking the work costs as much as doing it. The best automated citation checker tested on legal briefs reached only 60.5% overall accuracy, at nearly seventeen reasoning steps per document. Among intellectual property professionals, 88% report spending half their time reviewing AI output — the time saved on the first draft reappears in the review.
Tools tuned to miss nothing will flood you with noise. In corporate transactions, 20–35% of items flagged by leading AI diligence platforms turn out to be dead records — dissolved subsidiaries, expired liens, terminated consents — adding more than $36,000 per deal in unbudgeted legal time and inflating timelines against faster bidders.
Having the tool is not using it, and using it is not benefiting. Seventy-two percent of knowledge workers have access to AI meeting tools; 41% use them monthly. One analysis found AI meeting summaries omitting 97% of significant decisions and actions — a quiet failure no accuracy metric catches, because nothing in the summary is wrong; it is simply missing.
Go deeper: the full Research & Knowledge briefing — the longer analytical write-up, plus every practice we track in this domain with its maturity rating, the tools to consider, and the evidence behind our assessment.