Retrieval readiness
How Omnipresence scores one of your pages against a fan-out query: the relevance and n-gram thresholds, the Ready, Borderline and Weak bands.
Updated October 20, 2026
Readiness answers a narrow, useful question: if a model went looking for this query, is there a passage on this page worth lifting? It works on your own pages, it is scored per query, and it moves when you change the page rather than when the market does.
How a page is scored
- The live page is fetched and reduced to clean text.
- The whole page is split into chunks.
- Every chunk is scored against the query, and the best one is kept.
- That best chunk, and its two numbers, are what you see.
Scoring the whole page and keeping the best chunk matters: a page can have one excellent answer buried two thirds of the way down, and a scorer that only read the opening would call it weak.
The two numbers
Relevance, target 105. How well the best chunk answers that specific query. It is a scoring-model output rather than a percentage, which is why the target is a number to clear rather than a share.
N-gram, target 3. How much literal phrase overlap the chunk shares with the query. Retrieval is not purely semantic, and a passage that never uses the words the query uses is harder to match.
The two are deliberately different tests. Relevance can be high while n-gram is low when a page answers the question in entirely different vocabulary, which is a real and common failure.
The bands
| Band | Meaning |
|---|---|
| Ready | Relevance 105 or more and n-gram 3 or more. Both tests cleared. |
| Borderline | One of the two short. Close enough that a small edit usually fixes it. |
| Weak | Neither. There is no passage on this page serving this query. |
Borderline is where the cheap wins are. A page with good relevance and a low n-gram often needs one sentence that uses the query's own words.
The folder rollup
Retrieval by folder stacks Weak and Borderline counts per section of the site, so you can see where the work is without reading a page list. Sections with the most weak pages are usually the ones written for a buyer who already knows what they want, rather than for the question a model is asking.
Short queries
A very short fan-out query cannot produce meaningful phrase overlap, so the n-gram test is waived for those rather than failing every page against them. Without that, a two-word query would mark your entire site Weak.
When to score
After you change a page. The scorer fetches live, so there is no crawl to wait for.
Tracked pages are also re-scored on a weekly cycle, which is what keeps the folder rollup current without anyone pressing anything.
Do not expect presence to move the same day a score does. Each assistant re-fetches your page on its own schedule, and that lag runs from days to weeks and differs by assistant. The score is the leading indicator; the Visibility Score is the lagging one.
Questions about retrieval readiness
- Is the score about the page or about the query?
- Both. A page is scored against a specific fan-out query, so the same page can be Ready for one query and Weak for another. There is no single score for a page in the abstract.
- Why did a page score lower after I improved it?
- The scorer fetches the live page each time it scores, so a rewrite is reflected immediately. If the score fell, the chunk that used to answer the query no longer does.
- Does a Ready page get cited?
- Readiness is necessary rather than sufficient. A perfectly extractable page about something nobody asks about will still never be quoted.