Build a Business Knowledge Pack Before Asking AI to Market Your Company

A business knowledge pack is a focused collection of reliable material that explains what your company does and what it can honestly say. It gives an AI marketing workspace the context that an experienced colleague might otherwise supply in conversation.

It is not a data dump. A hundred files containing old offers, unexplained abbreviations and contradictory descriptions can make the task harder. A smaller collection with clear ownership and current information is easier to use and review.

In Réson, Knowledge and Profile provide places to establish business context. The preparation below helps you decide what to include before asking for public-facing work.

Start your business knowledge pack with a reference sheet

Write one page that a new colleague could understand without an introduction. Include the business name, official website, main products or services, intended customers and preferred enquiry route. Add a review date and the person responsible for maintaining it.

Then explain the boundaries. What do you not provide? Which requests need a separate assessment? Are there geographic or practical restrictions?

A hypothetical design studio might write:

We design brand identities for independent food businesses. Packaging design is quoted separately. We do not provide trademark registration or legal clearance.

That final sentence matters. Without it, a polished draft might expand “branding” into a service the studio does not sell.

Use plain descriptions here. The reference sheet should help someone make accurate decisions, not impress them with slogans.

Create a record of supported claims

Separate facts you can demonstrate from impressions you hope to create.

“Available in three sizes” can be checked against a current catalogue. “The perfect fit for everyone” is a much broader promise. “Designed for compact desks” describes intent; “fits every small office” implies evidence that the dimensions alone cannot establish.

For important claims, record the approved wording and its source. Include any condition that changes the meaning. If information can become stale, give it a review date.

A simple entry might read:

• Claim: The introductory consultation lasts 30 minutes.
• Source: Current booking-page description.
• Condition: This describes the consultation, not the complete service.
• Owner: The person responsible for bookings.

An old brochure is not automatically reliable evidence. Check whether its claims still apply before using it as source material.

Capture customer questions in customer language

Add questions that repeatedly come up before a purchase. “Can I book without knowing which service I need?” is more useful than “top-of-funnel education.” It tells the writer exactly what a reader is trying to resolve.

Answer directly. If the answer depends on a condition, state it. If someone must inspect a product or discuss a requirement before answering, explain that process rather than inventing a universal answer.

Customer questions can suggest strong content without becoming testimonials. Keep those categories separate. A question does not establish that the person endorses the company or agrees to appear in its marketing.

Anonymise examples. Most of the time, the question is useful without a name, email address or full conversation attached.

Include writing examples with editorial notes

A few well-chosen examples can teach more than a long list of adjectives.

Select writing that represents how the business should sound now. Beneath each example, explain what works: perhaps it answers the question immediately, explains technical terms carefully or uses humour without weakening the message.

Include an example you would not publish and explain why. “Too much sales language before the reader understands the offer” is actionable. “Not our vibe” leaves the next writer guessing.

Do not treat third-party writing as material you own or may reproduce. You can describe a preference without copying another company’s distinctive text.

Keep tone guidance separate from factual claims. An informal voice does not justify looser statements about pricing or performance.

Give assets useful instructions

An asset collection needs context. Identify the current logo, the relevant product version and any restrictions on use. Explain when a photograph must remain exact rather than becoming a generated interpretation.

A product image can be attractive and still be misleading if it changes a connector, label or physical feature. The reviewer needs enough information to notice that difference.

For photographs of people, establish permission for the intended use. For commissioned work, check the rights you received. Do not assume that possession of the file settles either question.

A service business may get more value from a few approved photographs of its actual process than from a large library of generic imagery.

Separate public context from internal information

Some material helps a team understand a business but should not appear in marketing. Internal margins, supplier negotiations and unpublished plans are common examples.

Réson provides visibility distinctions for business knowledge. Apply the appropriate setting and still review outgoing content. Classification is part of the workflow, not a reason to stop exercising editorial judgement.

Remove secrets and unnecessary personal data before adding files. If only one paragraph is relevant, consider preparing a clean reference rather than uploading an entire internal document.

A useful pack contains information for a defined purpose. It does not need to become a second copy of every company record.

Test the pack with a small assignment

Ask for one factual, low-risk draft, such as an explanation of how to request a quote. Compare each business-specific statement with your sources.

When something goes wrong, diagnose the information problem. Was a fact missing? Were two sources inconsistent? Did a vague description invite an overly broad interpretation?

Improve the source before adding more material. Otherwise, the same weakness may appear in a post, an email and a video script.

Keep a lightweight maintenance rule: every important source has an owner, and frequently changing details receive a fresh check before use. Retire superseded documents or make their status clear.

The goal is not to make AI know everything about the company. It is to supply dependable context for work you can inspect and improve. See Réson’s product overview and the company knowledge guide for the current workspace instructions.

Related reading

• What Réson does
• Build a consistent AI brand voice