The search is returning a large string of words-only and/or the beginning of the AI response contains some individual words before the sentence-based response.

Occasionally, on very large documents, the AI auto indexing will sometimes not be able to determine the actual meaning of the resource and the index will therefore be mostly a giant mass of text that is extracted from the URL. This is a rare occurrence, but it can happen on very large text pages such as terms of use agreements or large multi-page PDF documents.

This may also occur on webpages that contain a lot of superfluous content in addition to the main (primary) body content such as affiliate links, cross-selling lists, ads, large headers and footers with tons of links, etc

The solution is to simply add a summary paragraph of the content of the document into the index field and then resave it. You can also use the built-in search excerpt generator to create a more accurate search description for the specific resource. If needed, you can also “pad” the index with addition keywords or phrases that will help the AI engine understand the purpose and meaning of the URL resource.

This may also occur If your document is split into sub-documents by the indexer. When reviewing your imported content, if you discover that some of your large documents were split up into smaller sub-records, we recommend reviewing the search excerpts and index for those records and manually updating them if required. More information

Note: whenever you manually update the index field, we recommend marking the item to be excluded from the auto-indexing.

You can also automate the addition and updating of large documents directly through the API.