How search works

How Raffle finds answers

Raffle searches your sources in two ways at once:

• keyword search, which finds the sections that contain the words of the search;
• semantic search, which uses a multilingual language model to find sections that mean the same thing, even when they use other words.

The results of both are combined into one ranking. How recently a page was updated can also play a part.

The language model is shared and already trained. Raffle doesn’t train a model on your content, so there is nothing to wait for: new and changed content is searchable once Raffle has crawled or received it. See Sync frequency.

Answer sections

Each page is divided into sections. Raffle finds the most relevant section of each page, shows it as the snippet in the results, and ranks the answers.

The section shown may not contain the exact words of the search.

Synonyms and misspellings

The semantic search understands the meaning of a search, so synonyms are handled without any setup.

For your own terms, abbreviations or product names, add them as questions on the page they belong to, see Add questions to a page, or use Re-rank.

Order and ranking

The order of the answers and the section shown are decided by the search. You can’t set them by hand, but you can:

• put a chosen answer first for a question with Re-rank;
• write your own answer with Instant Answers when your website doesn’t have one.

Search data

Search, Chat and Summary use only the sources connected to the widget as the basis for their answers.

• Chat and Summary use the top search results as references for the answers they generate.
• Generated answers can’t be changed by hand, but you can influence them by improving the search results with Re-rank and Instant Answers.

GDPR compliance

Raffle is GDPR-compliant: the search does not use, set or store cookies, and anonymizes content in search questions that are passed through any Raffle widget.

When a user searches for a phrase, the original question is used to search for relevant answers. The question is then anonymized with “tags” before it is saved for search analytics. Simply put, Raffle does not keep questions that contain sensitive information in the database. These questions appear in the Trending Questions and Content Gaps lists.

Question tags:

• CPR/SSN/ID numbers are replaced with “<ID>”
• E-mails are replaced with “<EMAIL>”
• Phone numbers are replaced with “<NUMBER>”

Data clusters

Clusters group questions in a folder structure, so you can explore the data per widget at different levels.

The questions in the Trending Questions and Content Gaps lists are clustered by relevance: a question has to be related to other questions to be part of a cluster and shown in the list.

Trending Questions clusters are sorted by how often the questions are asked, and Content Gaps by how similar the questions are to the answers.

Cluster titles are based on the most common topics of a group, using all the data Raffle has collected. If a widget has been live for a year, the titles are based on the whole year. So a title may include words that aren’t in the questions for the selected time frame.