Our method

The Signal Stack.

A six-layer system for earning visibility in search and in AI answers. Each layer is a discipline on its own; together they are the difference between “we published content” and “the model recommends you.” Skip a layer and everything built on top of it is unstable. Most agencies sell layer four and call it GEO.

How we grade our own techniques.

The field is young enough that the gap between a proven technique and a bet is enormous. We grade each one before offering it, and we never sell an experiment as proven.

Proven

Rests on an established, documented mechanism. Only the execution needs verifying.

Directional

The logic holds and early results are strong, but the field is too young for guaranteed numbers.

Experimental

A working hypothesis, applied with full disclosure that it is exactly that.

Six layers, one system.

01

Access

Machines have to reach the content before anything else matters.

Search bots and training bots are not the same bot, and treating them as one is the most common technical mistake we find. OAI-SearchBot governs visibility inside ChatGPT Search; GPTBot trains models and does not affect citation. PerplexityBot, Claude-SearchBot, Googlebot, Bingbot — each is configured, verified, and reconciled against the CDN and firewall rules that block them more often than any client expects.

ProvenPer-crawler access audit against live responses
ProvenCDN and WAF rule reconciliation
DirectionalAgent-facing structured delivery, where rendering blocks access
02

Entity

The model has to know unambiguously who you are.

One unambiguous, repeated description of the company, anchored consistently across the Knowledge Graph, professional and industry directories, sameAs structured data, and a maintained llms.txt. This layer decides whether a model can form a correct opinion at all, before it decides whether to share it.

ProvenOrganization and Person schema with cross-linked @id
ProvensameAs corroboration across independent profiles
ProvenThird-party directory accuracy audit
Directionalllms.txt as a maintained surface
03

Architecture

The site has to be organised the way the category is actually searched.

A topic map built on real demand rather than assumption, in a hub-and-satellite structure: a small number of pillar pages, each surrounded by satellite content that earns the hub authority instead of competing with it. Marker-query collection defines the category vocabulary, entity-coverage checks grade each page against the full set of attributes the topic requires, and a coverage map shows the gaps while they are still cheap to fix.

ProvenHub-and-satellite taxonomy
ProvenEntity coverage scoring per page
ProvenCannibalisation audit and consolidation
04

Answer layer

Content engineered to be extracted and quoted, not merely read.

Every page opens with a direct answer in the first 50 to 80 words that a model can quote without inferring, and builds subheadings as the questions a person actually asks a machine. Every factual claim is grounded in a real source before the sentence is allowed to ship. No page goes out without review by a human who understands the subject.

ProvenShort-answer block and question-shaped headings
ProvenFAQ and structured markup on every commercial page
ProvenSource-grounded fact checking before publication
DirectionalCitation-density structuring
05

Authority economics

Authority is priced in money, not accumulated by accident.

Every inbound link is classified by anchor type and reconciled against the natural distribution for the category — an unnaturally clean exact-match share is one of the oldest documented penalty triggers in search. Any owned link infrastructure is costed against the price of a single earned placement before choosing between them. Most agencies bill link building as a fixed monthly line; we show the unit economics behind it.

ProvenAnchor distribution modelling against category norms
ProvenCost-of-ownership analysis before any infrastructure spend
ProvenEditorial and conference outreach
DirectionalDual-intent asset pairing on one topic
06

Proof and compounding

Visibility you cannot measure is visibility you cannot defend at budget time.

Every AI answer is scored on six dimensions: whether the brand is mentioned, where in the answer, in what tone, whether the description matches fact, who else stands beside it, and which page the answer leans on. The full chain is tracked end to end — model reads the page, model mentions the brand, a person follows the link, that becomes revenue — and every break in the chain gets its own diagnosis. "GEO does not work" is never an acceptable conclusion; the chain shows exactly where it broke.

ProvenMulti-run prompt baselines across all major systems
ProvenCitation gap analysis against named competitors
DirectionalCategory-volume tracking, not just share of voice
ExperimentalCross-model consensus mechanics

Every engine trusts differently.

Treating them as one search box is the second most common strategic mistake we see. Observed behaviour, tracked continuously.

ChatGPTCarries little native notion of source trust; leans heavily on a narrow publisher set for what enters the citation pool.
ClaudeScores and re-ranks roughly the top 300 organic sources per query before answering.
Gemini / AI OverviewsWeighs author and entity trust — E-E-A-T signals — close to a first-order factor.
Bing / CopilotA separate index. In some categories a large share of what ChatGPT cites traces back to it.
PerplexityCitation-dense by design; every cited URL is traceable and worth tracking on its own.

What we will not do.

Trust, once spent, does not come back at the same price.

  • We do not guarantee specific positions or citation frequency. The market decides that, not us.
  • We do not fabricate reviews or invent expert personas for trust signals, even where parts of the industry treat it as standard.
  • We do not manipulate behavioural signals or manufacture artificial traffic.
  • We do not attack competitors: no CTR manipulation, no campaigns against their assets.
  • We do not buy link infrastructure without costing it honestly against the alternative first.
  • Paid advertising is not our lane. We bring in the right partner when it genuinely helps.

Proof, not a highlight reel

natalchart.ai #1 in Google and Perplexity for its method.

A consumer AI product, web, eight languages, core product reachable without signup. The Signal Stack applied in full. Layers 1–3: a technical audit of 40+ live pages, with two silent risks caught before they cost visibility — contradicting facts between language versions of one page, and a rendering path that could quietly swap the homepage title. Layer 5: a managed outreach system with 380+ tracked touches under one rule, nothing sent twice, plus a reputation audit that caught four factual errors about the product on third-party AI directories and queued each for correction. Layer 6: the baseline showed the market already thinking in the product’s terms — its methodology core was already appearing in answers where its name was not.

The honest part: visibility on broad commercial queries is not yet dominant. That is the normal shape of a growing product that takes a narrow category first, then widens — shown as it is, not smoothed for a slide.

Common questions.

What is the Signal Stack?

The Signal Stack is CTO Monster’s six-layer method for earning visibility in Google, Bing and AI answers (ChatGPT, Perplexity, Gemini, Copilot, Claude). The layers are access, entity, architecture, answer layer, authority economics, and proof. Each builds on the one below it, and each technique is graded proven, directional or experimental before it is used.

What is the difference between SEO and GEO?

SEO earns organic positions in Google and Bing — a twenty-year-old discipline that still drives most first contacts. GEO (Generative Engine Optimization) earns mentions and citations inside AI answers from ChatGPT, Perplexity, Gemini and others — a discipline about two years old. Both run on the same foundation, which is why running them together usually costs 20–30% less than two separate contracts.

How do you measure AI visibility?

Every baseline is 20 to 50 buyer-shaped prompts, run several times a day across every major AI system and target region. Each answer is scored on whether the brand is mentioned, where, in what tone, whether the description is accurate, who else appears, and which page the answer leans on. One check proves nothing; appearing in 4 of 10 runs is 40% mention share, not the 0% or 100% a single search would show.

Do you guarantee rankings or citations?

No. We do not sell guaranteed positions or citation frequency — the market decides those. We sell the system, a measured baseline, and a weekly reconciliation against it. What we guarantee is that when something moves, the chain shows exactly where.

Where to start.

Starter

from $2,000/mo

Testing the category, building the base.

Layers 1–3 built, baseline AI-radar, entity signals live.

Signal Stack

from $8,000/mo

Established brand compounding authority.

All six layers, layer-4 content in production, continuous layer-6 monitoring.

Enterprise

after audit

Complex footprint, competitive category, multiple markets.

Dedicated citation infrastructure, competitive gap analysis, digital PR.

A standalone diagnostic — technical audit plus a baseline AI-radar across five systems — runs $2,500–$5,000 and can be booked before any further commitment.

See your six layers, free.

A 20-minute call: how the Signal Stack looks for your brand today, no commitment.

Book a free diagnostic call