How to optimize your content for the AI chatbot

AI chatbot overview

The AI chatbot generates answers by searching your knowledge base, retrieving the most relevant articles, and synthesizing a response. The quality of its answers depends on how well your content is structured and written. This article covers practical best practices for getting the most out of the chatbot.

How the chatbot finds and uses your content

When a reader asks the chatbot a question, it searches your knowledge base by looking for both the meaning behind the words and exact keyword matches. It identifies the most relevant articles based on how well individual sections score against the question, then reads the relevant sections of those top-scoring articles to build an answer.

The chatbot can pull from multiple articles in a single response (typically 3-5), but it only works with articles that score highest for that particular query. Content that doesn't rank well for a question won't be used, regardless of how good it is.

This has two practical implications:

1. Your article needs to score well for the right questions so the chatbot finds it in the first place. Titles, headings, and terminology all affect the score.
2. Related content in the same article is more likely to be used together, so grouping related information helps the chatbot give more complete responses.

Content best practices to optimize chatbot performance

In this section, we cover some of the ways you can structure your content to optimize the chatbot's ability to answer questions.

Write clear, descriptive headings

The chatbot uses your headings (H1-H4) to understand what each section of an article is about. A heading like "Reset your password" gives the chatbot clear context, while a heading like "Overview" or "Step 3" doesn't tell it much on its own.

Write each heading as a complete, descriptive phrase.

Here's a good test: Could someone understand what the section covers from the heading alone?

Keep sections focused on one topic

Each section (the content between one heading and the next) should cover a single concept. If a section addresses multiple unrelated topics, the chatbot may have trouble understanding what the section is about.

If a section is growing beyond about three paragraphs, consider breaking it into subsections or a separate article.

Put the most important information first

Within each section, lead with the core answer, definition, or key point. Don't bury important details like feature names, error messages, or critical steps deep in a long section. The chatbot is more likely to surface content that gets to the point quickly.

Use the same words your readers use

The chatbot uses keyword matching as part of its search, so exact terminology matters. If your readers call something "the print button," use "print button" in your article rather than "output control" or "browser print function."

A few specific tips:

• Include common synonyms in the same section when they're genuinely interchangeable (e.g., "sign in / log in").
• Include error message text verbatim so the chatbot can match on them.
• Use search phrases sparingly but strategically. Since the chatbot draws from the top of your search rankings, search phrases that boost an article's ranking will also increase the likelihood of the chatbot referencing that article.

Use numbered steps for procedures

The chatbot handles numbered lists well and can extract clean answers from them. For step-by-step instructions, use a numbered list rather than writing the steps out as a paragraph.

For example, this is easy for the chatbot to work with:

1. Open Settings.
2. Select Security.
3. Select Change password.

This is harder:

To change your password, open Settings, then navigate to Security, where you'll find an option to change your password.

Write descriptive link text and image alt text

The chatbot indexes your link text and image alt text and can include them in its responses.

• Use descriptive link text like "Refer to our SSO configuration guide" instead of "click here."
• Write meaningful alt text for images and screenshots (e.g., "Settings panel showing the SSO toggle enabled") rather than leaving it blank or writing "screenshot."
• Avoid using URLs longer than 200 characters as hyperlinks in article body text, since they may get truncated in the index.

Include text summaries alongside diagrams and screenshots

The chatbot cannot read the content of images. If a diagram or screenshot contains the answer to a reader's question, the chatbot won't be able to use it unless the surrounding text also explains what the image shows. Always include a brief text summary of key visuals, either using text before or after the image or by Adding image captions.

Write article titles as questions or tasks

Your article title carries significantly more weight in search results than a match in the article body. You'll have better performance if you write the title as the question or task that the article answers. For example, "How to configure SSO with Okta" will outperform "SSO Configuration" for most reader queries.

Your first H2 heading should reinforce the title's topic to give the chatbot a strong signal about what the article covers.

Use the meta description

The article Meta Description field is indexed and serves as a high-signal summary of your article. A well-written one- to two-sentence description that captures the article's purpose gives the chatbot another way to find and surface your content.

Refer to Meta descriptions for more information. And if you're not great at writing meta descriptions, consider using our feature to Generate article meta description with Owl Intelligence.

Keep tables reasonable and descriptive

The chatbot handles tables well for small-to-medium datasets. However, tables over 100 rows or 100KB are summarized rather than fully indexed, which means the chatbot won't search the actual data in those tables.

If you have large reference tables, consider splitting them across multiple articles. If the content isn't something you expect the chatbot to index or search, consider providing them as downloadable files.

Use descriptive column header names, since the chatbot uses them to understand the data.

Keep articles from getting too long

There's an upper limit on how much content per article the chatbot can index. As a general guideline, try to keep articles well under 10,000 words. Most articles will be well under this limit, but if you have a very long reference article (such as a full API reference), consider splitting it into separate articles by topic or endpoint.

Content that exceeds the limit is silently dropped during indexing, so it won't be searchable or usable by the chatbot. Splitting into separate articles won't reduce the quality of answers. The chatbot can pull from multiple articles in a single response.

Consolidate related content when it makes sense

Consider grouping related content together in a single article, as it can help the chatbot give more complete answers. For example, combining how-to steps with related troubleshooting tips in one article lets the chatbot address follow-up issues the reader may not have explicitly asked about.

That said, if grouping related content together leads to very long articles, don't do it! A tightly-scoped article with a clear title (e.g., "How to reset your password") will score better for that specific question than a section buried inside a broad article titled "Account management." The chatbot can pull from multiple articles per response, so splitting genuinely distinct topics is fine.

A good rule of thumb: Consolidate closely related content (how-to + FAQ + troubleshooting for the same feature), but give genuinely distinct topics their own articles with descriptive titles.

Exclude deprecated or placeholder content from search

Draft articles are automatically excluded from the chatbot's index, so you don't need to do anything extra for those. For published articles that contain deprecated information or placeholder text, check the Exclude from search results setting in the article's Display Settings to exclude them.

The chatbot builds answers from what it finds, so outdated or incomplete content can lead to incorrect responses.

Use hidden text for additional chatbot context

The chatbot processes all text in the article body, including text that isn't visible to readers. This means you can add hidden text to give the chatbot additional context without changing what your readers see. For example, you could add supplemental information using white text on a white background or a <span> with display: none. The chatbot updates as soon as you resave an article with text changes, including hidden text changes.

This approach is best treated as a targeted workaround for specific situations rather than a standard practice.