How to Prepare Documents for the Teamflect AI Knowledge Base

The quality and structure of the documents you add to the Teamflect AI knowledge base can affect how easily the AI Agent can find and use the information they contain.

A document that is easy for a person to read is not always equally easy for an AI system to process. For example, important information may be presented only in a screenshot, complex table, diagram, or scanned page rather than as searchable text.

This guide covers our recommended practices for preparing internal policies, HR documentation, performance management resources, employee guides, and other company materials for use with the Teamflect AI Agent.

For the most reliable results, we recommend using:

For PDFs, make sure the text can be selected, copied, and searched. If a PDF was created from a Word document, retain the original text layer rather than flattening the pages into images.

Teamflect's Company Resources uploader accepts several additional file types, including presentations, spreadsheets, images, audio, video, and archive files. owever, being accepted by the uploader does not necessarily mean that every file type will provide the same AI retrieval quality.

For knowledge that employees will rely on through the AI Agent, DOCX and searchable PDFs are therefore our recommended formats.

The most important information in your documents should be available as regular, machine-readable text.

This is particularly important for scanned PDFs. A scanned document may consist entirely of images of pages rather than actual text.

Before uploading a scanned document, we recommend checking whether you can:

If you cannot, consider converting the document into a searchable format before adding it to your AI knowledge base.

Do not rely exclusively on an image, screenshot, chart, or diagram to communicate information that the AI Agent needs to reference.

For example, instead of providing only a diagram showing your performance review process, add a written explanation alongside it:

"Contoso's performance review process consists of four stages: employee self-review, manager review, performance discussion, and final sign-off."

Similarly, if a screenshot demonstrates an important process, explain the relevant steps in the surrounding text.

You can still use visuals to make documents easier for employees to understand. The important point is that business-critical information should also be stated clearly in text.

Tables can be useful for presenting structured information, but complicated layouts can make information harder to interpret during document processing.

Where possible:

For business-critical information, we also recommend adding a short written explanation before or after the table.

For example, instead of relying exclusively on:

Add a sentence such as:

"Employees with 0–2 years of service receive 15 days of annual leave, while employees with 3–5 years of service receive 20 days."

This makes the rule explicit even outside the table.

Avoid treating your AI knowledge base as a general document repository. Instead, curate the documents around information that employees and managers are likely to need.

For example, an enterprise knowledge base might include separate documents such as:

Separating large collections of information into logical, authoritative documents can make them easier to maintain and update.

However, avoid splitting closely related information unnecessarily. A policy and the exceptions to that policy should remain together so that the complete rule can be understood in context.

Structure your documents using real heading styles such as Heading 1, Heading 2, and Heading 3.

Headings should describe the information that follows.

For example, prefer "Parental leave eligibility" over "Additional information." Similarly, prefer "Performance rating definitions" over "Ratings."

Clear headings make documents easier to navigate and help keep related information organized.

Each section should ideally focus on one subject.

Avoid very long sections that combine several unrelated policies or processes. For example, instead of one section called "HR Information" containing leave, benefits, onboarding, performance reviews, and payroll information, use separate sections such as:

Where possible, keep the heading and the information answering that topic close together.

Do not assume that the meaning of a statement will always be obvious from the rest of the document.

Important qualifiers should be written directly into the relevant section. This is especially important for:

For example, instead of writing "Employees receive 20 days." write "Full-time employees in the UK with three or more years of service receive 20 days of annual leave per calendar year."

The second version contains enough context to make sense independently.

Define acronyms when they first appear and use terminology consistently throughout your documentation.

For example: "Individual Development Plan (IDP).'' After defining it, you can use "IDP" throughout the rest of the document.

Avoid using several different terms for the same process unless those differences are intentional.

This becomes particularly important in global organizations where different teams or countries may use different terminology for the same HR process.

For policies and other frequently updated materials, include information such as:

For example:

Title: Global Performance Management Framework

Owner: Global HR

Version: 2.1

Effective date: January 1, 2027

Audience: All employees and people managers

Include important version information within the document itself rather than relying only on the filename.

The AI Agent relies on the information made available through its knowledge sources. If multiple documents contain different versions of the same policy, this can create ambiguity.

For example, avoid simultaneously using:

"Performance Rating Framework 2026"

and:

"Performance Rating Framework 2027" as active knowledge sources if the 2026 version is no longer applicable.

Review your knowledge base regularly and remove or disable:

Keep the current authoritative version available whenever possible.

Avoid placing important information exclusively in:

If employees need the AI Agent to know something, state it clearly in the main body of the document.

For example, if a policy contains a link to an external benefits website, do not assume that the AI Agent will automatically retrieve and understand all of the information available through that link.

Include the information employees need directly within the uploaded knowledge source where appropriate.

You do not necessarily need to upload one very large handbook containing every company policy.

For large collections of information, consider separating stable topics into clearly named documents.

For example, instead of "Global Employee Handbook.pdf" you might maintain:

However, each document should remain authoritative and understandable on its own. Avoid separating a rule from its exceptions or placing essential context in another document unless necessary.

For larger organizations, some practical knowledge-base use cases include:

Upload global and country-specific HR policies, employee handbooks, leave processes, benefits information, and common HR procedures. Employees can then use the AI Agent to find information without manually searching through multiple documents.

Provide performance review guidelines, rating definitions, competency frameworks, review timelines, calibration guidance, and manager instructions. For example, managers could ask:

Provide manager handbooks, leadership frameworks, coaching guidance, and internal people-management processes that managers need throughout the year.

Provide career frameworks, competency definitions, development resources, and internal learning guidance.

Provide new-hire documentation, onboarding processes, organizational information, and frequently asked questions.

Add your organization's own guidance explaining how employees and managers are expected to use Teamflect.

For example:

Before making a new collection of documents available to a large employee population, test it using realistic questions.

Choose questions employees are actually likely to ask rather than simply checking whether the AI Agent can repeat a sentence from the document.

For example, if you upload a parental leave policy, test: "What is our parental leave policy?"

Then test more specific scenarios:

Similarly, test information contained in tables, regional variations, eligibility conditions, and exceptions.

This helps identify areas where the source document itself may need to be clearer.

For larger organizations, you may want to begin with a smaller, curated knowledge base and a limited employee population.

For example:

This approach allows your organization to validate both the source material and the employee experience before a wider AI rollout.

Before uploading a document to the Teamflect AI knowledge base, check the following:

Following these practices helps create a knowledge base that is easier to maintain and provides the Teamflect AI Agent with clear, authoritative company information to reference.

Teamflect supports uploading a broad range of file types to Company Resources. However, a file being accepted for upload does not necessarily mean that every file type will provide identical AI retrieval quality.

For this reason, DOCX files and searchable, text-based PDFs are currently the safest formats for AI knowledge sources.

Scanned documents, image-only information, complex tables, and highly visual layouts should be treated with additional care. Whenever information is important, provide it as clear body text rather than relying solely on visual presentation.

Exact processing behavior can vary depending on the content and structure of the source document, so we recommend testing representative documents and realistic questions before a broad organizational rollout.