Dr. Connor Robertson Books

Published October 27, 2026 ยท Independent educational guide

Build an AI-Assisted Reading Glossary Without Inventing Definitions

A reading glossary is useful when it helps you understand a passage without replacing that passage with a more confident definition. Familiar business words can carry different meanings in different contexts. An AI assistant can help organize candidate terms and draft own-word explanations, but the resulting list still needs to preserve where each meaning came from, what remains unresolved, and which distinctions matter to your reading question.

This guide describes an original educational workflow for supplied reading material. Its six practice cases are deliberately fictional written notes. They are not quotations from a book, chapter summaries, manuscript excerpts, actual model outputs, or client records. No claim is made that a named author used these definitions or that a model produced the proposed interpretations. The aim is a glossary you can check against the material you actually have permission to use.

Decide what the glossary should help you do

Begin with a reading question, such as understanding how a passage distinguishes a process measure from an outcome. A glossary for that purpose will differ from a dictionary of every unfamiliar word. Include terms that affect the question, the comparison, or the reasoning you are trying to follow. This keeps the exercise connected to reading rather than turning it into a long list that looks complete but does not clarify the argument.

Separate three possible tasks. One task records how a term is used in a supplied passage. Another supplies outside background about a concept. A third proposes a definition your own team might adopt. These can all be useful, but they need different evidence labels. A background definition is not automatically the source author's definition, and a proposed local convention is not automatically a description of current practice. Keep those boundaries visible in the glossary entry.

Set a small initial scope that you can review carefully. You might select a few supplied paragraphs, record their locators, and identify the terms needed to understand them. A narrow scope does not establish that the glossary covers a whole book or a field. It gives you a manageable starting point. Expand after checking whether the entries preserve the meaning of the selected material, not merely after finding more words to include.

Keep term labels separate from meanings

The W3C SKOS Primer provides a useful limited distinction: concepts can have labels and documentary notes. A label is not the whole account of a concept. This guide uses that distinction as vocabulary context, not as a claim that a reading notebook implements SKOS or conforms to a technical standard. You can keep the same practical separation in an ordinary document by recording the term, its local meaning, and the note that supports the interpretation.

Give each entry a working identifier so that two uses of the same word can remain separate while you investigate them. For example, Margin-A and Margin-B can refer to two candidate meanings rather than two confirmed synonyms. The identifier organizes your notes; it does not prove that the candidates are distinct concepts in the source. Preserve the occurrence locators so you can return to the passages before making that decision.

An entry should identify the exact term form, the source locator, an own-word explanation of the local use, and any qualification necessary to understand it. It can also include a related term or an unanswered question. Do not force every entry into a polished one-sentence definition if the supplied passage does not define the term. A precise note that says definition not supplied is more useful than an invented definition presented as the author's meaning.

Supply the assistant with evidence and a bounded task

Give an assistant only the material appropriate for your task and access permissions, together with your reading question and a clear description of the output you want. Ask it to identify candidate terms in the supplied passages, preserve occurrence locators, draft explanations in its own words, and distinguish explicit definitions from inferred meanings. Tell it to mark unresolved entries rather than fill gaps from memory. These are suggested instructions, not a tested prompt or a promise that the assistant will follow them correctly.

Ask for the evidence boundary in each entry. An explicit definition should point to the supplied passage that states it. A local-use inference should name the phrase or comparison that prompted the inference. Outside background, if requested, should be separately sourced. This lets you review the relationship between the explanation and its support. A fluent glossary can otherwise turn a reasonable guess into an apparent quotation-free but still unsupported attribution.

NIST's AI Risk Management Framework is intended for voluntary use across AI design, development, use, and evaluation. That broad context supports taking evidence and evaluation seriously; it does not validate this particular reading workflow or certify a glossary. The practical checks in this guide remain narrow checks of supplied notes and original teaching examples. They do not establish model accuracy, professional review, or the correctness of an unseen book interpretation.

Case A: the same word can describe different quantities

Case A contains two fictional notes. Note P1 uses margin for a quantity of twenty invented units. Note P2 uses margin for twenty percent. The number twenty appears in both, but the units differ. A proposed glossary entry that merges them into margin equals twenty would remove a distinction needed to interpret the notes. Retain two occurrence records and describe the quantity stated by each before deciding whether they refer to the same underlying concept.

The exercise does not supply a denominator for the percentage or a real financial statement. It therefore cannot establish a conversion between the two quantities. Do not invent a revenue amount to make twenty units equal twenty percent. If an assistant supplies a plausible denominator, mark it as an added assumption and return to the source question. The absence of a denominator is a gap to investigate, not an invitation to complete the example with an attractive number.

A useful provisional entry can say that the word appears with an amount in P1 and a percentage in P2, while the relationship between those uses remains unresolved. This is a reading result, not a financial recommendation or an accounting definition. An external dictionary might explain several common meanings, but it would still not establish which meaning the supplied notes intend. Keep background vocabulary separate from the local evidence.

Case B: distinguish elapsed time from active work

Case B supplies an invented elapsed cycle of sixty minutes and active work of fifteen minutes. The elapsed quantity is four times the active quantity, and their arithmetic difference is forty-five minutes. Those statements follow from the supplied numbers. They do not establish why the difference exists, whether anyone was idle, whether the process could be improved, or whether a real team performed poorly. No workflow was observed for this example.

The glossary should preserve two entries or two explicitly distinct measures: elapsed cycle and active work. Each needs its starting and ending basis if that basis is supplied. If it is not supplied, record the missing definition. A summary that calls both measures completion time makes later comparisons ambiguous. A reader cannot tell whether a proposed change affects the whole interval, the active portion, or an entirely different measure that was never defined.

Suppose a later fictional note reports fifteen minutes but omits the measure name. Do not choose active work solely because that value appeared earlier. A matching number is not a definition. Link the ambiguous occurrence to the two candidates and state the question needed to distinguish them. This keeps the glossary useful when a source switches shorthand or relies on context that is absent from the excerpt.

Case C: a role is different from an identified person

Case C defines an invented owner role as review coordinator, while leaving the assigned person unknown. The role description explains a kind of responsibility in the fictional note. It does not identify a person, prove that someone accepted the assignment, or establish authority to approve an action. A proposed glossary entry that replaces owner with a guessed person's name would add information the supplied material does not contain.

Record the role meaning and the assignment question separately. A role entry might explain the responsibility described by the passage, while an adjacent question asks who currently fills it in the relevant situation. Do not make a glossary carry operational facts that belong in a separate assignment record. A definition can remain stable while the person filling a role changes, and a named person can appear in a passage without being assigned that role.

This distinction also helps when comparing a book idea with a local workflow. An author may describe a responsibility in general terms, but that does not appoint anyone in your organization. If you propose a local assignment after reading, label it as your proposal and identify the acceptance or approval needed. The glossary clarifies language; it is not evidence that the proposed workflow has been adopted or that an individual has consented to perform a task.

Case D: preserve scale words from headings

Case D describes a fictional table whose heading states that displayed quantities are in thousands of invented units. A displayed value of seven therefore represents seven thousand units under that supplied heading. A separate note contains a literal seven units. The represented quantities differ by a factor of one thousand. The example is not actual money, a company result, or a recommendation about financial reporting.

A glossary entry for the measure needs to retain the scale information attached to the table. A detached row label and value can be misleading if the heading is omitted. Ask the assistant to keep the table title, relevant column heading, unit, and scale with each extracted measure. If the excerpt lacks those fields, the proposed explanation should identify the missing context rather than silently assume ordinary units.

Scale words can look like formatting when they are actually part of meaning. The same applies to a footnote that limits a measure to a subset of cases or a note that excludes a particular category. A glossary built from isolated words cannot recover context that was never supplied. Before using an entry to calculate or compare values, inspect the relevant headings and qualifications and retain the ones needed for the intended comparison.

Case E: an automation label does not authorize an action

Case E gives a fictional automation scope of draft only. A proposed explanation describes a process that sends the drafted message. Sending goes beyond the supplied meaning. The glossary should preserve the stated scope and identify the proposal's added action. No actual message was drafted for a recipient or sent in this example, and no operating authorization is implied by the word automation.

A short definition can accidentally change the boundary of a workflow. Terms such as assistant, automatic, review, and approved need to remain connected to the actions and decisions the passage actually describes. Do not infer that a system may act independently merely because a source says it helps prepare work. An own-word explanation should retain the distinction between producing a draft, presenting it for review, accepting it, and carrying out an external action.

When adapting an idea to your own work, keep a separate implementation question beside the reading entry. The entry describes the supplied term's local meaning; the implementation note asks what actions you propose and who may authorize them. This separation lets you explore useful applications without rewriting the source into a claim that it endorsed your specific process. It also prevents a glossary from becoming an unreviewed instruction set for an integrated tool.

Case F: count selected occurrences without claiming complete understanding

Case F contains four fictional term occurrences, G1 through G4. The selected material includes G1, G2, and G4, leaving G3 unread. Three of four supplied occurrences have been selected, a selection fraction of three quarters. That number does not measure semantic coverage, accuracy, or completeness of the glossary. The omitted occurrence might contain a definition, a qualification, or a different use; its significance is unknown until it is read.

The example also gives the word lead two candidate meanings without supplying enough context to choose between them. Keep the candidates distinct and retain the unresolved state. Do not merge them because their spelling matches, and do not select one because it is more familiar in business conversation. An assistant can propose a question that would distinguish the candidates, but the question itself is not evidence for either answer.

A useful result can state both the narrow count and the open meaning question. The note might record that three supplied occurrences were selected and that the interpretation of lead remains unresolved because the relevant context has not been consulted. That is a complete description of the exercise's current state even though it is not a complete glossary. Avoid turning an incomplete but organized output into a claim that the whole source has been understood.

Review the proposed entries against the passages

Read each drafted explanation beside the passage that supports it. Check whether the assistant preserved qualifications such as may, sometimes, under a stated condition, or for a particular group. Removing one of these can transform a local use into a universal definition. Also check whether the explanation has imported a familiar outside meaning that the passage does not establish. A term can sound ordinary while functioning as specialized shorthand in that source.

Inspect role words, unit words, and action words carefully. A role can become a person, a proportion can become an amount, and drafting can become sending during a seemingly small rewrite. These are meaning changes rather than stylistic preferences. Record the specific mismatch and revise the entry so that its scope matches the evidence. Do not accept a smoother sentence merely because it is easier to read if it answers a different question.

Use unresolved as a meaningful result. If two passages appear inconsistent, retain both locators and describe the difference before trying to reconcile them. The next step may be to read surrounding material, identify an edition difference, or consult a separately sourced background definition. Do not let the assistant invent a harmonized definition and then present that invention as though it resolved the author's terminology.

Build connections without flattening distinctions

A glossary becomes more helpful when related entries point to one another. You might connect an elapsed-time measure to an active-work measure while stating that they are not interchangeable. You might connect a role definition to a separate assignment question. These relationships support reading because they show which terms need to be considered together. They do not require collapsing all related terms into one broad definition.

Keep a small related-term note that explains the relationship in ordinary language. Is one term broader, a component, a possible alternative label, or an unresolved candidate? State only the relationship supported by the supplied material. The SKOS Primer discusses several kinds of vocabulary relationships, but this guide does not implement its formal model or infer a standard relationship from a casual reading note. The purpose here is a transparent educational notebook.

When a source changes terminology, preserve the earlier term form and the later term form with their locators. A later explanation may clarify an earlier shorthand, but it may also introduce a different scope. Check before replacing every earlier entry. This is especially useful when a reading group combines notes from separate sections or editions. A shared spelling convention can improve navigation without establishing that every source uses the term identically.

Maintain a glossary as a reading aid, not a substitute for the source

Give each revision a date and a short reason. For example, an entry may change from unresolved to a supported local-use explanation after surrounding paragraphs are read. Retain the earlier uncertainty in the correction note when it explains an important change. Another reader should be able to see why the definition changed without guessing whether the source changed, the excerpt expanded, or the original interpretation was mistaken.

Use the glossary while mapping an argument, comparing notes, or preparing a discussion question. Return to the source when the meaning of a term affects a conclusion. A glossary is a compressed interpretation, and compression can leave out details that matter to a new question. Do not treat the existence of an entry as proof that the term has one settled meaning across a book, an industry, or your own organization.

If a reading group shares entries, include the evidence labels and unresolved questions with them. A polished export that removes every caveat can turn a useful draft into misleading authority. Keep the source locator, local meaning, outside-background label, and implementation proposal distinct. This allows other readers to contribute corrections while preserving the boundary between what the passage supports and what the group would like to investigate next.

Frequently asked questions

Can AI define terms that the supplied passage does not explain?

It can propose background or candidate meanings, but those need a separate label and support. Do not present a plausible outside definition as the source author's definition. An entry that records definition not supplied can be more accurate than an invented explanation. Keep a question for the surrounding material or an independently sourced reference.

Should I merge every occurrence of the same word?

No. Matching spelling does not establish matching meaning. Case A uses margin with different stated quantities, while Case F leaves two meanings of lead unresolved. Retain occurrence locators and candidate meanings until the supplied context supports a relationship. Related terms can be linked without being treated as synonyms.

What belongs in a practical reading-glossary entry?

Keep the term form, source locator, own-word local explanation, evidence type, necessary qualifications, and unanswered questions. Include units or scale when they affect meaning. Mark outside background and proposed local definitions separately. The entry should help another reader reconstruct why you interpreted the term that way rather than merely supply a polished sentence.

Does an owner definition identify who must perform the work?

Not by itself. Case C defines a fictional responsibility while leaving the assigned person unknown. A role, a named person, an accepted assignment, and authority to approve an action are different facts. Keep implementation and assignment questions separate from the reading definition, and do not treat a source's general discussion as a local appointment.

Does selecting most occurrences mean the glossary is complete?

No. Case F's three selected occurrences out of four give a selection fraction of three quarters, not semantic coverage or verified understanding. The unread occurrence could change the interpretation. State the supplied scope and the unresolved question rather than turning a count of selected units into a claim about the meaning of an entire source.

Are these examples quotations or a method attributed to a book?

No. All six cases and the suggested workflow are original educational material. No manuscript, chapter quotation, actual model output, client result, professional review, or author endorsement is asserted. Narrow checks verify only the supplied fictional arithmetic and consistency. The guide helps readers examine material they have, without inventing what an unseen book says.

Original fictional practice data

Original fictional teaching cases, not actual client records, reservations, model outputs, professional review or operating approval. No executable utility is supplied.

Six original fictional teaching cases; definitions and scope remain in the guide
fictional caseauthored inputsbounded result
Afictional noteP1 margin=20units;noteP2 margin=20percentsame word/different stated quantities;not interchangeable
Bfictional elapsedcycle60min;activework15min60/15=4;45min difference;different measures not performance finding
Cfictional ownerrole=reviewcoordinator;personassignedunknownrole definition not identity or accepted assignment
Dfictional tableheadingthousands;display7 versusliteral7unitsrepresented7000versus7;factor1000;not actual money
Efictional automation scope=draftonly;proposal=sendsmessageproposal exceeds supplied meaning;no send performed
Ffictional selected occurrencesG1G2G4ofG1-G4;termleadhas2candidate meanings3/4occurrence selection NOTsemanticcoverage;G3unread;meaningunresolved

Primary sources and editorial scope

AI-assisted website editorial. Sources read October 27, 2026 America/New_York. Original examples checked independently. No record change, actual client dataset, observed model output, professional review, security certification or operating approval is asserted.

Related practical questions