Is your knowledge AI ready? Start before it is written down

AI drafts from whatever it is given, and it cannot tell which file is out of date. It can also fill gaps with text that sounds right. The NIST profile on generative AI calls that risk confabulation: confidently stated content that is erroneous or false. Give it two conflicting checklists, and a draft can blend them into a third procedure that nobody uses. So readiness is not about having everything written down. It comes down to three things: a current source for the task, a person who owns it and an approval step before anyone learns from the draft.

Take a quick inventory for one role before you touch any software. Look for every place where someone could learn how the job is done today. Write each one down in a simple list, with the task it covers and the person who last changed it.

• Written files: SOPs, checklists, binders, onboarding packets and old slide decks.
• Recordings: training videos, screen recordings, voice notes and recorded calls about how to do a task.
• Group chats where a manager explained a task once, and people still scroll back to find it.
• Vendor manuals and guides for the equipment or software you use.
• People: the one or two staff members everyone asks, and what they are asked most.

Run each source through the same short check. A source does not need to be perfect. It needs to be current, owned and safe to use, and the table shows what to fix when it is not.

Then the expert is the source, and a recording is the fastest way to capture it. Ask the person who does the task most reliably to talk it through while doing it, on a phone. Ask them to say why each step matters, not only what it is. A photo of their checklist and a short list of common mistakes help too.

That raw material becomes a draft source, which the owner reads and approves. In Elvina, Company Knowledge stores recordings, files and notes as sources with an owner, available now. Creation Studio drafts lessons and job aids from approved sources only, and a person approves each draft before staff see it; both are available now. Today the capture and drafting are done by our team with you on the platform.

• Record where the work happens, so the steps and the tools are both visible.
• Keep each recording to one task, so the owner can review it in one sitting.
• Note who recorded it and when, so everyone knows how current it is.

Every source needs one owner who can settle disagreements. Two approvers with different habits turn a lesson into a compromise nobody follows. The owner does not need to write well. They need to know the work and have the authority to say how it is done now. Give them a short review slot each week, and make it clear that anything they have not approved does not reach staff. If two locations do the task differently, the owner decides which way becomes the standard, or approves both as named variants. Write that decision into the source, so the next draft does not reopen it.

A source is stale when the work changed and the file did not. Watch for changes in equipment, software screens, prices, suppliers and customer rules, because each one breaks a procedure quietly. People change too. The Bureau of Labor Statistics reports that in January 2026, workers had been with their current employer for a median of 4.1 years, and 2.4 years in leisure and hospitality. When the person who knows a task leaves, the knowledge leaves too, unless someone captured it. Set a review date for each source, and review it sooner when someone who holds knowledge gives notice.

Some material should stay out of any AI tool, however useful it looks. Keep it out from the start, because removing it later is harder than never adding it.

• Customer records, account numbers and anything with a customer's name attached.
• Employee files, such as pay, reviews, medical notes or discipline records.
• Passwords, door codes and system logins, even inside an old checklist.
• Drafts nobody approved, such as a manager's half-finished notes from last year.
• Rules you are not sure are still true. Mark them for the owner instead of guessing.

You do not need a project plan to start. These short steps get one role ready, and each one fits around normal work.

• Pick one role and the task new hires get wrong most often.
• Gather every source for that task into one folder, including the group chats.
• Name one owner, and have them mark what is out of date.
• Record the expert doing the task once, from start to finish.
• Have the owner approve one draft lesson before you touch the next task.

A fictional thrift store chain wants to train donation sorters. The knowledge sits with one longtime sorter who decides what goes to the floor, what gets bundled for resale and what is recycled. Nothing is written down.

• The owner lists the sources: a faded price chart, a chat thread about electronics and the longtime sorter.
• The sorter records herself working through one donation bin, explaining each call as she makes it.
• The store manager becomes owner of the sorting source, and approves one named variant for a store with a smaller back room.
• Our team drafts a lesson and a one-page sorting guide from the recording and the chart, using drafting on the platform that is available now.
• The manager corrects the electronics rule, approves both and sets a review date for the next season.

Measure how ready the knowledge is and what the training built from it changes. Keep the baseline from before the cleanup.

• Tasks in one role with a current, owned source, counted before and after.
• Corrections the owner makes per draft, which shows how clean the source was.
• Questions new hires bring to the expert about tasks that now have a lesson.
• Sources past their review date, checked once a month.
• Mistakes a manager sees in the first weeks on the covered tasks.

Is our knowledge AI ready? It is ready for one task if there is a current source, one person who owns it and an approval step before staff see anything drafted from it. You do not need everything written down. Start with the task new hires get wrong most, and fix only that source first.

• Do we need to write SOPs before using AI? No. A recording of an expert doing the task, a checklist photo and a list of common mistakes are enough to draft from. The owner then reads the draft, corrects it and approves it. A full written SOP can come later, built from the approved lesson.
• Can we use vendor manuals and training videos as sources? Often, but check the terms first. What you wrote yourself is yours to use. Vendor material may limit copying or adapting. Also remove personal data, such as customer names in screenshots and recordings, before any file becomes a source.

Public sources, checked on the date shown.

• NIST AI 600-1 lists confabulation, the production of confidently stated but erroneous or false content, among the risks of generative AI. Source: https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf (checked 2026-10-02).
• BLS reports median tenure with the current employer of 4.1 years in January 2026, and 2.4 years in leisure and hospitality. Source: https://www.bls.gov/news.release/tenure.nr0.htm (checked 2026-10-02).