Discovery and verification
The two kinds of fan-out search a model runs, how to tell them apart in Omnipresence, and why each one calls for a different piece of work.
Updated October 20, 2026
A model answering a buying question runs two kinds of search, and the difference decides what work is worth doing.
Discovery
A discovery fanout goes looking for candidates. "Project management tools for design studios", "best emergency plumber Bristol", "alternatives to the thing the user named". The model does not yet know who belongs in the answer and is assembling a set.
If you are not in the answer at all, your problem is here. You are not in the sources the model reads when it builds the candidate list, and the work is being present on those pages: your own page that serves that query, or the third-party roundup that already answers it.
Verification
A verification fanout checks a specific claim or brand the model has already surfaced. "Is this company reputable", "does this product do that", "pricing for that named tool". The candidate exists and the model is deciding how to describe it, or whether to keep it.
If you are named and then described weakly, hedged, or dropped from the final answer, your problem is here. The work is evidence: a page that confirms the claim, a review presence, a specific and checkable statement of what you do.
Telling them apart
Each fanout is labelled, so the list can be read either way. A rough test by eye: a discovery query names a category and a constraint, and a verification query names a thing.
Why the split changes the plan
The two failures look identical in the Visibility Score and have nothing in common as work.
| Symptom | Likely fanout kind | Work |
|---|---|---|
| Never named | Discovery | Be in the candidate sources: a page for the query, or a placement on the roundup that answers it. |
| Named, then hedged | Verification | Give the model something to confirm: specific claims, dates, numbers, third-party corroboration. |
| Named last, consistently | Verification | The evidence is thinner than the brands named before you. |
| Named only on one assistant | Either | Compare that assistant's cited sources against the others first. |
Using it in the work queue
When you commit to working a query, flag that fanout as being optimized. That stamps the date every before-and-after measurement is drawn from, and puts it in the queue your agent reads.
Flag it when you have actually agreed to do the work. It is a commitment, not a bookmark, and a queue full of maybes is worth nothing to the next session.
Questions about discovery and verification
- Which kind should I work on first?
- Discovery, if you are not in the answer at all. Verification, if you are named and losing to a better-evidenced rival.
- Can one query be both?
- In practice a query is one or the other, because a model asks them for different reasons at different points in the answer.
- Do verification searches mention my brand?
- Often, yes. A model that has surfaced you then goes looking for confirmation, and what it finds decides how it describes you.