When a maintained source index is a better first step than a chatbot

A chatbot can retrieve relevant text without knowing whether the organisation has authorised that text as a public answer. A working discussion and a final report may use the same terms while representing different stages of a decision. Relevance alone does not resolve that distinction.

This fictional planning decision concerns a small set of public questions. It chooses a maintained route to approved passages before adding conversational generation. It does not claim that a chatbot cannot be designed with a properly governed public corpus.

Inspect the proposed source set

An organisation wants a chatbot to answer questions about a completed programme. The proposed collection includes its approved report, partner meeting notes and internal drafts. The team assumes that including more material will make answers more useful.

One actual reader question is whether the programme continues into the following year. The approved report states that the pilot has ended and directs readers to the organisation for any later announcement. A partner meeting note discusses a possible continuation, but records no approved decision.

The two passages are easy to retrieve together because they share the programme name and the word “continuation.” The working note is not inaccurate as a record of discussion. It is unsuitable as authority for a public claim that the programme will continue.

Identify the missing product decision

The issue is not a low recognition score or a missing keyword. The team has not defined which material may serve as public evidence, who maintains that selection or how an unresolved question should be handled.

A confidence threshold would not supply those decisions. A system could be highly confident that the meeting note contains relevant words while still presenting a proposal as an organisational commitment. The content owner must establish the source boundary.

The owner confirms that the approved report and two public notices are the sources for the current question set. Partner deliberations remain outside the public collection. That boundary applies whether the eventual interface is a search page, a source index or a chatbot.

Compare the immediate options

The first option is a broader conversational tool with an approved corpus, retrieval checks, answer evaluation, maintenance responsibilities and an escalation route. It may be worthwhile, but the current brief has not supplied the necessary operating arrangement or evidence of a wider question demand.

The second option is a small, maintained question-to-source index. It answers the observed tasks by naming the question, linking to the approved passage and making the issuing source clear. It does not generate a new institutional statement from working notes.

The organisation chooses the index for the first delivery. The choice rests on the narrow current task and the approved material already available, not on a general rejection of AI tools.

Publish a bounded answer route

For the continuation question, the index uses the heading “Is another programme period announced?” It states that the cited final report describes the pilot as ended and links to the relevant passage. It directs questions about future plans to the organisation's designated enquiry route.

A second entry addresses where to find the final outputs, linking to the approved publication record. A third question about an unannounced future funding arrangement has no approved public answer. It receives the enquiry route rather than a speculative synthesis from partner notes.

The selected result is useful even without a conversational interface: readers can find the current approved position and understand where a question remains open.

Check authority and maintenance together

In the fictional review, the continuation task leads to the approved report passage. The outputs task leads to the intended publication record. The unresolved funding question does not expose or paraphrase internal discussion.

The editor can identify the source for each entry and the person responsible for reviewing it when a new public announcement is issued. The index does not automatically change because someone adds a meeting note to the internal document store.

The owner also verifies that the report's description has not been expanded into a claim that continuation is impossible. “No approved announcement in this source set” and “there will never be another programme” are different statements.

Keep a concrete trigger for revisiting the tool

The organisation can reconsider a chatbot if it develops a maintained approved corpus, a broader set of reader questions, evaluation examples and a clear route for uncertain answers. That future scope should include how source changes are reviewed and how generated claims are checked against them.

For now, the index resolves the observed task and leaves the unanswered question with its proper owner. Read public and private material when preparing the source boundary, or describe the reader questions before selecting the interface.