Finding ambiguity in content

In many organizations, the knowledge base serves as the trusted knowledge repository, ensuring content accuracy. However, when organizations go through changes – such as a new product release, mergers or acquisitions, a consolidation of business systems, or evolving corporate policies – it becomes difficult to maintain a single source of truth. Moreover, if technical writers are spread across various teams, they might have a siloed view of information. Updating only a few articles with current information while leaving others unchanged can result in ambiguities across the entire knowledge base. With the rise of AI search engines and chatbots, ambiguous content can lead to hallucinations, thus eroding trust in new technologies and even the content itself. So, how can we identify and eliminate content ambiguity in knowledge repositories?

Why ambiguity happens

Even when technical writers strive to ensure accurate content, ambiguity can arise due to several factors:

• Technical writers often work independently across different departments. This leads to contradictory information when stakeholders provide conflicting guidance without coordination.
• When a feature’s documentation is updated, related articles are left with the old content, creating discrepancies.
• Organizations struggle to identify and update all instances where information appears, leaving outdated content scattered throughout their knowledge base.

If the organization’s information practices are well developed, content ambiguity can be detected early and fixed during the editorial review process.

Types of content ambiguity

There are five important types of content ambiguity:

Factual inconsistencies are direct conflicts in which different content types state contradictory facts. For example, a company’s website states that "We have 500+ employees worldwide", while a press release refers to "Our team of 400 dedicated professionals".

Temporal ambiguity refers to time-related inconsistencies or outdated information that conflicts with newer, updated content. For example, a policy written in 2019 states that "Remote work is not permitted", yet, in the FAQ section, updated in 2020, there is a passage stating that "We offer flexible remote work options".

Scope ambiguity results from unclear boundaries of what is covered by statements, policies, or descriptions. For example, a warranty states that "All manufacturing defects are covered." But it doesn't specify if this includes wear and tear, cosmetic issues, or damage from normal use.

Inferential conflicts occur when multiple logical conclusions can be drawn from the same information, creating competing interpretations or contradictory inferences. A restaurant menu, for example, describes a dish as "heart-healthy" under the "Light Options", but the nutritional information shows 1,200 mg of sodium and 1,000 calories.

Terminological inconsistency occurs when the same concept is referred to by different terms, or different concepts use the same terminology. For example, a software company uses "client," "customer," and "user" interchangeably throughout their documentation. This might create confusion about whether these refer to the same group or different categories.

Finding ambiguity

There are many ways to detect and fix content ambiguity issues. This includes a combination of people, processes, and tools.

Tech writer training

Providing regular training to technical writers on best practices and content frameworks will ensure compliance with organizational processes. Since content ambiguity happens because of information silos, it is important to train technical writers in collaboration and coordination.

Process checks

Editors and subject matter experts have a holistic perspective of the entire knowledge base and know about possible dependencies. Adhering to style guides while drafting content and having a stringent editorial review process can eliminate ambiguities during the content creation stage.

Tools using GenAI

Given the advent of GenAI, it becomes easier to send large amounts of content to Large Language Models (LLMs). The right prompt can detect ambiguities in content within seconds. We will look at this in more detail below.

Using GenAI to detect ambiguity

GenAI apps powered by LLMs can be used to detect content ambiguity. Instead of providing the entire knowledge base as input, writers can identify related articles using semantic matching, and the top five articles can be sent to GenAI apps. Here is a sample prompt using Claude.ai that is used to detect five content ambiguities.

Task Description

Analyze the following set of articles from my knowledge base to identify any potential ambiguities, contradictions, or inconsistencies between them. For each type of ambiguity found, provide specific examples with direct quotes from the articles, explaining why they represent an ambiguity or contradiction.

Articles for Analysis

[Insert the 5 articles here, clearly labeled as ARTICLE 1, ARTICLE 2, etc.]

Ambiguity Types to Identify

For each of the following ambiguity types, identify if they exist between any of the articles:

1. Factual Contradictions: Direct conflicts in which contradictory facts are stated in articles.
2. Terminological Inconsistency: Identical concepts referred to by different terms, or different concepts using the same term.
3. Temporal Ambiguity: Time-related inconsistencies or outdated information that conflicts with newer content.
4. Scope Ambiguity: Unclear boundaries of what is covered by statements, policies, or descriptions.
5. Inferential Conflicts: When combining information from multiple articles leads to contradictory conclusions.

Output Format

1. First, provide an "Executive Summary" with a high-level overview of the most critical ambiguities found.
2. Then, for each type of ambiguity found, create a section with:
   • The type of ambiguity
   • The specific articles involved (by number)
   • Direct quotes showing the contradiction or ambiguity
   • Explanation of why this creates confusion
   • Potential impact on knowledge base users
   • Recommendation for resolution
3. For each article pair comparison, indicate the confidence level of the identified ambiguity (high, medium, low).
4. If a particular ambiguity type is not found, briefly note that it was not detected.
5. Finally, provide a prioritized list of the ambiguities that should be addressed first, based on:
   • Severity of potential misunderstanding
   • Centrality of the topic to the knowledge base
   • Likelihood of the user encountering the contradiction

Additional Instructions

• Focus on substantive contradictions rather than stylistic differences.
• Consider both explicit statements and implicit assumptions.
• Make note of instances when ambiguities might be intentional or reflect genuine uncertainty in the field.
• When analyzing terminological inconsistencies, distinguish between true ambiguity and simple synonyms that are clearly referring to the same concept.
• For each ambiguity identified, suggest whether resolution would require: a) updating one or more articles, b) adding clarifying context, c) creating a new bridging article, and/or d) consolidating information.

Closing remarks

Content ambiguity represents a critical challenge that extends beyond technical writing. It directly impacts user trust and organizational credibility, especially as AI systems can amplify these inconsistencies through hallucinations and conflicting responses. The solution lies in a systematic approach that combines comprehensive writer training, robust editorial processes, and sophisticated GenAI tools that can detect ambiguity at scale. Organizations must recognize that addressing content ambiguity requires enterprise-wide commitment, with technical writers serving as champions of information integrity. By embracing proactive ambiguity detection and resolution, knowledge repositories can serve as trusted single sources of truth, preserving both user confidence and organizational reputation.