Well-structured documents make it easier for your AI agent to find the right information and return accurate answers.

You do not need to rewrite every document from scratch, but documents usually work better when they are:

• Clear
• Well organized
• Up to date
• Free of unnecessary duplication
• Written with descriptive headings
• Focused on one topic at a time
• Easy for a person to read

The same qualities that make a document easier for a person to understand usually make it easier for Document Search to use.

Use documents for written knowledge

Documents are best for information that reads like written content.

Common examples include:

• Policies
• FAQs
• Manuals
• Guides
• Procedures
• Help articles
• Service information
• Product documentation
• Internal documentation
• Employee guidance

If your information is mainly organized as records, rows, columns, or lookup values, use Data Search instead.

Give each document a clear purpose

A document should have an obvious subject.

For example:

Good

• Cancellation Policy
• Employee Leave Policy
• Hotel Check-in Guide
• Product Installation Manual
• Customer Refund Process

Less clear

• Miscellaneous Information
• Notes
• General Content
• Everything Customers Need to Know

Focused documents make information easier to maintain and easier to find.

Use descriptive headings

Headings help organize the document into meaningful sections.

For example:

Cancellation Policy

Free cancellation period

Late cancellations

No-shows

Refund processing

Exceptions

This is better than one long block of text with no section structure.

Headings should describe the information that follows them.

Avoid vague headings such as:

• Other
• More information
• Details
• Notes

when a more specific heading is available.

Keep related information together

Keep information about the same topic in the same section.

For example, if a cancellation fee depends on how close the cancellation is to the appointment, keep the rule and the fee together.

Avoid placing the rule in one section and an important exception several pages later unless the document naturally requires it.

Clear context helps the AI agent understand how pieces of information relate to each other.

Avoid unnecessary duplication

Do not repeat the same rule across many documents unless there is a clear reason.

For example, if your refund policy appears in:

• A customer FAQ
• A service guide
• An internal manual
• A pricing document

and those versions are not identical, the AI agent may encounter conflicting information.

Keep one authoritative version where possible.

If the same information must appear in several places, make sure the versions remain consistent.

Remove outdated information

Old documents can create incorrect answers even when newer information also exists.

Before uploading or updating knowledge, check for:

• Old prices
• Old opening hours
• Old contact details
• Expired promotions
• Previous policies
• Old product names
• Retired services
• Outdated procedures

If information is no longer valid, remove or replace it.

Do not keep outdated content just because it may be useful for historical reference unless your AI agent actually needs that history.

Make the current version obvious

If you maintain different versions of a document, make it clear which one is current.

For example:

Current Cancellation Policy - October 2026

is easier to manage than:

Cancellation Policy Final v7 Updated New

Clear naming also helps your team know which source should remain active.

Use complete statements

Write important rules as complete, explicit statements.

Instead of:

Late cancellation: 50%.

Write:

Cancellations made less than 24 hours before the appointment are charged 50% of the service price.

The second version includes the condition, timing, and result in one place.

This reduces ambiguity.

Define important terms

If your document uses terms that may not be obvious, define them.

For example:

A “late cancellation” is a cancellation received less than 24 hours before the appointment start time.

This is especially useful when a business process depends on a specific definition.

Keep exceptions close to the rule

If a rule has an exception, keep the exception near the main rule.

For example:

Cancellations made less than 24 hours before the appointment are charged 50% of the service price.

This fee does not apply when the appointment is canceled by the business.

This is clearer than placing the exception in a separate “General Notes” section many pages later.

Use lists for clear sets of information

Lists work well when the content naturally contains several items.

For example:

Customers should bring:

• A valid photo ID
• Their confirmation number
• Any required documents
• A payment method

Lists make distinct requirements easier to understand.

Use tables when the table is genuinely useful

Tables can be appropriate for compact reference information.

For example:

Keep tables simple when possible.

If a table becomes very large or is mainly used for exact record lookup, the information may be better suited to Data Search.

Do not turn structured datasets into long documents

If you have hundreds or thousands of rows of structured information, use Data Search.

Examples include:

• Product catalogs
• Inventory
• Pricing tables
• Vehicle records
• Location lists
• Insurance factors
• Customer records

Document Search is better for written knowledge.

Data Search is better for structured lookups.

Avoid screenshots of important text

Whenever possible, keep important knowledge as actual text rather than only inside screenshots or images.

For example, a policy should be written in the document itself rather than appearing only as a screenshot of another system.

Text is easier to search, update, and maintain.

Prepare PDFs carefully

PDFs can work well for Document Search, but the content should still be readable and structured.

Good PDFs usually have:

• Selectable text
• Clear headings
• Normal reading order
• Simple page layouts
• Consistent sections

Be more careful with PDFs that are:

• Scanned images
• Highly visual brochures
• Multi-column layouts
• Forms with text spread across many boxes
• Pages dominated by diagrams
• Poorly exported

If the important information is difficult for a person to copy or read in order, consider creating a cleaner text-based version for your AI agent.

Separate marketing material from operational knowledge

Marketing brochures and operational documents often serve different purposes.

A brochure may say:

“Experience effortless luxury from the moment you arrive.”

A guest guide should say:

“Check-in begins at 3:00 PM. Guests arriving after 10:00 PM should use the after-hours entrance.”

For practical customer questions, the second type of information is usually more useful.

You can keep marketing content when it is relevant, but make sure operational facts are written clearly.

Write for the questions customers actually ask

Think about how customers will ask for the information.

For example, if your document contains:

“Reservation amendments are subject to the conditions defined herein.”

customers may ask:

“Can I change my booking?”

Make sure the document explains the rule clearly enough that both a customer and the AI agent can understand the connection.

You do not need to copy every possible customer question into the document.

You do need to make the underlying answer clear.

Use FAQ format when it fits naturally

FAQ sections can work well for common customer questions.

For example:

Can I cancel my appointment?

Yes. Appointments can be canceled without a fee more than 24 hours before the scheduled start time.

What happens if I cancel late?

Cancellations made less than 24 hours before the appointment are charged 50% of the service price.

FAQ format is useful when the information is already naturally question-based.

Do not force an entire manual into FAQ format if normal sections are clearer.

Keep one topic per section

Avoid sections that mix unrelated subjects.

For example, do not put cancellation rules, parking instructions, payment methods, and dress code under one heading called:

Guest information

Break them into clear sections instead.

This makes the document easier to scan and maintain.

Include the context needed to understand numbers

A number by itself may not be enough.

Instead of:

Fee: 50

write:

The late-cancellation fee is 50% of the service price.

Instead of:

Check-in: 3

write:

Check-in begins at 3:00 PM.

Include units, percentages, currencies, time zones, and conditions when they matter.

Be consistent with names

Use consistent names for:

• Products
• Services
• Locations
• Departments
• Membership levels
• Plans
• Policies

For example, if the official name is Premium Membership, avoid switching between:

• Premium Membership
• Premium Plan
• Gold Membership
• Premium Package

unless those are actually different things.

Consistent terminology makes the knowledge easier to maintain and reduces ambiguity.

Make dates clear

When a date matters, use a clear format and include the year when appropriate.

For example:

This policy takes effect on October 1, 2026.

is clearer than:

This policy takes effect next month.

Relative dates become outdated quickly.

Make location-specific rules clear

If information applies only to one location, state that directly.

For example:

The Manhattan location is open until 8:00 PM on weekdays.

Do not write:

We are open until 8:00 PM.

if that is only true for one location.

Location context is especially important for businesses with multiple branches or properties.

Make audience-specific rules clear

If a section applies only to a certain customer group, say so.

For example:

These cancellation terms apply to group bookings of 10 or more guests.

This prevents a rule from appearing broader than intended.

Keep documents maintainable

Do not optimize only for the AI agent.

Your team still needs to update the content.

A good document should be:

• Easy to edit
• Easy to review
• Clearly owned
• Easy to replace when information changes

Simple, well-structured documents are usually better than highly complex ones.

Split very large documents when it improves clarity

You do not need to split every long document.

A large manual can remain one document if it has a clear structure.

But splitting can help when one file contains several unrelated subjects.

For example, instead of:

Complete Company Information.pdf

you might maintain:

• Returns Policy
• Shipping Policy
• Warranty Guide
• Product Setup Guide

This can make ownership and updates easier.

Split by meaningful topic, not simply by page count.

Do not split content too aggressively

Very small fragments can also become harder to manage.

For example, creating a separate document for every individual FAQ may create unnecessary maintenance.

Group related information when it belongs together.

The goal is clear organization, not the maximum number of files.

Review conflicting content before uploading

Before adding several documents, ask whether they contain different answers to the same question.

For example:

Document A

Refunds are processed within 5 business days.

Document B

Refunds are processed within 10 business days.

Decide which rule is correct before relying on both documents.

Do not expect the AI agent to decide which conflicting source your business intended to follow.

Test questions after uploading a document

Once the document status is Ready, test the questions customers are likely to ask.

For example:

• Ask for a specific policy
• Ask the same question using different wording
• Ask about an exception
• Ask something that is not in the document

Confirm that the response reflects the source correctly.

Test detailed questions

Do not test only broad questions such as:

“What is your cancellation policy?”

Also test:

“Can I cancel 12 hours before?”

“What is the fee for a no-show?”

“Does the late-cancellation fee apply if you cancel the appointment?”

Detailed questions help reveal whether important conditions and exceptions are clear in the source.

Improve the source before adding complicated instructions

If Document Search struggles because the source itself is unclear, improve the document first.

For example, if three different sections describe the same refund rule differently, adding more AI Instructions may not solve the underlying problem.

A clean source is usually more reliable than complicated instructions trying to explain a confusing source.

A simple document structure

A practical document might look like this:

Cancellation Policy

Standard cancellations

Appointments can be canceled without a fee more than 24 hours before the scheduled start time.

Late cancellations

Cancellations made less than 24 hours before the scheduled start time are charged 50% of the service price.

No-shows

Customers who do not arrive and do not cancel are charged the full service price.

Exceptions

The fee does not apply when the business cancels the appointment.

Refund timing

Approved refunds are processed within 5 business days.

This structure is easy to read, update, and test.

Tips for preparing documents

• Give each document a clear subject.
• Use descriptive headings and sections.
• Keep related rules and exceptions together.
• Remove outdated or conflicting information.
• Avoid unnecessary duplication.
• Use complete statements instead of shorthand.
• Include units, dates, locations, and conditions when they matter.
• Keep important information as searchable text whenever possible.
• Use Data Search instead of documents for large structured datasets.
• Test realistic customer questions after uploading or updating a document.
• Improve unclear source material before adding complicated workflow rules.

Related guides

For more information, see:

• How can I upload my files?
• What is a Document Search node?
• Should I use AI Answer, Document Search, or Data Search?
• How should I prepare CSV and Excel files for Data Search?
• How should I test my AI agent before publishing?