How we work · The delivery, in plain words

No secret sauce. Just work
most firms won’t do
properly.

Our methodology page explains how we measure. This page explains everything else — who does what, what happens each month, and the lines we won’t cross. If a vendor can’t describe their delivery this plainly, ask why.

01 · The runs

Real people on the answers. Real machinery around them.

When we say we run your 75 prompts across five engines, we mean a trained analyst asking the actual consumer products — the same ChatGPT, Perplexity, Copilot, Claude, and Google AI Overviews your buyers use. Not API shortcuts. That distinction matters more than it sounds: what an engine’s API returns is often not what a buyer sees in the product, and some answer surfaces have no API at all. Measuring the wrong surface precisely is still measuring the wrong thing.

Every observation is logged the moment it’s run: timestamp, engine, prompt, the full answer, every brand named, every source linked, and how you were characterized. That log — roughly 1,125 rows a month — ships to you with every report. The division of labor is deliberate: analysts capture what buyers actually see, because fidelity demands a human; software tallies, cross-checks, and trends what they capture, because consistency demands a machine. We automate the math, never the looking.

The judgment calls are governed by a written scoring rubric — what counts as a mention, how sentiment is bucketed, what happens when an engine answers a question with a question — so two analysts score the same answer the same way. We publish the rubric, because a score you can’t reproduce is a score you shouldn’t trust.

And the humans don’t work alone. Around the analyst runs sits a full technical stack — the instrument panel that extends what the humans see and double-checks what they record:

InstrumentWhat it does
SERP & AI Overview data feedsProgrammatic capture of Google's AI answer surfaces at query scale — broader coverage than any human could run, cross-referenced against the analyst observations.
Crawler & log diagnosticsWe watch your server logs and crawl your site the way GPTBot, ClaudeBot, and PerplexityBot do — verifying the engines can actually read what we've built, not assuming it.
Third-party visibility trackersWe run the same monitoring platforms other agencies resell as their whole product — as a cross-check on our own numbers, never as the product. When our ledger and a tracker disagree, we investigate why and tell you.
Entity & structured-data validationSchema validators, knowledge-graph lookups, and consistency scans across your web presence — confirming the machines resolve you as one coherent brand.
Source & citation crawlingAutomated monitoring of the third-party pages your category's answers cite — review platforms, comparison articles, communities — so we see movement in your source-of-truth map the week it happens, not at quarter's end.
Processing & cross-check pipelineEvery logged answer runs through automated extraction and tallying — brands, links, sentiment flags — reconciled against the analyst's scoring. Two independent reads of every observation.

The division of labor holds through all of it: machines for scale, verification, and arithmetic — humans for the buyer-visible answers no API faithfully reproduces. We automate the math, never the looking.

The obvious objections · Answered in public

“Manual? In 2026?” Yes. Here’s why that’s the strong position.

Three reasonable pushbacks we hear, taken seriously instead of dodged:

Objection 1

"Manual doesn't scale — you're a body shop."

You're not buying hours; you're buying a fixed-price program with a money-back guarantee attached. The careful human work is our cost to carry, not your invoice to audit. In this market, full automation isn't the upgrade — it's the corner-cut: automated trackers measure API surfaces your buyers never see. The expensive method is the credible one, and firms whose margins depend on not looking can't copy it.

Objection 2

"Humans make mistakes. Dashboards are consistent."

Dashboards feel consistent because you can't check their work. Our process is more verifiable, not less: a published scoring rubric so two analysts score identically, three-run replication against engine randomness, software cross-checks on every tally, and the raw log shipped with every report. Anyone on your team can audit any row. Try asking a dashboard for that.

Objection 3

"We could just do this ourselves."

You could — for a month. The value isn't running prompts once; it's holding the ruler still: the identical protocol, every month, scored consistently, trended against your consensus set — plus the remediation the measurement exists to steer. Bookkeeping is also possible in-house. Firms exist because consistent tedium, executed well, is a service. And if you try it yourself first, genuinely: you'll come back our best-educated client.

The expertise behind the discipline

We didn’t discover algorithms last quarter.

This firm was founded by a practitioner who spent a decade reverse-engineering ranking systems for a living — because that’s what serious SEO always was: forming hypotheses about an opaque algorithm, testing them against real results, and discarding what the data killed. When answer engines emerged, we brought that same discipline to the new black box: systematic observation of how these systems select sources, resolve brands, and build consensus — run on our own properties first, with our own money at stake, before we ever charged a client for the findings.

That history is why this page reads the way it does. We’ve watched every “algorithm-beating secret” of the last ten years die at the next update — and the practitioners who survived every one of those updates were the ones betting on fundamentals, measurement, and legitimacy. We’re not new to studying opaque systems. We’re new to an industry that finally makes showing your work a competitive advantage.

02 · The work between the runs

Four layers, labeled so you know what you’re paying for.

Layer 1 · Fundamentals · weeks

Open the doors

We make sure the engines can read you at all: AI crawler access (blocked more often than anyone expects, usually by a security default nobody chose), clean structured data, and a technically sound site. This is the unglamorous 80% — and if another agency already handles it well, we coordinate with them instead of billing you twice.

Layer 2 · Content · weeks to months

Write what answers quote

Engines lift direct claims, real numbers, and honest comparisons. We build the pages your buyers' questions imply — pricing with actual figures, "you vs. the alternatives" pages that genuinely compare, even "who this isn't for." Written by people, structured for extraction, approved by you.

Layer 3 · Identity · ongoing

Become one clear thing

Models recommend brands they can resolve confidently. We make your company read as one consistent entity everywhere it appears — your site, your profiles, the directories and references engines learn from — so every mention anywhere reinforces the same you.

Layer 4 · Authority · months, compounding

Earn the trusted rooms

Your baseline's source-of-truth map shows which third-party sources your category's answers actually draw from — review platforms, comparison articles, communities, trade press. We run the slow, legitimate campaign to get you into them: real reviews from real customers, expert commentary under real names, outreach to the writers who maintain the comparison pages. This layer moves consensus sets. It is also why the program is measured in quarters.

The lines we don't cross

Things we will not do, at any price

Some of what's sold as AI visibility is manipulation with an invoice. Our no-list, in writing:

  • §1No fake or incentivized-and-undisclosed reviews. Review campaigns we run go to your real customers, through the platforms' own rules.
  • §2No astroturfing. We don't pose as neutral voices in communities to praise you. When we participate anywhere on your behalf, it's under a real, disclosed identity.
  • §3No AI-article flooding. Publishing dozens of automated posts a month is the most commonly sold tactic in this industry and the fastest way to trade your domain's future for a vendor's monthly invoice. We build fewer, better, quotable pages.
  • §4No manipulation of the models themselves. No prompt-injection tricks, no hidden text aimed at crawlers. Beyond being wrong, it's brittle — one model update and the "win" becomes a liability with your name on it.

If a tactic only works while nobody notices, it isn't a strategy. It's a countdown.

03 · What we need from you

Your side of the ledger.

The program runs on a few things only you can provide. In month one: two hours with someone who talks to your buyers — sales calls, objection logs, lost-deal notes are where the real prompts come from — plus site access for the technical layer and sign-off on the prompt set that becomes your fixed ruler.

‍Ongoing: timely content approvals, introductions to happy customers when we run review campaigns, and a subject-matter voice (often fifteen minutes of an expert’s time) when commentary opportunities land. That’s the whole ask — and it’s why the guarantee’s third condition exists: the roadmap has to actually run to be guaranteed.

‍Every month, from us: the re-run report with your trend line, what we did, what moved, and what’s queued next. Fifteen minutes to read; one call a month if you want it; a board-ready deck each quarter.

‍The program & guarantee →