An agency can show a client rankings that are holding steady and still get the same question every quarter: "So why are we not in the AI answers?" Organic performance and AI visibility have come apart, and the standard report has nothing to say about it. The way through is to stop reporting a number and start reporting evidence.
Organic performance and AI visibility have come apart. Rankings hold, and the AI answers still do not mention you. The standard report, rank tracking and traffic volume, has nothing to say about the thing actually happening: the referral pathway is eroding. Agencies that answer with a vanity visibility score are papering over it. The way through is to pivot to forensic Generative Engine Optimisation, and it starts by naming a number the rank tracker cannot see.
Traditional SEO measures Share of Voice through click-through rate. In the AI era that is incomplete. Unlinked Share of Voice (USoV) is the gap between an AI's ingestion of your content and its citation of it. If an AI crawler, verified by forward-confirmed reverse DNS and published IP ranges, pulls your content but the assistant's answer never cites the source, you have supplied the value and received none of the credit. That is asymmetric value extraction: the publisher provides the data, the model withholds the attribution.
For an agency, the Evidence Report is the document that changes a retention meeting. Show a client several thousand verified crawler requests against a 0% citation rate, and the question stops being "are we ranking?" and becomes "why is our intellectual property being ingested without attribution?" That single reframe moves the agency from a vendor delivering a service to an advisor delivering a technical audit. And unlike a score, it is auditable line by line, which is what makes it survive a hard client conversation. It is the same distinction we draw in receipts versus an inferred score.
Phase 1 — the baseline audit. Use forensic crawl logs to establish the client's current USoV: which crawlers are accessing the site, and which specific topics are failing to produce citations. That gap list is the whole plan.
Phase 2 — make the page quotable. Work the exact topics the client has demand for but is not cited on:
Phase 3 — evidence-based iteration. This is where the agency earns the retainer. After the content work, re-scan the exact topics that were failing and show the client the before-and-after: did the assistants begin citing the page, or not? The movement is measured, sourced and dated, evidence of lift on that topic, not a claim of causation. Some changes land and some do not; the point is that, for the first time, the client can see which. This is the loop; the difference between a report and a hunch is covered in the three numbers behind AI visibility.
By quantifying Unlinked Share of Voice, the agency sets a new benchmark for performance, one built on evidence rather than estimates. The client is no longer paying for rankings; they are paying for the forensic monitoring and evidence-led improvement of their standing in the AI ecosystem. That is the transition from service provider to technical partner.
It is the gap between an AI’s ingestion of your content and its citation of it. If a verified AI crawler pulls your content but the assistant’s answer never cites the source, you supplied the value and received none of the credit. That is asymmetric value extraction, and a rank tracker cannot see it. Naming and measuring it is the core of the forensic GEO pitch.
It moves the agency from a vendor delivering a service to an advisor delivering a technical audit. Show a client several thousand verified crawler requests against a 0% citation rate and the question stops being "are we ranking?" and becomes "why is our IP being ingested without attribution?" That reframe anchors a retainer around evidence rather than estimates.
Phase 1, the baseline audit: use verified crawl logs to establish current Unlinked Share of Voice and the topics failing to produce citations. Phase 2, make the page quotable: lead with the self-contained answer, make claims traceable, cut what no one is asking. Phase 3, evidence-based iteration: re-scan the exact topics and show the client the before-and-after, measured and dated.
No. Every content change is a hypothesis, and Phase 3 is the test, not a promise. Some changes land and some do not; the point is that, for the first time, the client can see which. The movement is measured, sourced and dated — evidence of lift on a topic, never a claim of causation.
The evidence-vs-estimate case, applied to a specific tool.
Read it →Which measurement to actually put in a client report.
Read it →How each citation is graded, so your report stands up.
Read it →Unsourced captures which AI assistants cite you, proves which crawlers really fetched your pages, and re-checks after you act — evidence, not a score.
© Unsourced — the evidence layer for AI search.