Writing good content

Learn how to structure and write content that your bot can retrieve and use accurately.

How retrieval works

When a visitor asks a question, DGbot:

1. Converts the question into a vector (a numerical representation of meaning)

2. Searches your knowledge base for chunks of text with similar vectors

3. Sends the most relevant chunks to the AI model along with the question

4. The model writes an answer based only on those chunks

The quality of the answer depends entirely on whether the right chunks were retrieved and whether those chunks contain clear, complete information.

Structure your content well

Use headings. Headings divide content into retrievable chunks. A page with headings like "Return policy", "Shipping times", and "Warranty" can surface each topic independently. A wall of text forces the retrieval system to guess which paragraph is relevant.

One topic per section. Mix "shipping" and "returns" in the same paragraph and the bot may give a combined or confused answer to either question. Keep topics separated.

Use numbered steps for processes. "To return an item: 1. Log in to your account. 2. Go to Order history. 3. Click Return." is much easier for the bot to extract than a narrative paragraph about the same process.

Put the answer first. Don't bury the answer at the end of a long preamble. Start each section with the key information, then add context.

Writing style

Use plain language. Write at a reading level accessible to someone unfamiliar with your industry. Acronyms, technical jargon, and internal product codes should be defined on first use.

Be specific. "Delivery takes a few days" is worse than "Standard delivery takes 3–5 business days." Specific answers build trust; vague answers frustrate visitors.

Present tense, active voice. "We ship orders within 24 hours" not "Orders will be shipped within 24 hours by our fulfilment team."

Include the specific details visitors need. Phone numbers, email addresses, URLs, business hours, prices, deadlines. If your content is vague, the bot's answers will be too.

What to avoid

Duplicate content. If the same information appears in slightly different wording across multiple pages, the bot may return inconsistent answers. Keep one authoritative source per topic.

Marketing language without substance. "We offer the best customer experience in the industry" contains no retrievable fact. Visitors asking about your customer support process need specifics, not superlatives.

Long legal or compliance text. Dense terms-and-conditions documents produce poor retrieval because every chunk looks similar to every other chunk. Summarise the key points in plain English as a separate source.

Outdated information. Prices, policies, and product names change. Set a reminder to audit sources quarterly. Outdated content is worse than no content — the bot confidently gives wrong answers.

Contradictory information. If your FAQ says returns are accepted within 30 days and your website says 14 days, the bot will return one or the other — or both — inconsistently. Resolve contradictions before training.