Discipline 05 — Distribution

When Someone Asks AI Who to Trust, Your Name Appears.

We structure your presence so ChatGPT, Gemini, Perplexity, and Claude cite you as the authority in your field. Built from structured data, not prompt tricks.

Six mandates per quarter. Senior attention only. Complete discretion.

Request Digital Identity Audit arrow_forward What It Is arrow_downward

One Answer. No Page Two.

The answer names three to five people. The rest do not exist for that question.

language models measured monthly, 10 runs per query

25–32

queries built and signed off before work begins

60%

of runs across four retrieval models within 90 days

01 The Definition

What Is AI Recommendation Infrastructure?

AI Recommendation Infrastructure is the structuring of a digital presence so language models find, verify, and cite a name when someone asks who to trust in a field. It is built from structured data, entity signals, and citation architecture. It is not SEO. It is not prompt hacking. It is making a person legible to the systems that now answer most research questions.

psychology

Legible to Models

Structured so language models can find, verify, and cite.

hub

Not a Trick

A source the model can quote. Not manipulation.

Built From Sources

The only way to be cited is to become a source worth citing.

02 The Five Behaviours

What AI Does With a Name.

Language models treat names in one of five ways. Which one depends entirely on what they can find.

verified

Behaviour 01

Names the person

Cited as authority.

volume_off

Behaviour 02

Stays silent

Names no one, or names only others.

auto_fix_high

Behaviour 03

Invents facts

Hallucinates a role, biography, or controversy.

shuffle

Behaviour 04

Confuses the person

Attributes another person's work.

warning

Behaviour 05 · Worst

Recommends a competitor

Actively names someone else as the authority in the field.

The first is the goal. The other four are liabilities. The worst outcome is not silence. It's active competition.

03 The Cost of Absence

The Invisible Loss

A question gets asked. The AI names five people. The name is not among them.

A media interview goes to someone else A keynote invitation goes to someone else A deal introduction goes to someone else A panel seat goes to someone else

None of these are visible. All of them happen weekly for people whose AI presence is unmanaged.

You cannot count what you were never considered for. You can only fix the input that excluded you.

04 The Self-Test

Test Your Own Situation in Five Minutes.

Run these five questions. Note which models name you, which stay silent, and which invent.

"Who are the most credible [your field] experts?"

Ask on ChatGPT

"Who should I contact about [your domain]?"

Ask on Perplexity

"Who leads in [your industry]?"

Ask on Gemini

"Summarise [your name]'s background."

Ask on Claude

"[your name]" — check if an AI Overview appears

Search on Google

The situation is strong.

1–2

There is a gap.

Or it invents facts — this is the work.

05 The Distinction

How Is This Different From SEO?

SEO ranks pages in search results. AI Recommendation Infrastructure determines whether a model names the person in an answer.

SEO

Rank pages in search results.

target

Objective

Rank pages in search results.

web

Output

Ten blue links.

link

Mechanism

Keywords, links, on-page signals.

visibility

Visibility

Google Search Console.

schedule

Time to result

6–12 months.

arrow_forward

AI Recommendation

Be named in AI answers.

psychology

Objective

Be named in AI answers.

chat

Output

A single named answer.

schema

Mechanism

Structured data, entity signals, citation architecture.

query_stats

Visibility

Direct query testing across four models.

bolt

Time to result

60–90 days on retrieval models.

SEO feeds AI. AI decides who gets named. The two must be built together.

06 Measurement Methodology

How Do You Actually Measure AI Visibility?

Language models are non-deterministic. The same question asked twice gets two different answers. This must be measured, not assumed.

Control Center

Non-determinism acknowledged. Measured as distribution.

Every query is run 10 times per model, across four models, in a single measurement window. Reported as a distribution — how often the name appeared, in what position, with what framing.

Live · Monthly cadence

Models Tested

psychology

4models infoChatGPT · Gemini · Perplexity · Claude

Runs Per Query

repeat

10runs infoPer model, per query

Query Set

list_alt

25–32queries infoBuilt & signed off in audit

Data Points

data_usage

1,000+per cycle infoRaw outputs saved

Query set composition

Signed off

Standard queries

Field-specific queries

8–15

Competitive queries

Model versions recorded

Yes

Raw outputs provided

Yes

Monthly cadence

Yes

What's reported

Monthly

percent

Presence rate — how often the name appeared

format_list_numbered

Position — first, second, mentioned, absent

fact_check

Accuracy — facts the model reports

groups

Competitors — named in the same queries

trending_up

Month-over-month — changes tracked

memory

Model versions — recorded and reported

descriptionWritten monthly report

find_in_pageRaw outputs saved

callQuarterly review call

07 The Guarantee

05 GATE

Gate 05 · Milestone Guarantee

Within 90 days of launch, the name appears in at least 60% of runs across all four retrieval models for the agreed query set.

If it does not, we keep working at no fee until it does. Training-data models are measured separately. The timeline is controlled by the model provider, not us. Reported transparently.

Request Audit arrow_forward

08 The Model Split

Retrieval Models and Training Models Are Not the Same.

ChatGPT, Gemini, Perplexity, and Claude behave completely differently. Treating them as one thing reads as superficial.

Retrieval Models

Live search every query.

Examples

Perplexity, ChatGPT with search, Gemini with grounding.

bolt

How it answers

Live search every query.

schedule

Time to reflect new sources

Days to weeks.

verified_user

Guarantee applies

Yes — Gate 05 covers retrieval models.

compare_arrows

Training-Data Models

Trained data only.

storage

Examples

Claude, ChatGPT without search.

hourglass_bottom

How it answers

Training data only. Months or years out of date.

update

Time to reflect new sources

Next training cycle. 6–18 months.

block

Guarantee applies

No — provider controls the timeline. Measured separately.

Retrieval models shift in weeks. Training models shift on the model's schedule, not ours.

09 The Correction Process

What If AI Invents Facts or Says Something Wrong?

When a model has nothing verified to cite, it invents. When it has contradictory sources, it chooses one. When it has partial information, it fills the gaps.

The Four Failure Modes

auto_fix_high

Hallucination

AI invents a role or event. Verified sources reduce it.

content_copy

Contradiction

AI repeats a fragmented claim. One controlled narrative reduces it.

shuffle

Confusion

AI attributes another person's work. Distinct entity signals separate the name.

volume_off

Silence

AI names no one. Structured data gives the model something to retrieve and cite.

The Honest Position

Structured sources reduce hallucination. They do not eliminate it.

Models still confabulate. The frequency drops. It does not reach zero.

Less invention. More citation. Not perfect.

The Correction Process

Audit

Documents exactly what the model says, with raw outputs.

Source correction

Verified sources published that contradict the wrong claim.

Reinforcement

Corroborating citations built from credible third parties.

Retesting

Six components. Delivered as one system. Each layer builds on the one before it.

01Entity

Entity Architecture

A distinct, verifiable entity the model recognises. Name, role, domain, and relationships defined across every structured surface.

checkNamed Entity + Relationships

02Data

Structured Data Layer

Schema markup on every owned surface. The universal language language models read to understand who a person is.

checkSchema · SameAs · Person Markup

03Citations

Citation Building

Third-party sources that reference the name against a topic. The corroboration that makes a claim credible to a model.

checkThird-Party Sources

04Narrative

Narrative Consolidation

One consistent story across every surface the model reads. Fragmentation is a signal to a model. Consistency is a stronger one.

checkOne Controlled Narrative

05Testing

Model Query Testing

Monthly testing across four models, 10 runs per query. Reported as a distribution. Raw outputs saved.

check1,000+ Data Points / Month

06Recalibrate

Recalibration

Adjustments as models retrain and competitors shift. The entity stays. The strategy evolves with the platforms.

checkQuarterly Recalibration

12 The Mechanism

How Does Structured Data Actually Cause Citation?

Not a trick. A source. Here's the mechanism in five steps.

The model needs a source it trusts.

It draws from indexed, verifiable, corroborated information.

Structured data tells the model who the person is.

Schema markup defines role, domain, and entity relationships.

Third-party citations confirm the entity.

When credible sources reference the person against a topic, the model treats the association as reliable.

Consistency across surfaces increases weight.

One source is weak. Fifteen consistent sources is a signal.

The model cites the person because the model has a source it can quote.

Not a trick. A source.

13 The Timeline

What Happens Week by Week.

Six phases. Retrieval models shift in weeks. Training models shift on the model's schedule. Both are measured.

Wk 1–3Diagnostic

Phase 01 — Diagnostic

Map what exists.

Baseline testing across four models. Citation audit. Entity map. The query set is built and signed off before any work begins.

searchBaseline testing list_altQuery set signed off descriptionWritten report

Wk 3–5Foundation

Phase 02 — Foundation

Build the layer everything else sits on.

Structured data. Entity architecture. Narrative consolidation. Every structural element a model uses to understand who a person is.

schemaStructured data hubEntity architecture mergeNarrative consolidated

Wk 5–9Citations

Phase 03 — Citation Building

Publish and index third-party sources.

Third-party sources published and indexed. Every source that references the name against a topic is a corroboration the model can use.

publishSources published linkIndexed verifiedCorroborated

Wk 6–10Retrieval

Phase 04 — Retrieval Shift

Retrieval models begin citing.

Perplexity, ChatGPT-with-search, and Gemini-with-grounding begin citing the name. First controlled results appear in AI answers.

boltRetrieval shift psychologyFirst citations trending_upPresence rising

Wk 10–12Verify

Phase 05 — Verification

Measure full control. Sign off Gate 05.

Full measurement across all four models. Every query accounted for. Gate 05 signed off — or we keep working at no fee until it is.

verified_user60% measured descriptionFull report flagGate 05 signed

6–18 moTraining

Phase 06 — Training-Data Shift

Training models catch up on their next cycle.

Claude and ChatGPT-without-search catch up on their next training cycle. Timeline controlled by the model provider, not us. Reported transparently.

storageTraining cycle update6–18 months visibilityReported separately

14 Two Honest Realities

movie

Synthetic Media

AI can generate fake video, audio, and images of real people.

Deepfakes are now cheap and common. The question is what can be done about it — and what cannot.

check_circle Monitor with the same 24-hour alert system

check_circle Prepare a written response protocol

check_circle Publish verified video and audio so models have a trusted source

check_circle Coordinate with legal on takedown where platform policies permit

We cannot prevent creation. Remove from every platform. Guarantee no synthetic media will ever exist.

visibility_off

The Black Box

We cannot see inside language models. We infer behaviour from outputs.

check We can test the outputs. We cannot see the weights.

check We can measure change. We cannot explain every variance.

check We can build sources the model trusts. We cannot force the model to cite them.

Test rigorously. Report distribution. Show raw outputs. Change one input at a time. Measure again.

15 Rented Land

What If the Platform Changes Its Rules?

OpenAI, Google, and Anthropic control the platforms. They can change citation behaviour overnight. No contract exists. No SLA exists. The system is designed to survive it.

hub

Entity architecture

Models need to know who someone is. That never changes.

schema

Structured data

Schema markup remains the universal language.

link

Citation sources

Third-party references remain citable through any model.

bookmark

Named frameworks

Original frameworks remain attributable to their creator.

The platforms change. The entity, the data, and the frameworks don't.

16 Three Starting Points

Where You Start Changes the Sequence.

library_books

Scenario 01

Existing assets

A website, media mentions, or published work already exist.

check_circle Kept where they support the entity
check_circle Structured where they are unstructured
check_circle Aligned to one narrative in all cases
check_circle Cited where they are already trusted

add_box

Scenario 02

Nothing exists

The hardest case for AI — and the most fixable. Models have three options: silence, someone else, or invention.

check_circle Foundational entity created and verified
check_circle Structured data published across owned surfaces
check_circle Named frameworks give the model something to cite

Timeline is longer than a search-only build. The outcome is the same.

fingerprint

Scenario 03

Common name

The name matches others. The audit identifies every same-name entity across all four models.

check_circle Attach the name to a specific domain, role, and frameworks
check_circle Distinct entity signals separate the name from others
check_circle Named frameworks give the model a distinguishing signal

Scope increases. Timeline holds.

17 The Competitive Reality

What If a Competitor Is More Established?

If a competitor has 20 years of published work and Wikipedia entries, the strategy is not to outrank them on their territory.

block

Not their territory.

The strategy is not to outrank them on the domain they have occupied for two decades.

explore

Claim a distinct, adjacent domain.

The strategy is to claim a distinct, adjacent domain they have not occupied.

bookmark

Named frameworks separate the two.

Named frameworks and distinct entity signals separate the two.

verified

In the client's specific domain, the client becomes the source.

In the client's specific domain, the client becomes the source AI cites.

Timeline holds. The domain changes.

Five boundaries. What this is — and what it will never be.

No tricks that expire

The only way to be cited is to become a source worth citing.

block

Not SEO. SEO ranks pages in search results. AI Recommendation Infrastructure determines whether a language model names the person.

block

Not prompt hacking. We do not manipulate model outputs through crafted prompts. We build sources the model trusts.

dangerous

No baseline

Without a baseline, there's no measurement.

help

No methodology

No query set. No runs per query. No distribution.

analytics_off

No verification

You can't prove progress. You can't compare against a fixed standard.

We build the entity, not the trick. No methods that break on the next update.

21 The Change

What Changes After AI Recommendation Infrastructure?

Five shifts. One measurable outcome.

Before

Absent or misrepresented.

help

Asking AI returns others

The name is not among the cited.

auto_fix_high

AI invents facts

Roles, events, or controversies that aren't true.

shuffle

AI confuses the name

Attributes another person's work to the wrong person.

volume_off

Nothing exists for AI to cite

The model has no verified source to draw from.

analytics_off

No way to measure AI presence

No baseline. No distribution. No comparison.

arrow_forward

After

Cited and verified.

check_circle

Asking AI returns the name among the cited

Controlled sources occupy the AI answer surface.

check_circle

AI cites verified sources instead

Structured, corroborated information replaces invention.

check_circle

AI recognises a distinct entity

Distinct entity signals separate the name from others.

check_circle

Structured sources exist across every surface

Every surface a model reads has been structured.

check_circle

Monthly testing, written reports

10 runs per query, four models, raw outputs saved.

05 GATE

The Measurable Outcome

The name appears in at least 60% of runs across four retrieval models within 90 days.

Training-data models are measured separately. The timeline is controlled by the model provider, not us. Reported transparently.

Request Audit arrow_forward

22 Fit & Audience

Which System — and Who It's For.

AI Recommendation Infrastructure sits inside three systems. The Identity Audit determines which one fits.

check

Tier 01 — Presence

Foundation infrastructure in place

check_circleTechnical SEO + schema markup built into the foundation
check_circleEntity optimization — name, role, and authority defined
check_circleFoundational page-one control for the name and close variants
check_circleDigital headquarters live and indexed

check

Tier 02 — Authority

Full system + AI infrastructure

check_circleEverything in Presence, delivered
check_circleFull search infrastructure — page-one control across 15–20 queries
check_circleFull AI recommendation infrastructure — LLM citation setup
check_circleExtended content architecture across 20+ surfaces

check

Tier 03 — Legacy

Ongoing defence and recalibration

check_circleEverything in Authority, ongoing
check_circleQuarterly search & AI visibility review
check_circleReputation monitoring with 24-hour alert system
check_circleCrisis response protocol + priority strategist access

check

Included across every tier

The guarantees that hold everywhere

check_circleIdentity Audit runs first — nothing is built on assumptions
check_circleMilestone Guarantee — every missed gate means the next phase is free
check_circleEvery asset you own, always — no lock-in, no hostage tactics
check_circleComplete discretion — your name never used without written permission

check

Group 01

Being checked

check_circleAn invitation is being considered
check_circleAn application is under review
check_circleAn introduction is about to be made
check_circleSomeone senior will run the search within 48 hours

check

Group 02

Public role approaching

check_circleA public appointment is being announced
check_circleA keynote or published work is about to launch
check_circleA board seat or advisory role is being finalised
check_circlePress will run the search before the announcement

check

Group 03

AI has been tested

check_circleThe name is missing from AI answers
check_circleThe name is confused with someone else
check_circleAI invents facts about the person
check_circleAI recommends a competitor instead

check

Group 04

Body of work exists

check_circleYears of expertise, real results, published work
check_circleAI answers don't reflect any of it
check_circleFragmented presence across a handful of platforms
check_circleThe gap between reality and AI answers is the entire problem

check

Group 05

Nothing exists at all

check_circleSearches return an empty page
check_circleAI either invents or stays silent
check_circleNo legacy contradictions to correct
check_circleCleanest starting point. Timeline is longer than a search-only build.

check

Group 06

Already cited correctly

check_circleAI already names the person correctly
check_circleAnswers reflect the current role accurately
check_circleNo confusion with other namesakes
check_circleThen this is not urgent — and we will say so during the audit

check

The Direction

AI is moving from answering to acting

check_circleBots that research, shortlist, and recommend on behalf of a buyer
check_circleAn agent asked to "find the best [domain] expert in [city]"
check_circleDraws from the same sources as today's language models
check_circleThe entity that is structured, cited, and consistent is the entity the agent recommends

check

The Preparation

What the build already does for agents

check_circleEntity architecture — the agent needs to know who someone is
check_circleStructured data — the universal language of agent-readable surfaces
check_circleThird-party citations — corroboration the agent can verify
check_circleNamed frameworks — attributable, quotable, distinct

check

The Stakes

Why the work matters more, not less

check_circleAgents act without a human in the loop
check_circleRecommendations happen at machine speed
check_circleBeing absent from an agent's source list is not visible until the deal is lost
check_circleThe work is the same. The stakes rise.

check

The Preparation

How we prepare for it today

check_circleModel version tracking — behaviour changes against each version
check_circleQuery set evolution — expanded as agent questions change
check_circleRecalibration — the entity stays, the strategy evolves
check_circleRaw outputs — the record of what each model said, when

23 Reporting & Measurement

Control Center

Everything tracked. Every movement measured.

Monthly written report. Query-by-query results. Raw model outputs saved for verification. Quarterly review call.

Live · Updated monthly

Presence Rate

percent

60%+ trending_upGate 05 target

Queries Tested

list_alt

25–32queries infoSigned off in audit

Models Tracked

psychology

4models infoChatGPT · Gemini · Perplexity · Claude

Data Points

data_usage

1,000+/ cycle infoRaw outputs saved

Presence by model

Live

Perplexity (retrieval)

82%

ChatGPT with search (retrieval)

76%

Gemini with grounding (retrieval)

71%

ChatGPT without search (training)

44%

Claude (training)

38%

Google AI Overviews

68%

Retrieval models shift first. Training models lag on their next cycle. Both are measured. Both are reported.

Recent activity

Last 30 days

check

Perplexity now cites the name in 8 of 10 runs

3 days ago

verified

Wikidata entity verified with current role and affiliations

6 days ago

auto_awesome

ChatGPT citation added for "who leads in [field]"

10 days ago

trending_down

Two competitor names displaced from the top of the answer

14 days ago

warning

Hallucination detected in one model — source correction published

19 days ago

check

Gate 05 target measured — 60% threshold confirmed

28 days ago

description Monthly written report

find_in_page Query-by-query results

call Quarterly review call

24 Frequently Asked Questions

Straight answers. No fine print.

Everything you'd ask about AI Recommendation Infrastructure.

Structuring a digital presence so language models find, verify, and cite the name when someone asks who to trust in a field.

SEO ranks pages in search results. AI Recommendation Infrastructure determines whether a model names the person in an answer. Different mechanisms, connected systems.

Models invent when they have nothing verified to cite. We provide structured, verifiable sources. That reduces invention. It does not eliminate it.

25–32 queries, 10 runs each, across four models. Reported as distribution, with raw outputs provided.

No. Models weight sources by credibility, consistency, and corroboration. The only way to be cited is to become a source worth citing.

The build starts from zero. Foundational entity, structured data, named frameworks, third-party citations. Timeline is longer than a search-only build.

Yes. Retrieval models respond in weeks. Training-data models respond on their next cycle, 6–18 months. Both are measured separately.

Gate 05 of the Milestone Guarantee. The name appears in at least 60% of runs across four retrieval models within 90 days. If not, we keep working at no fee until it does. Training-data models are measured separately.

Request the Audit

See exactly what AI says when a name is searched.

The diagnostic that tests every model with the questions your stakeholders actually ask — and documents what comes back.

schedule3-week audit verified_userGate 05 guarantee visibility_offComplete discretion

Request Audit arrow_forward See Search Infrastructure

Six mandates per quarter. Senior attention only. Complete discretion.

Discipline 05 · Distribution

Control what AI says when a name is searched.