EP 279- Beyond the Link: How Citation Labs Rebuilt Its Agency for AI Visibility (Part 2)

Garrett French (CEO), Valerie Cecil (operations) and James Wirth (Sr. Director, Strategy and Growth Marketing) of Citation Labs return for Part 2 of their Near Memo conversation with Greg Sterling and Mike Blumenthal. Part 1 laid out the six layers of AI visibility; this one is about what running an agency against that model actually looks like: what gets built, by whom, how it's checked and how it's billed.

AI visibility is a content-supply problem — models cite whatever well-structured, third-party evidence fills the gaps in their query fan-outs — so an agency that once sold links now wins by manufacturing that missing evidence, proving its impact the way paid search proves return on ad spend, and pricing the work by the gap it closes rather than by the link.

Citation Labs' Six Layers of AI Visibility

James recaps the framework from Part 1 (01:44–05:20): each layer is a step in how a model turns a prompt into an answer, and a gap at any step is something the team can work on. Garrett groups the first two layers as the eligibility zone and the last two as the preference zone.

Takeaways

• Gap-filling content gets cited within days. When a satellite page covers detail that was missing from the answer space, the team sees it cited "as soon as we start measuring," even without link building behind it. Models are "hungry" for accurate, well-structured content that answers the prompt faster. (09:28–10:07)

• Models ignore the same facts on the brand's own site. Content that sits on the client's domain gets passed over, while the identical content on a third-party page gets used. Greg's read: brand-sourced claims are less trusted in an evaluation context. (10:07–10:38)

• Satellite sites publish what clients won't. One client rejected on-site content because it named competitors. Citation Labs runs its own comparison sites that review the client and every competitor "fairly," pro and con. Whether a site discloses the client's ownership is set by the client. (07:13–11:40)

• "Query fan-out placements" are dry by design. Valerie's team writes narrow, detail-heavy pieces aimed at a single prompt set, then echoes them across several surfaces (websites, LinkedIn, chambers of commerce). They read flat to a human because they exist to fill what the model found missing. (07:13–08:56)

• The old link builder can't call it link building anymore. In the last 18 months, Garrett says, the work stopped being link building, and James won't let him call it off-site either: it's a content agency that now rewrites client sites for LLM structure too. One client barred the words "link building" from its 2027 statement of work. (13:05–14:48, 47:40–49:36)

• The next frontier is "brain crawlers." Citation Labs is building subject-matter-expert interview processes, using query fan-outs to decide what to ask. The goal is extracting expertise that isn't on the web yet, the evidence a model is asking for and can't find. (14:48–16:10)

• AI made the work bigger, not easier. Some satellite sites run to thousands of pages, so volume went up rather than effort down. Every piece still gets human review, and Valerie pushes back on AI output nobody reads, like "forty page guides" shared unread. (21:15–23:18)

• Claude is the analyst's workbench. James runs each six-layer "visibility probe" as a Claude skill fed by MCP connectors. They pull DataForSEO data through the Link Launch link-gap tool (rebuilt on Replit), Xofu visibility data and internal sources. (24:46–26:32)

• Assume the model will invent, then check it. James challenges findings in a second frontier model, checks each tool's own front end, cross-checks against Semrush or Ahrefs, and increasingly uses Search Console and GA4 as the source of truth. Time saved on spreadsheets now goes to validation. (26:58–28:35)

• Borrow the paid team's numbers to model SEO ROI. Paid search has return on ad spend; SEO rarely builds the equivalent. James gets value per conversion and customer lifetime value from the client's paid team to calculate a simple ROI. Mike notes many small businesses can't name the value of a client at all. (32:14–33:40)

• Control pages are what earned the C-suite's trust. Citation Labs showed what happened to pages it built links to, similar pages it didn't, and competitor pages. That comparison justified the spend for years and now carries into AI visibility. In one pilot against four other agencies on the same budget, it was the only one to hit the KPIs, and James joined from the client side. (34:26–35:46, 38:49–39:43)

• The new unit of work replaces price per link. The Xofu team estimates units of work per descriptor or prompt cluster, and the client chooses which to fund. Each unit (a fan-out placement, a satellite page, a traditional editorial link) is held near the old per-link price, with a separate tracking fee when prompt sets get large (Garrett wants 4,000). (40:41–43:12)

Related Reading

• EP 277 — Correcting the Machine: Citation Labs on the Six Layers of AI Visibility (Part 1) — Part 1, where James walks through each layer and its lever.
• Three Brands Built for the Next Era of Search Visibility — how Citation Labs, Xofu and ZipSprout fit together.
• Xofu — the AI-answer visibility tracker behind the six-layer probes.
• How Xofu Finds the AI Visibility Gaps Killing Your Brand Mentions — Citation Labs Podcast on the gap-finding method.
• How Many Links Does My Page Need? Link Launch by Citation Labs — the link-gap analysis tool James rebuilt on Replit.
• How to Track and Fix BOFU Visibility in LLMs — the bottom-of-funnel prompt work Valerie references.
• ZipSprout — Citation Labs' local sponsorship platform for smaller clients.
• EP 276 — GEO Is CRO: Ross Hudgens on Chasing Influence, Not Citations — Siege Media's finding that versus pages predict AI traffic, the same comparison content Citation Labs puts on satellite sites (confirm final URL/slug before publishing).
• EP 278 — Track What You Can, Let Go of the Rest: Dana DiTomaso on Attribution After GA4 — the measurement side of the impact-reporting discussion (confirm final URL/slug before publishing).

Practitioner Notes

• Satellite sites need a disclosure line before local clients copy them. Undisclosed "fair comparison" sites owned by one competitor sit close to the patterns Near Media flags in its Maps spam and review-network audits, where shared ownership is the tell. For local clients, recommend disclosed ownership or editorial placements on independent local publishers instead.
• Fair comparison content fixes the refusal problem from EP 276. Ross Hudgens found versus pages correlate best with AI traffic, and Citation Labs' client refused them on its own domain. A third-party comparison page is the workaround, and it matches James's finding that models favor third-party content over identical brand-site copy.
• Paid-team ROI inputs fit Near Media's lead-source attribution work. James's value-per-conversion and lifetime-value inputs are the missing denominator in monthly lead-source reports for law firms. Pairing signed-case value with self-reported source turns an attribution report into an ROI report.
• ZipSprout sponsorships double as local brand mentions. Community sponsorships produce the off-site corroboration the brand series argued AI answers depend on. They are a local-scale version of the third-party evidence Citation Labs builds nationally.

Concepts

• Six layers of AI visibility — ranked, retrieved, cited, mentioned, recommended and framing: the steps from a model's search results to how it describes a brand.
• Eligibility zone / preference zone — Garrett's grouping: ranked and retrieved make a brand eligible; recommended and framing decide whether the model prefers it.
• Query fan-outs (QFOs) — the sub-queries a model runs to ground an answer, which reveal what information it is looking for.
• Query fan-out placement — a narrow, detail-heavy piece of content written to fill a specific gap in a prompt set's fan-outs.
• Satellite site — a Citation Labs-run site that publishes factual comparison content about a client and its competitors, with ownership disclosed or not as the client chooses.
• Unbranded editorial citation (UEC) — Citation Labs' term for its traditional editorial link building.
• Visibility probe — James's six-layer gap analysis for a prospect or client, run in Claude over MCP-connected data.
• Brain crawlers — interview processes that extract a subject-matter expert's knowledge so it can be published where models will find it.
• Unit of work — the billable replacement for the link: one piece of work aimed at a descriptor or prompt cluster, priced near the old per-link rate.
• VizFam — Garrett's name for the client-side stakeholders Citation Labs works to get promoted.

Pick your starting point:

00:00 Intro: Citation Labs, Part 2
01:44 The six layers of AI visibility
05:51 What the framework changed in operations
07:13 Query fan-out placements
08:56 Satellite sites and comparison content
10:07 Why models ignore the brand's own site
13:05 "We can't say we're just doing link building"
15:32 Building "brain crawlers"
16:15 How Citation Labs is structured
20:49 How the agency uses AI internally
24:46 Claude, MCPs and Link Launch
26:32 How to reality-check AI analysis
28:35 ZipSprout and small clients
32:14 Modeling SEO ROI like paid search
34:26 Proving impact to the C-suite
39:45 From price-per-link to units of work
44:10 Lessons from 24 months of change

Full Transcript --> on-line