EmarationThe scan finds it. The plan fixes it.
Emaration works to improve your AI visibility by running a five-layer audit, then fixing what it finds: structured data and llms.txt so engines can read you, answer-led content so you can become the answer, and GA4/GSC measurement so you can see the visits AI sends you. Every AI-assisted task runs through one audited pipeline — metered and disclosed — and is reviewed by a person before it ships.
Most agencies hide what their AI does. We do the opposite: we name the model we use (Anthropic's Claude — it's in our sub-processor register), we meter every call it makes — token counts and exact cost, attributed to the client and the feature it served — and we measure the answer engines your customers use from the outside, with the sample size printed on the report. Humans approve every artifact before it leaves the building.
The measurement is the work. The disclosure is the moat.
Start membership — $99/month →Or start free — $0, no cardRead on
Four rules that sit above the methodology.
The system below is how we do the work. These are the commitments that decide whether we ship it at all. They are written before the engagement so neither side has to relitigate them mid-flight.
01: A human in the loop, every time
AI is a force multiplier on expert judgment, not a substitute for it. We deploy AI where it deepens insight, accelerates throughput, or removes ambiguity. We put a qualified human between the model and the client every time the deliverable matters. We say this in our pricing, our deliverables, and our hiring.
02: Full AI disclosure, every time
We tell you what AI did. We tell you what humans did. Every time. No black-box deliverables. No "our proprietary intelligence" hand-waving. Every audit, every report, every recommendation carries a plain-English note on what was model-generated, what a person wrote, and who wrote it.
03: Accessibility is a default, not a feature
WCAG-AA is a floor, not a ceiling. Co-founder Jordan Williams lives with vision and hearing loss. That is a structural commitment baked into the products we build, the websites we ship, and the content we write. Color choices, contrast, keyboard access, screen-reader semantics, and a read-aloud control on every page are non-negotiable.
04: The mission funds the mission
10% of net profit flows to EOCS. Every paid engagement contributes toEmaration's Outreach and Community Support — our community-focused assistive-technology initiative, focused on accessibility and Deaf-Blind community needs. You get an EOCS receipt with your audit, and the math goes public quarterly once there's a first quarter of profit to report. We will not grow the agency in a way that starves the initiative.
Three of these four are policies most agencies cannot adopt without rebuilding their operating model. That's the point: a commitment a competitor can't match is worth more than a feature.
If you're paying an agency in 2026, you're paying for one of three things.
- A human doing the work by hand: slow, expensive, doesn't scale.
- An agency that quietly pipes your account through an AI vendor it never names: cheap for them, and you can't audit what you can't see.
- An agency that names its AI, meters every call, measures the answer engines from the outside, and puts a human's name on every deliverable — us.
Option three isn't a secret model stack — it's receipts. When you can see what the AI did, what it cost, and who reviewed it, you can judge the work instead of trusting the pitch.
Related reading: what multi-AI orchestration actually means for your campaign, and why most agencies can't prove ROI.
One audited AI pipeline: every call has a receipt.
Our generation-side AI — audit narratives, content drafts, fix artifacts — runs through a single call path to one named vendor: Anthropic's Claude, listed in oursub-processor register. One path means nothing is off the books: every call is recorded with its token counts and exact cost, attributed to the client and the feature it served. Failed calls are recorded too, with the reason, so failure volume is as visible as success.
We don't sell a secret model stack, and we won't pretend to. What we sell is the audit trail: ask what AI touched your work, what it cost, and who reviewed it, and we answer from records, not memory.
Scheduled monitoring: rule-based, no model in the loop.
After a report lands, scheduled checks re-read your history and flag what actually moved: a score drop past a threshold, a critical finding your previous report didn't have, a Core Web Vitals metric falling out of the "good" band when your Google data is connected.
An alert says "something moved — look." It is not a deliverable and it approves nothing: every client-facing artifact still passes the human gates below.
AI answers move. We measure them like it.
Every model from every vendor changes over time, and the same engine can answer the same question two different ways in the same afternoon. We don't claim to control that. We claim to measure it honestly:
- Repeated trials per prompt, per engine, as the run's budget allows — and every rate names its trial count.
- The sample size on every rate: findings read "cited in X% of N model answers," never a bare percentage.
- Pinned measurement settings, so a repeated reading doesn't move on sampling randomness alone.
- Failed trials disclosed as failed trials, never papered over.
The honest defense against a moving model isn't a secret dashboard — it's disclosed measurement plus a human reading the output. When an engine's answers about you change, your visibility history shows it, and a person — not a threshold — decides what to do about it.
The five-layer scoring rubric — one number that can't lie.
The audit reviews five layers on every page it crawls — technical and on-page SEO, content quality, business-presence signals, answer-engine (AEO) readiness, and an accessibility read for common WCAG Level A barriers — plus two site-wide checks: how the crawl itself went, and whether your robots.txt lets the AI crawlers in at all. It turns the findings into one deterministic 0–100 score with an A–F grade. Deterministic means exactly that: the same findings always produce the same number, with no model in the scoring path. Here is the whole computation, in plain English:
- Count penalty. The score starts at 100. Every critical finding subtracts 8 points, every warning 3, every informational note 1 (floored at zero).
- The anchor-cap rule. A critical finding in a foundational category thencaps the score outright, no matter how much else passes: a critical crawl failure caps it at 40, an indexability failure at 50, an accessibility failure at 55, a performance failure at 60, a schema failure at 65 — and when several apply, the lowest cap wins. A money page Google can't index is unrankable; averaging that against a hundred passing checks would produce a comfortable, false number. The number never lies by averaging.
- Grade. 90 and up is an A, 80+ a B, 70+ a C, 60+ a D, anything below an F.
Every finding behind the score cites the specific data point that produced it (the citation rule under "Human in the loop", below), and each carries a provenance stamp — what was measured live, what came from a cached source, and what is a static heuristic.
How the AI-visibility probes work.
We ask the engines your customers use — ChatGPT (OpenAI), Claude (Anthropic), Perplexity, Gemini (Google), and Grok (xAI) — the buyer questions from your intake, and we record what actually comes back: whether the answer cited you, whether it cited a competitor instead, and which competitor keeps winning the prompts you should own.
- Trial counts are disclosed on reports. AI answers vary run to run, so each prompt is probed in as many trials per engine as the run's budget allows, and every rate on your report names its sample size — findings read "Across N model trials …" or "cited in X% of N model answers," never a bare percentage.
- No fabrication. Probe rows reflect exactly what the engines returned. A failed trial is reported as a failed trial (and how many failed), never papered over.
- Read-only and budget-capped. The probes only ask and measure — they never write anywhere — and every run stops at a hard spend ceiling.
Our analytics posture: we count events. We don't track you.
We count events. We don't track you. Analytics stores no tracking cookies, no fingerprints, and no IP addresses — event counts only. Abuse rate-limiting on the instant audit uses salted one-way, per-day hashes in place of addresses: an address is hashed before it can reach storage, is never stored raw, and the hash changes every day. Measurement here is a tiny first-party beacon that counts a handful of page steps: the home page, the audit page, and the welcome page after you sign up. Alongside it, the server tallies a few funnel events — an instant scan starting, that scan finishing, a signup landing. What we keep is a count per event per day, nothing else: no pathnames, no per-visitor records, nothing that says who.
- Opt-outs are honored. The beacon respects Do-Not-Track and Global Privacy Control, and visitors without JavaScript never fire one — so beacon counts are observed events only, and we never extrapolate them. The server-side tallies carry no identifiers at all: they record that an event happened, never who did it.
- Nothing third-party. No analytics vendors, no advertising pixels, no cross-site identifiers. The first-party browser storage this site does set is listed by name, key for key, on the cookie policy, and none of it says who you are.
Five stages. Humans at every gate.
The pipeline runs inside our five-stage client engagement model. Here's how AI plugs into each stage.
Discover
The five-layer audit crawls and scores your site; the visibility probe asks the answer engines about you. A human reads every finding and owns the priority list.
Define
Who your best customers are, what they ask, and where you sit in your market. AI drafts from what you already publish; a human owns scope, money, and expectations.
Design
How the site is organized, what content to write, which channels to use, what structured data to add, and how a visitor becomes a customer. AI-drafted, human-led.
Deploy
AI-drafted artifacts plus vetted vendors execute. Every deliverable carries a human review stamp before it ships.
Defend
Analytics pipeline plus scheduled monitoring alerts. The methodology keeps working after the launch.
See the five stages run end-to-end in our illustrative case studies.
Human in the loop: by rule, not by intention.
Four non-negotiable gates sit on top of the pipeline. None of them are theater.
Pre-send review
Every deliverable we author — audits, reports, roadmaps — is read end-to-end by a founder before it goes out. No exceptions, no automation, no "the AI said it was fine." Content you draft and approve for your own site through the member portal is yours to approve — clearly labeled as AI-drafted, published only on your say-so. Where a lane runs an editorial lint, the rules being applied are ours, not a brand guide of yours we hold; the lanes that run none say so on the card you approve.
Voice and brand check
We have an explicit list of words and patterns we won't ship. Our brand-voice rules are enforced at the output template, not by trust.
AI-image disclosure
AI-generated images are marked on the image itself — the exact wording is in the note on visuals below. This gate we keep by hand: nothing we run opens an image file to check it for us.
Citation enforcement
Every finding in an audit references a specific data point. No data point, no claim.
This is the reason the work can be handed to lawyers, accessibility consultants, and boards without flinching.
A category-of-one posture you cannot buy off a shelf.
SaaS marketing tools hand you a dashboard and leave the judgment to you. Most agencies hand you the judgment and hide the machinery.
We hand you both: a deterministic score that can't lie by averaging, answer-engine measurement with the sample size printed on it, an AI pipeline where every call has a receipt, and a human gate no deliverable can skip.
The disclosure is the structural answer. The measurement is the renewable one. The human review is the safety.
Your onboarding audit is where you see this at work, and it's waived with membership.
The scan finds it. The plan fixes it.
Your onboarding audit puts the methodology on your site: what's leaking, ranked by severity, in plain English, with a scoped remediation plan a human wrote — not a sales pitch. Onboarding (our automated five-layer audit and setup) has a list price of $1,000, and it's waived with membership.