Read a sample: 1. A Useful Assistant, Not an Authority
A volunteer at a small historical society is preparing for a new project about a neighborhood gathering place. A box of photographs sits on the table. An interview with a longtime resident is being arranged. The society’s catalog contains useful records, but visitors do not always know the words the catalog uses.
There is more work than the volunteer team can finish in an afternoon. Someone needs to suggest interview topics, describe photographs, and think about how people might search for them. An artificial intelligence tool may help with each task. It can offer a first list of questions, suggest descriptive words, or propose alternate search terms in seconds.
It cannot, however, tell the team what actually happened simply because it produces a fluent answer. It does not know who has given permission, which details a community considers sensitive, or whether a confident-sounding caption is supported by the photograph. Those judgments belong to people, working with the sources and with one another.
That distinction is the foundation of this book. AI can be a useful assistant for local history work, but it is not an authority on the past. Treat its responses as proposals to examine, not facts to accept.
The tools discussed in this book respond to instructions, often called prompts. You type a request in ordinary language, and the tool generates a response. For example, you might ask it to suggest open-ended topics for an interview about a community garden, or to propose search terms for a collection of meeting notices.
A prompt does not have to be technical. It does need to give the tool a clear job and enough appropriate context to make the response useful. You might write:
> Suggest five broad, open-ended interview topics for a project about a community garden. Do not assume that every participant had the same experience.
The response may include topics you had not considered. It may also include topics that do not fit the project. You remain responsible for deciding which suggestions to use.
AI-generated language can be especially persuasive because it is often smooth, specific, and neatly organized. That polish is not evidence. Depending on the tool and the information supplied, a response can blend sound suggestions with invented details, misunderstand a record, or repeat an assumption embedded in the prompt. A tool might confidently name a person in a photograph when the available description does not identify anyone. It might suggest that a building served a particular purpose because similar buildings often did. Neither statement becomes true because it appears in a tidy paragraph.
The safest working habit is to separate three things:
- The source: the photograph, recording, document, or other material being described.
- The source’s claim: what a label, narrator, or document says, which may itself require context or checking.
- The AI suggestion: a proposed question, tag, search term, or wording that a person must review.
These categories can overlap in a final record, but they should not be confused during the work. An AI response is not an original source for a local event.
This book focuses on three jobs where a volunteer can benefit from suggestions without handing over historical judgment.
Before an oral history interview, a team may know its broad purpose but need help thinking through possible subjects. AI can suggest themes and draft questions in accessible language. For a project about a neighborhood market, it might suggest asking about daily routines, changes over time, or the market’s role as a meeting place.
A useful question gives the narrator room to decide what matters. Compare “What do you remember about visiting the market?” with “Why was the market the heart of the neighborhood?” The second question presumes an answer. Even if that description appears in a project proposal, it may not reflect the narrator’s experience.
AI can help produce options, but an interviewer should choose and revise them. Project leaders and community members may notice missing perspectives or language that feels awkward, leading, or unwelcome. A question list is a guide, not a script that takes priority over the person being interviewed.
Interview planning also does not require submitting an existing interview to a tool. General project goals may be enough to generate a starting list. If the team later wants assistance with actual testimony, it must first consider the permissions and sensitivities attached to that material.
A tag is a word or short phrase used to describe an item so that people can find it. Depending on the collection, tags may describe subjects, activities, places, or other documented features. A photograph showing people planting seedlings might receive a subject tag such as “gardening,” provided that the image or its accompanying documentation supports the description.
AI can suggest candidate tags from a brief, permitted description. It can be helpful when volunteers are trying to think consistently across a group of items. But a plausible tag can still be wrong. “School garden” should not be added merely because children appear near plants; the image may show a family garden, a park, or something else.
Tags also carry choices about language. A term that seems ordinary to one volunteer may be outdated, overly broad, or inconsistent with how a community describes itself. Human review should consider not only whether a tag is supported, but whether it is respectful and useful.
Search terms are the words people might enter when looking…