Skip to content
Resources / GA Applications

Guide / AI assistance

Give AI a narrow, source-grounded job and keep important decisions with people.

AI assistants can help staff find approved information, summarise source material, prepare drafts and extract structured fields. They can also be wrong. A useful implementation defines sources, permissions, task boundaries, uncertainty, evaluation and human review before connecting the assistant to customer communication or operational actions.

Questionwhat the reader needs
Evidencedefinitions and boundaries
Next stepa useful action path
Why it matters

Find

Retrieve approved material with source links.

CHAPTER 01

Choose bounded assistance tasks.

Different tasks carry different risk.

01.1

Find

Retrieve approved material with source links.

01.2

Draft

Prepare content clearly labelled for review.

CHAPTER 02

Prepare sources and permissions.

Model quality cannot repair unmanaged knowledge.

02.1

Sources

Name owners and review dates.

02.2

Access

Retrieve only what the user may see.

CHAPTER 03

Evaluate failures before release.

Test unsupported and conflicting questions.

03.1

Uncertainty

Allow the assistant to say it cannot answer.

03.2

Review

Name the person responsible for important output.

Continue through the resources system.

Each route answers a different operating question while retaining the same standards for evidence, responsibility and a usable next step.

Questions worth resolving

What to clarify before committing.

Will an assistant always be correct?
No. Source grounding and human review reduce but do not remove errors.
Can it use private documents?
Potentially, under an architecture covering access, provider terms, logging and retention.

Bring the real problem

Choose one assisted task and its approved sources.

Bring representative questions, documents and the decision that follows.