Practical guide
AI fluency for information specialists: document a search
Build a reproducible fictional evidence search with explicit inclusion rules, version handling and access limits.
Information specialists can use AI to propose search terms while retaining control over retrieval and inclusion decisions. This original exercise builds a small search log for guidance on evaluating workplace AI outputs. It focuses on a reproducible search process, not a claim that a search is exhaustive or that every retrieved item is reliable.
Turn the request into a retrieval question
A search needs boundaries that explain what belongs in its result set.
The fictional requester wants practical guidance for knowledge workers checking AI-generated text, not model-training techniques or general news about AI. Define the intended audience, document type and use of the result. The deliverable is a short source shortlist with reasons for inclusion, not a broad bibliography.
Ask AI to suggest alternative phrases for checking, verification and generated text. Review whether each suggestion keeps the original scope. A term such as model validation may retrieve useful background but also change the question. Record that choice rather than silently expanding the search because more results appear.
Keep a search log another person can follow
The log should record actual searches, not a plausible reconstruction after the fact.
Use fields for platform, date, query, filters, retrieved result identifiers and screening decision. If the exercise is only a search-plan walkthrough, leave executed-result fields blank. Do not populate them with invented hit counts or URLs to make the plan look completed.
AI may organize a query list, but it should not claim to have searched a database it cannot access. Record access restrictions and the implications for coverage. A search on one public website does not establish what a subscription database or an internal collection would contain.
Screen against the question rather than keyword presence
An item can match the words while failing the intended use.
The fictional screening set contains a practical checking guide, a news announcement and a technical paper on training loss. The practical guide appears directly relevant; the announcement may provide context; the technical paper does not automatically address the workplace checking task. State those dispositions with reasons.
Do not discard an item solely because its title lacks the preferred phrase. Inspect the available abstract or text and mark what was actually read. If only a snippet is accessible, record provisional relevance rather than implying full-document review. Keep excluded items with reasons when that helps reproduce the search.
Handle versions without multiplying evidence
Several copies of one document should not become several independent sources.
Introduce a corrected version of the practical guide and an older mirror copy. Record the relationship and identify which version should be used for the current note. If the correction affects a recommendation, revisit the relevant summary rather than updating only the access date.
Ask AI to flag matching titles and identifiers, then confirm the relationship in the supplied metadata. Similar titles can also belong to different works. Do not merge them automatically. Version handling should reduce false duplication without erasing genuinely distinct documents or their publication histories.
Report the search’s limits with the shortlist
A usable result explains both what was found and what the method could miss.
Deliver the retrieval question, query plan or executed log, inclusion rules, screened items and version decisions. Describe platforms not searched and text not available. Avoid calling the result comprehensive unless the scope and method justify that statement, and do not equate a large result count with adequate coverage.
To review the exercise, add an item that is relevant despite an unexpected title and ask whether the rules can include it consistently. This guide concerns retrieval and screening. The knowledge manager resource starts later, with the ownership and maintenance of guidance already in an internal collection.
Sources and scope
NIST: AI risk management. Skills England: workplace AI foundations.
These references provide background, not validation or endorsement of this exercise. The case details, calculations and suggested review questions are original instructional material. Use them to discuss observable work, not to infer customer outcomes, professional credentials or performance in every setting. Before adapting the exercise, confirm the relevant facts, approved tools, data permissions and decision owners. If you change the case, revisit the expected answers and checks as well. These examples describe practice tasks, not a promise that a particular product includes the fictional features.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England