The correlation is the differentiator.

Every AI visibility tool reads one number and calls it a strategy. That is a guess. We triangulate three sources, pressure-test the findings against each other, and refuse to recommend anything the data does not corroborate. This is the full methodology. Documented, versioned, and running against every engagement.

01 · Why correlate

Single signals lie.

Rankings alone tell you where you stand in Google. They tell you nothing about whether ChatGPT names you. An AI mention scan alone tells you whether you were named in one query on one day. It tells you nothing about whether the fix will lift search alongside it. A schema audit tells you what is broken structurally. It tells you nothing about whether fixing it moves the numbers customers actually care about.

Each signal is a partial view. Correlation is evidence. The industry sells single-source dashboards because they are cheaper to run and easier to render as reports. That does not make them true.

We correlate three sources on every engagement. AI answers. Search performance. Site structure. A finding only becomes a recommendation when at least two of the three agree.

02 · The three sources

Three measurements. In parallel.

Every engagement runs three parallel scans against the same query set. Each source has its own tooling, its own outputs, and its own failure modes. Cross-referencing them is the discipline.

Source A

AI answers.

Hundreds of query variants per category, run against ChatGPT, Claude, Google AI Overviews, and emerging surfaces. Repeated across days to detect variance. Verbatim capture of whether the assistant named a specific business, which one, and what it cited to build the answer.

What it surfacesAI mention rate
Source-graph capture
Competitor entity claims
Structural retrieval gaps

Source B

Search performance.

Google Search Console pulled against the same query set. Impressions, clicks, average position, and click-through rate on the categories the AI answers scan measures. Bing Webmaster Tools where available. What is actually driving traffic today versus what is invisible.

What it surfacesQuery impressions
Avg search position
Click-through rate
Page-two traps

Source C

Site structure.

Rendered-markup audit of every retrievable page. Schema validity per page, canonical hygiene, NAP consistency, heading hierarchy, and robots access per crawler. Not the pretty page. What the crawler sees when it actually reads the site.

What it surfacesSchema validity
Canonical drift
Entity claim gaps
Crawler access map

The bar we hold

A finding only becomes a recommendation when at least two of three sources agree. Single-source recommendations are categorically refused.

03 · The loop

Baseline. Correlate. Ship. Measure. Repeat.

The engagement is not a report. It is a discipline that compounds. Every cycle sharpens the model. Nothing carries forward on hope.

01
Baseline
Where you stand now, across every signal. AI mention rate per query per engine. GSC performance. Rendered markup and schema audit. Nothing extrapolated. Nothing inferred. Just what is.
Day 0
02
Correlate
Which signals point to which decisions. A missing entity signal that shows up in the AI source graph and the schema audit is a fix. A gap that shows up in only one is a note. Every finding gets mapped to the sources that agreed on it.
Day 1 to 3
03
Ship
The pages, fixes, and entities that move the number. Your team, or ours. The spec is the same either way. Every fix carries an expected lift on both AI citation and search rank, sourced from the correlation that surfaced it.
Day 3 to 28
04
Measure
Re-scan on the same query set, against the same engines, on the same three sources. Cache-busted. Delta reports published per signal, per surface, per engine. What moved. What did not. What we got wrong.
Day 30
05
Repeat
The next decision list is smarter than the last. What worked stays. What did not gets rewritten. If a category is structurally unwinnable, we tell you that too, and where the ceiling is. Guessing does not enter the next cycle.
Cadence

04 · What we ship

A build spec. Not an audit report.

The deliverable at the end of a scan is a documented brief. Content strings, schema graphs, page structure, engine-specific trust-layer fixes. Mapped to the exact findings that call for them. Scoped against measurable expected lift on both AI citation and search rank. Something your team can ship against on Monday morning.

What we ship

Build spec

Per-fix scoping, sourced back to the correlation
Content strings ready to hand to a writer
Schema graphs ready to hand to a developer
Expected lift documented per fix
Refuse list of findings we would not act on
Re-scan on cadence to close the loop

What we do not ship

Audit report

PDFs of what is broken with no path to fix
Findings without the source that surfaced them
Recommendations built to justify the retainer
Metrics scored on an internal proxy
Guesses dressed as strategy
Anything you cannot verify yourself

05 · What we refuse

The lines we will not cross.

The discipline works because we hold it. These are the rules we tell every client on the first call.

×
Single-source recommendations. If only one signal supports a fix, it becomes a note in the report, not a recommendation. Guessing does not enter the spec.
×
Speculative AI-only bets. We only recommend structural fixes that move both AI citation and search rank. If we cannot show the correlation, we do not ship the fix.
×
Retainers without receipts. Every re-scan publishes deltas per signal, per surface, per engine. If the numbers did not move, we tell you. If a category is unwinnable, we tell you where the ceiling is.
×
Content dishonest to the business. No expansion into adjacent verticals you do not actually serve. No credentials you do not have. No claims that would not survive scrutiny from a real customer.
×
Naming clients without permission. Case entries publish anonymized by default. Named data ships only when the client consents to it, in writing, on the specific numbers.

06 · Tools

We built our own.

The scanner is ours. We wrote it, we break it, we fix it. Named Atlas internally. It runs hundreds of variants per category query against ChatGPT, Claude, and Google AI Overviews. Records every named business. Maps the source graph the assistants read to build the answer. Retains dated evidence for every scan so re-runs are strictly comparable.

Correlation is done against Search Console (or Bing Webmaster where available) and against a rendered-markup audit that runs on every scan cycle. The tooling is not the product. The tooling is what makes the discipline possible. The product is what we do with what it finds.

07 · Version notes

The methodology evolves.

Every scan carries the version it was run under. When the discipline changes, the version does too. Prior engagements are traceable back to the version they were measured against. Nothing gets rewritten silently.

v2.4.1
Added source-graph capture across every AI scan, so we can map exactly which pages the assistant read to build its answer.Released 08.15.26 · Current
v2.4.0
Expanded query variant sweep from ~100 per category to hundreds. Added variance detection across scan days.Released 06.02.26
v2.3.0
Formalized the two-of-three correlation rule. Made single-source recommendations refuseable at the tool level, not just editorially.Released 03.10.26
v2.0.0
Structural audit added as the third source. First engagements ran with the full three-source methodology.Released 11.14.25

See it run on your business.

Your first scan across all three sources. AI answers, search performance, site structure. Cross-referenced. Delivered as a build spec your team can ship against. Every finding traced back to the source that surfaced it. Every recommendation carrying the correlation that earned it.

Book a visibility check →

Or write directly to [email protected]