See where your buyers look and your brand is missing.

What a cycle does

Probe, diagnose, prescribe, re-measure

1. Ask what your buyers ask

   A prompt corpus built per sector and per language. Each raw response is stored with its provider, model version, country and cost, and kept.

2. Work out why you are missing

   Absence is sorted into one of four causes, using the crawl of your own site on one side and the record of cited sources on the other. A single score would tell you nothing you could act on.

3. Get the named change

   Pages to write, schema blocks to paste, placements to pursue, posts and audiences to use. Drafted, ordered by how often buyers ask, and reviewed before you see them.

4. See what moved

   Probes run again on a cycle. The result is compared with the baseline taken when the fix was prescribed, next to unfixed gaps on the same site.

Four reasons a model does not mention you

Each reason has a different fix. Telling them apart is the job.

• entity_absent

  You are not in the source material

  Nothing on your site, and nothing about you elsewhere, answers the question. The fix creates the entity.

• present_not_extractable

  You answer it, but it cannot be read

  The content exists, but sits in a PDF, a carousel or prose with no structure. The fix is structure, not new writing.

• third_party_sourced

  Someone else is the source

  A directory, a trade title or a review site is what the model reads. The fix is a placement, not a page.

• competitor_owns_source

  A competitor owns the cited source

  The cited source names them and not you. The fix targets that source.

If the probe did not return it, it is not in your report

Every stored response carries the same fields. A reader can open any finding and see exactly what was asked, where, by whom, and what it cost.

Every gap, citation and presence figure is stored against the raw response that produced it. A value without provenance is left out, never estimated.

surface

Where the question was asked

model_version

The exact model that answered

query_text

The question, word for word

language

The language of the question

country

The market of the question

requested_at

When it was asked

raw_payload

The full response, kept

cost_micros

What the call cost

Findings are written in the language of the question. Nothing is written in English and then translated.

Raw responses and the record of what was shipped are archived, never deleted.

For agencies

Run it across your client book, under your own name. Your reviewer for each market approves the work in that language, and a language opens only when a reviewer is in place.

For European mid-market firms, the agencies that serve them, and European creators and solo businesses.

What procurement will ask about

• Every sub-processor is named. A request to a provider outside the allowlist fails and is never rerouted.

• Unveilio never writes to your website, your content system or your DNS. It writes the change. Your team applies it.

• Connections to your ad and analytics accounts are read-only today. Making changes in ad accounts is a later step, and it opens only after the controls for it are in place.

• Output that has not passed human review carries a machine-generated label.

See the sub-processors and residency facts