Is your knowledge AI-ready? Find out before your people rely on it
You've deployed Copilot — or you're about to. The tool will only ever be as good as the documents beneath it. Our AI readiness review audits your contracts, policies and knowledge hubs so AI retrieves the right content, answers accurately, and your people can act on what it says.
Why AI gives wrong answers from your documents
When an AI assistant answers from your knowledge base, it inherits everything about the source: the ambiguity, the outdated versions, the buried conditions, the 70-word sentences. The answer just arrives faster, in a friendlier voice, with more confidence than the document ever earned.
Four failures stack:
retrieval (the AI surfaces the wrong clause or an outdated version because the knowledge was never structured for machine reading);
accuracy (the AI misreads what it found);
comprehension (a faithful answer still carries the source's complexity, so users can't act on it); and
interaction (users ask leading questions, the AI obliges, and human bias compounds machine bias into decisions that feel checked but aren't).
Most "AI readiness" work addresses the first failure only. Our review assesses all four, because a correct answer that nobody understands, or a fluent answer nobody verifies, fails just as expensively.
What the review covers
Your knowledge, scored for both readers. We sample your document estate — contracts, policies, procedures, knowledge hubs — and assess machine readability (structure, chunking, versioning, metadata, terminology consistency) and human readability (plain language, comprehension load, cognitive-diversity barriers) against our benchmarks.
Your AI, tested against ground truth. We put your actual AI system through realistic questions on your actual documents and score where its answers are right, partial, outdated or confidently wrong.
Your people, tested through the pipeline. Where scope allows, we run a live test: real users, real tasks, your AI, your documents — measuring whether they reach correct decisions and where the pipeline misleads them.
A prioritised fix plan. You receive a benchmarked scorecard, the failure points ranked by risk, and a roadmap — what to restructure, what to rewrite, what to retire, and how to design the interaction against bias. Fixes can be delivered by your team, or by ours through AI knowledge curation and plain language redesign.
Who this is for
Legal, risk, governance, knowledge and IT teams deploying AI over document estates — especially where a confidently wrong answer becomes a compliance breach or a liability. Our clients include banks, government agencies, health services, utilities, telcos and universities across Australia, Hong Kong, the UK and the US.
Why Inkling Legal Design?
We've been making legal knowledge work for its readers since 2018 — and building and testing AI that reads legal terms since before most firms had a Copilot licence. The FT has recognised our AI work every year from 2022 to 2026, including an Innovative Lawyers award win in 2024 and recognition in 2026 for our AI-powered predictive drafting tool, built on our global user-testing data. Our founder Sara Rayment sits on the ICC's AI taskforce and served as a 2025 delegate to UNCITRAL Working Group I on data and AI in decision-making.
We're independent of every AI vendor. Our success metric isn't your licence count — it's whether your people, using your AI, on your documents, get it right. Measurably.
FAQ
What does "AI-ready" mean for documents? An AI-ready document is structured and written so AI systems can retrieve the right content and represent it accurately: clear sections that stand alone, plain language, explicit conditions, current versions and consistent terminology. The same qualities that make it human-ready.
Why does Copilot give wrong answers from our policies? Usually because of the knowledge, not the tool. AI answers inherit the ambiguity, buried conditions and outdated versions in the source documents — then deliver them fluently. Restructuring and rewriting the source fixes most of the answers; the review shows you exactly where.
How long does an AI readiness review take? A typical review runs three to four weeks: document sampling and scoring, AI answer testing, an optional live user test, and delivery of the scorecard and fix roadmap. Larger estates are phased.
Which AI tools do you work with? We're model- and vendor-agnostic. The review tests whatever your organisation runs — Microsoft Copilot, custom RAG assistants, or legal AI platforms — because the information layer beneath them is where the fixes live.