News
Knowledge Sources Are Now More Relevant, Transparent and Reliable
Kojable has released two connected improvements to how Knowledge Sources are used during article generation.
The platform now applies a stricter relevance process before trusted information can enter generation. It also distinguishes between information that was reviewed, retrieved, included in the generation context and meaningfully reflected in the finished article.
When strongly relevant context is overlooked, Kojable can make one carefully controlled second attempt—but only when doing so improves the article without weakening its quality, evidence, originality or safety.
What changed
Knowledge Sources must now earn their place in an article
Knowledge Sources allow teams to provide company information, research, internal expertise, proprietary insights and other trusted material that can help generated content become more accurate and specific to the business.
But making more information available to an AI model does not automatically improve an article.
The information must be relevant to the exact subject being covered, and the platform must be honest about whether that information actually influenced the final result.
Kojable now applies stricter topic-level relevance checks and more transparent source-usage reporting throughout the generation process.
Answer: Knowledge Sources are only eligible when they have a genuine connection to the article, and Kojable only describes them as grounding the article when their contribution can be detected in the finished content.
Selecting relevant information
Trust and recency are not enough on their own
Previously, a Knowledge Source could be considered during generation because it was recent, trusted or related to the company’s wider business context, even when it had only a weak connection to the article’s specific topic.
That could make relevant first-party expertise compete with company information that happened to be available but did not belong in the article.
Kojable has now introduced a stricter relevance process.
Before Knowledge Source content can be included, it must demonstrate a meaningful connection to the article through signals such as:
- Topic and keyword relevance
- Matching titles, headings or tags
- Connection to the article’s topic cluster
- Meaningful overlap with the subject being covered
Trust, recency and source authority still matter.
They can strengthen information that is already relevant, but they can no longer make unrelated information eligible on their own.
Answer: a Knowledge Source must first match the article’s topic. Being trusted or recent cannot compensate for being unrelated.
Why this matters
A company may have a large collection of valuable internal material, but not every document belongs in every article.
Stricter relevance selection makes the platform less likely to:
- Introduce unrelated company information
- Force a proprietary point into the wrong subject
- Distract from the reader’s intent
- Make an article appear personalised without adding genuine value
- Allow broad company context to overwhelm topic-specific evidence
The aim is not to use the largest possible amount of first-party information.
The aim is to identify the information that improves this particular article.
Transparent source usage
Kojable now separates four different stages of Knowledge Source usage
Providing information to an AI model does not prove that the model used it in the final response.
Kojable now distinguishes between four separate stages:
-
Reviewed for relevance
The source was examined to determine whether it matched the article.
-
Relevant information retrieved
Specific source material passed the relevance process and was selected.
-
Included in the generation context
The selected information was supplied during article generation.
-
Meaningfully reflected in the finished article
The final article contains a detectable contribution from the source.
These stages should never be collapsed into a single “used” label.
A source can be relevant and included in the generation context without appearing meaningfully in the finished article.
Answer: Kojable only says a Knowledge Source grounded an article when the final content contains a detectable contribution from that source.
Honest reporting when context is overlooked
When relevant information was included in generation but did not appear in the finished article, the platform reports that distinction clearly.
It should not claim:
- The article was grounded by the source
- The source materially influenced the result
- The model followed the supplied context
- The final article contains proprietary insight that cannot be detected
This gives users a more accurate explanation of what happened during generation.
A controlled second attempt
Strongly relevant context receives one more opportunity when it is overlooked
Sometimes trusted, safe and strongly relevant Knowledge Source information reaches the article-generation context but is not reflected in the first draft.
Kojable can now make one focused second attempt in this situation.
The second attempt asks the model to add one useful source-grounded element where it naturally improves the article.
That contribution might be:
- A practical example
- A workflow or methodology
- A decision rule
- A measurement approach
- A relevant use case
- An audience-specific recommendation
- A distinctive business point of view
The system does not force every source into every article.
It does not ask the model to repeat the source verbatim or insert company information simply to prove that the source was present.
Answer: the second attempt is used only for strongly relevant context and only when the source can improve the article naturally.
The revised article must still pass quality checks
A second attempt is not accepted automatically.
The revised article must preserve:
- Topic relevance
- Search and reader intent
- Factual and evidential quality
- Originality
- Citation requirements
- Safety requirements
- Overall readability
- The intended article structure
- Appropriate use of the Knowledge Source
When the second attempt makes the article worse, uses the source unnaturally or still fails to reflect the source meaningfully, Kojable retains the original article.
Source usage never overrides article quality.
What this means for users
More relevant content without forced personalisation
These improvements provide several practical benefits.
-
More relevant content
Only Knowledge Sources with a genuine connection to the article topic are eligible for use.
-
Less forced personalisation
The platform does not insert unrelated company information merely to make an article appear customised.
-
Better use of proprietary expertise
When valuable first-party information is relevant but overlooked, the system receives one carefully controlled opportunity to incorporate it.
-
More honest explainability
Users can distinguish between information that was checked, retrieved, included during generation and actually reflected in the finished article.
-
Stronger quality protection
Knowledge Source usage never overrides article quality, safety, evidence requirements or citation policies.
-
Greater confidence in the finished article
The article is evaluated according to what it actually contains—not simply according to what information was available to the model.
A dependable content process
More context should produce better judgement, not more noise
Knowledge Sources should not simply increase the amount of information placed into an AI prompt.
A dependable process should:
- Identify information that is relevant to the exact topic
- Use it only where it improves the reader’s understanding
- Preserve the distinction between retrieval and actual influence
- Protect the quality and intent of the article
- Explain honestly what contributed to the finished result
These changes are part of Kojable’s wider effort to make AI-assisted content more grounded, auditable and aligned with the reality of each business.
The goal is not to force company information into every article.
The goal is to use the right information, in the right context, and communicate accurately about what shaped the result.
Frequently asked questions
Knowledge Sources relevance and grounding questions
What are Knowledge Sources?
Knowledge Sources are trusted materials supplied to Kojable, such as company information, internal expertise, research, proprietary insights, methodologies and other approved first-party context that may help improve relevant generated content.
Does every relevant Knowledge Source appear in an article?
No. A source can be relevant without being necessary for the final article. Kojable only uses source information where it adds value naturally and preserves the article’s quality and intent.
When does Kojable say a Knowledge Source grounded an article?
Kojable only describes a source as grounding the article when a meaningful contribution from that source can be detected in the finished content.
What happens when relevant source information is overlooked?
When strongly relevant, trusted and safe information was included during generation but is absent from the first draft, Kojable can make one focused second attempt to incorporate a useful source-grounded element.
Is the second version always accepted?
No. The revised article is accepted only when it uses the source meaningfully while preserving relevance, evidence, originality, safety, readability and overall article quality. Otherwise, the original article is retained.
Can Knowledge Sources override evidence or citation requirements?
No. Knowledge Source usage does not override article quality, safety, evidence requirements or citation policies.
Next step
Bring relevant company knowledge into the content process
Use trusted company information where it genuinely improves an article—and understand whether that information actually influenced the finished result.