AI Settings: Community Language Model (CLM)

Recommended Reference Boards Start with 1-3 high-quality boards (or Knowledge Base/Articles) so answers are consistently grounded. Avoid low-signal areas (for example, off-topic boards) during initial setup unless your test objective requires them.

Date Range and Filtering

Date Range & Filtering: choose Refresh Frequency and open Advanced settings.

Refresh Frequency

Refresh Frequency controls how often CLM re-runs ingestion to incorporate new or updated content and regenerate summaries. Shorter frequencies reflect changes faster; longer frequencies reduce churn.

Recommendation The refresh frequency should be optimized as per your own requirement on the basis of how engaged the community is, but generally 6 hours is a good balance to begin with. If you are actively modifying content and want faster visibility, temporarily reduce the refresh interval during the test window.

Advanced Date Range and Filtering Settings (Global)

Advanced Date Range & Filtering Settings (global): historical range, half-life recency, tag filtering, archived inclusion.

Setting	What it controls	Recommended for initial setup
Historical Data Range	How far back CLM ingests content (preset or custom).	Use Last 2-5 years(default) depending on the history you want to test.
Half-life Recency (weeks)	How quickly older content becomes less influential in relevance. Higher = older content stays relevant longer.	You can start  with the default 52 weeks for balanced approach. Generally speaking however, Fast-changing communities (e.g., frequent releases, time-limited offers, seasonal programs): use a shorter half-life (e.g., 8-26 weeks) so newer content is prioritized. While, slow-changing communities (e.g., stable products, long-lived how-tos): use a longer half-life (e.g., 52+ weeks) so older, authoritative content stays relevant.
Include Tag-based Filtering	Tag filtering limits which content CLM indexes (and which sources AA uses) based on tags. This helps keep answers focused on specific products, categories, or programs- but only works well if tags are used consistently. Practical tip: If you enable tag filtering, confirm your key support boards have enough tagged content- otherwise AA may answer less often or fall back to experts more frequently. Since only selected tags will be considered, others will be left out.	Keeping it Off is recommended initially. You should use tag filtering when your community uses tags reliably and you want more targeted answers. Otherwise, leave it off to avoid unintentionally excluding relevant sources.
Include Archived Posts	Whether archived content is included in indexing.	Off initially; enable only if you want archived content to be used(not recommended).

Per-board overrides (Edit modal on a source)

Per-board date range configuration: use Global Settings, choose a Preset Range, or set a Custom Range.

You can edit these settings if one or more boards need a different historical scope than the rest. By default, all boards would have the Global Date Range & Filtering settings enabled. It is recommended to use these settings only if needed, to not cause any confusion with regard to the behavior of the system. 

Example: when to use per-board overrides If you have an “Archive” board that is large and noisy, you can: • Keep your global settings broader for your main support boards, and • Use a narrower preset range (or exclude archived content) just for the archive node.

CLM build status and ingestion

CLM Builder overview (right side): total sources, ingested sources, contributors, last build time, and ingestion progress.

Use the CLM Builder panel to confirm ingestion is complete (for example, Content Ingestion shows 100% completed) and to see when the last build ran. Verify this before you expect reliable AA answers.

How CLM ingestion works (what you should expect during testing)

CLM ingests and processes community content in batches and in parallel to reduce build time on larger communities:

• Batch processing groups messages (up to about 100 at a time) to reduce repeated LLM calls.
• Multi-threaded ingestion processes multiple boards simultaneously.
• Incremental ingestion tracks progress, so refresh cycles typically process only new or modified content.

What “Build” vs “Refresh” means

• A Build is the process of ingesting your selected Reference Boards and generating/refreshing the CLM knowledge base so it can be searched and cited by AnswerAssist. Once, you’ve selected the reference boards you want to ingest into the CLM, and have determined the boards/settings in for the AA, you can start a build by clicking on the ‘arrowhead’ icon on the right (add pic)
  
  
• A Refresh is the scheduled re-run of ingestion (based on your Refresh Frequency) to pick up new or updated content. A refresh is usually faster than the first build because it primarily processes new/changed items.

How long does it take? (what you should expect)

Build and refresh times depend on your content volume, number of boards selected, and how much content has changed.

• First build (initial indexing): typically takes the longest. Expect it to take longer if you include many boards or a large date range. Usually, from several minutes to a few hours, depending on the volume of boards you ingest.
  
  
• Subsequent refreshes: typically complete much faster because they mostly process deltas (new/updated content in any particular board). 

For initial setup: Start with 1–3 Reference Boards(or KBs/Articles) and a smaller historical range. Once you confirm quality and behavior, expand coverage gradually.

What happens if you enable AA before CLM is fully built

AA can be enabled at any time, but it relies on CLM to find sources and produce strong, cited answers.

• If CLM is not fully built, AA may:
• produce fewer answers (confidence is lower because fewer relevant sources are available),
• show weaker/limited citations, or
• fall back to Expert Suggestions more often (if enabled), because it can’t find enough evidence to answer confidently.
  
  

Best practice: Enable AA only after CLM ingestion shows Completed (or plan to treat early AA behavior as “expected to improve” as CLM finishes building).