Skip to content
Omnipresence

AI searches

The searches behind an answer, and the sources they pull.

Updated October 20, 2026

An assistant rarely answers from memory. It runs its own searches first, reads what comes back, and writes from that. Those searches, and the pages they surface, are the layer you can actually influence.

Fan-out

One prompt becomes several searches. They are usually not your keywords and often not phrased the way anyone would type. They are, however, the exact strings that decide which pages are in the running for your answer, which makes them more useful than a keyword list.

Each one is a discovery search, assembling a list of candidates, or a verification search, checking a brand or claim the model already surfaced. Those two failures look identical in the score and have nothing in common as work.

Retrieved is not cited

A page the assistant fetched is a candidate. A page it quoted shaped the answer. Omnipresence records both and labels which is which, because optimizing toward the wrong one wastes a quarter.

Matching names honestly

Deciding whether an answer mentioned you is harder than a string search: brands get abbreviated, misspelled, and confused with common words. Matching is corroborated across runs rather than trusted on one hit, and you can correct a mapping when it gets one wrong.

Concepts rather than clicks: where ChatGPT gets its information.

In this section

  1. 01Fan-out searchesThe searches an assistant runs before it answers, how Omnipresence captures them, and why they are the most actionable thing in a new project.
  2. 02Discovery and verificationThe 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.
  3. 03Cited and retrieved sourcesThe difference between a source an assistant fetched and one it actually quoted, and how to use the source list to find what stands between you and the answer.
  4. 04Source opportunitiesThe deduped list of sources models read for your market, with the role each page plays, whether your brand is on it, and how often it has been seen.
  5. 05Claiming a sourceMark a source as claimed once your brand is genuinely on the page, record what you found in notes, and keep the completion signal honest.
  6. 06Brand name matchingHow Omnipresence decides which names in an answer are brands, how to correct a mapping it got wrong, and why corroboration is used instead of one hit.

Not tracking yet?

Seven days free, on your own model key. Point it at one brand and read what comes back.

Start your free trial