Frameworks & Templates4 min read

AI Skills for Employees: Audit the Input Pack First

Teach employees to test source relevance, authority, completeness, freshness and permission before an AI tool turns weak inputs into fluent output.

Bokili EditorialPublished October 8, 2026· Verified October 8, 2026

Source cards passing through authority, completeness, freshness and sensitivity checks before entering an AI workflow

AI skills for employees start before the first prompt. A well-written instruction cannot rescue an input pack built from the wrong policy, an incomplete spreadsheet, an outdated product note or material the employee was not authorised to use. Training should therefore teach people to audit the sources they bring to an AI tool, not only the words they type into it.

The practical aim is small: pause for two minutes and decide whether the source pack is fit for this task. That check prevents employees from spending ten minutes refining an answer that was grounded in weak inputs from the start.

Prompt quality begins with source quality

If the input pack is incomplete, stale or unauthorised, a fluent output can make the problem harder to see.

Why input judgement belongs in AI skills for employees

Training often starts with prompt structure and ends with output review. The missing step is input readiness. The UK Government Data Quality Framework treats quality as fitness for purpose and recommends addressing issues as early in the data lifecycle as possible. It separates dimensions such as completeness, consistency, timeliness, validity and accuracy because a source can pass one check and fail another.

NIST’s AI Risk Management Framework Playbook similarly asks organisations to document an AI system’s purpose, context and requirements, and to examine whether data is adequate, relevant and proportionate to that purpose. The UK Government AI Playbook adds that AI tools lack guaranteed accuracy and contextual awareness, so users need testing, human control and organisational data-handling rules.

Use the READY input-pack check

READY before you prompt

Relevant

Every source contributes to the exact task, audience and decision. Remove material that only adds noise.

Endorsed

Identify the authoritative version and owner. A convenient file is not automatically the source of truth.

Accurate enough

Record known errors, missing fields and conflicts instead of silently passing them to the model.

Dated

Check when each source was created or updated and whether that timing fits the intended use.

Yours to use

Confirm that the tool, data classification and intended processing are approved for the material.

READY is not a promise of perfect data. It is a short decision about whether the pack is fit for this use, what limitations must travel with it and whether the task should proceed at all.

Reading is a start. Practice makes it stick.

Start learning

Worked example: summarising a policy change

An employee is asked to draft a manager briefing on a revised travel policy. Their folder contains the signed policy, a presentation from last year, an undated FAQ and a spreadsheet of expense examples. Without an input audit, the AI may blend old limits with new wording and present the result with equal confidence.

The READY check removes the old presentation, confirms the signed policy as authoritative, flags that the FAQ has no date, and keeps the spreadsheet only as an example rather than a rule source. The prompt can then state the hierarchy explicitly: use the signed policy for requirements, use the examples only to illustrate, and mark any question the policy does not resolve.

Build a clean input pack

1
State the task
Write the intended output, audience and decision in one sentence.
2
List the sources
Name each file, page or dataset and its owner or publisher.
3
Run READY
Record one line for relevance, authority, known issues, date and permission.
4
Set the hierarchy
Tell the model which source governs when two materials conflict.
5
Carry the limits forward
Require the output to flag gaps, stale material and unsupported conclusions.

Do not hide weak inputs inside a stronger prompt

A longer prompt can describe format, tone and method. It cannot make a missing record appear, turn an obsolete instruction into current policy or grant permission to process sensitive content. When the pack fails READY, the right action may be to find the current source, ask the owner, redact material, switch to an approved environment or stop.

Input-pack release check

The task and intended use are explicit.
The authoritative source is named.
Missing, conflicting or low-quality material is visible.
Dates and version information are recorded.
Sensitive content is handled in an approved environment.
The prompt states what the model must not infer.

Related Bokili guides cover the next controls: define the source of truth, run an omission check, test in a clean context and check whether redacted data can still identify someone. The input audit is the bridge between choosing the material and checking the result.

Ten-minute input-pack audit

Choose one low-risk AI task you perform at work.
List every source you would give the tool.
Apply the five READY checks to each source.
Remove one irrelevant or outdated item.
Write the source hierarchy into the prompt.
Add one sentence telling the model how to report missing evidence.
Keep the result only after checking it against the authoritative source.

Better AI use does not begin with more elaborate prompting. It begins with deliberate inputs. Bokili gives employees short, role-relevant practice so source judgement becomes part of everyday AI work.