AI visibility is the outcome. Editorial Intelligence is the operating system.

Most organisations don’t have a visibility problem.

They have an evidence problem.

The terminology around AI-mediated discovery is still unsettled. Some people call the work GEO or Generative Engine Optimisation. Others use AI search visibility, answer-engine optimisation, LLM visibility or related terms. This framework does not depend on one acronym winning.

The durable question is more useful: what does an organisation need to know, prove, structure and publish so people and AI systems can understand it accurately and usefully?

AI systems can work with the public and accessible signals around an organisation: owned content, customer proof, media coverage, reviews, communities, databases and other sources. Technical optimisation can make useful material easier to retrieve or interpret. It cannot create expertise, customer evidence or a distinctive position that does not exist.

But there is another failure mode too: the expertise and evidence may already exist and simply be stranded in the wrong asset, page or format.

That is the distinction this framework is built around.

AI visibility is not a prompt-tracking problem first. It is an evidence, positioning, knowledge-routing and reputation problem that technical optimisation can then amplify.

Editorial Intelligence provides the upstream system: capture useful signals, turn them into evidence and knowledge, connect that evidence to the answers that need it, develop coherent positions, activate them across owned and third-party channels, then learn from what buyers and AI systems reflect back.

A useful shorthand is:

Questions → evidence → editorial decision → format and distribution → visibility → learning

The shorthand is not a replacement for the wider Editorial Intelligence cycle. It is simply the AI-discovery application of it.

The operating model

Signals → Evidence → Knowledge → Editorial Intelligence
        ↓
Renew / route / report
        ↓
Narratives → Activation → Reputation
        ↓
AI visibility → Buyability → Commercial outcomes
        ↓
New signals

The final arrow matters. Buyer interactions, customer outcomes and market response create new signals. Those signals should improve the evidence base rather than disappear into the next campaign.

This is what makes AI visibility useful to Editorial Intelligence. It becomes another surface for learning what the organisation is understood for, what it can prove, where the public information environment is weak and where strong existing evidence has not yet reached the right answer.

GEO is the legibility layer — but visibility intelligence can travel upstream

Generative Engine Optimisation is useful. Important information should be accessible, well structured and easy for machines to interpret. Entities should be clear. Technical barriers should be removed. Organisations should understand how they appear across relevant AI systems.

But technical GEO works with what exists.

Technical GEO asks: how do we improve the discoverability and legibility of what we have?

Editorial Intelligence asks: what does the organisation need to know, prove, reconnect and publish so there is something worth discovering?

The sequence matters.

GEO / visibility work	Editorial Intelligence
Audit representation	Capture organisational signals
Identify weak answers	Distinguish page, evidence-routing and knowledge gaps
Optimise entities and structure	Connect evidence across sources
Improve technical signals	Build reusable narratives
Improve discovery	Compound learning over time

A technically perfect page cannot compensate for an organisation saying roughly the same thing as everybody else. Schema cannot create expertise. An FAQ cannot manufacture customer evidence. Making a generic claim easier to retrieve does not make it more credible.

Equally, a new article is not always the answer. Sometimes the organisation already has a stronger statistic, customer example or practitioner explanation elsewhere. The editorial job is to move that evidence into the answer that needs it.

The working application GEO-informed thought leadership takes this upstream logic into commissioning: use the answer environment to identify what is already known, what evidence can be reconnected and which important questions genuinely justify new research or reporting before deciding what to publish.

A dashboard can tell you where to look. Reporting tells you why it matters.

Visibility tools can surface a weak answer, a competitor gap or a cluster of prompts worth investigating. That is useful diagnostic information. It is not yet proof that the underlying question matters to real people, nor does it explain why the answer is weak.

The next step can be journalistic rather than technical: go to the places where the problem is being experienced and test the signal against reality.

In one Making Tax Digital project, AI-visibility analysis had already surfaced quarterly updates as an area worth investigating. That could have remained a prompt and page-optimisation exercise. Practitioner conversations at an industry event and evidence from an SMB advisory board showed something more useful: people were genuinely confused about what had to be submitted, when, what categories meant, whether software had connected correctly and how quarterly updates differed from the final declaration.

The dashboard identified where to look. The field evidence explained why it mattered.

That distinction should shape GEO work:

use visibility data to identify questions and weak representation;
test important signals against customer, practitioner, sales, support or market evidence;
separate synthetic prompt demand from demonstrated information need;
use what you learn to decide whether to renew an answer, route stronger existing evidence or report into a genuine knowledge gap.

This is the GEO journalist role in practice. It does not replace technical visibility work. It stops the technical layer from pretending it knows more about audience demand than the evidence supports.

The buyer journey can begin before your website

Some of the orientation, comparison and shortlist work that once happened through search results and websites can now happen inside an AI assistant.

A buyer may arrive on an owned property already carrying a provisional view of the category, the available options and the criteria that matter.

That gives content two different jobs.

Before the visit: become useful source material

AI-mediated discovery needs material that can be retrieved, compared, corroborated or synthesised. That puts more weight on:

clear product and category facts;
original research and proprietary data;
named expertise;
customer proof;
explicit use cases;
useful comparisons;
distinctive concepts and frameworks;
claims that are specific enough to verify;
positions that are supported consistently across sources.

After orientation: help the buyer decide

A buyer who reaches the website may need less generic education and more decision support:

focused case studies;
implementation detail;
pricing context;
security and compliance evidence;
ROI and outcome proof;
use-case specifics;
comparable customers;
clear differentiation;
material that helps somebody defend the decision internally.

This does not mean every B2B buyer begins in an LLM. It means content strategy should account for decision stages that the organisation may no longer observe directly.

Buyer questions come before prompt lists

Prompt tracking is useful. Treating prompts as the new keywords is less useful.

LLM queries are contextual and variable. Systems can expand a question into additional searches or sub-questions. A fixed list of prompts therefore gives you a sample of representation, not a stable equivalent of keyword rankings.

Start with the buyer instead.

Map the questions, constraints and decision criteria that actually matter:

What problem are they trying to solve?
What requirements narrow the shortlist?
What objections or risks matter?
Which use cases change the answer?
What comparisons do they make?
What language do customers use when they describe the problem themselves?

Sales calls, customer interviews, support conversations, reviews, communities, events and webinars are useful sources because they contain the language and criteria that real buyers bring to the decision.

Those signals can then generate a bounded set of prompt scenarios for testing.

The principle is simple: prompts are a measurement sample. Buyer questions are the strategic unit.

Product positioning is an evidence node

Product and category pages are not only conversion copy. They are important evidence nodes in the organisation’s public information environment.

If differentiation is vague on the website, unsupported in customer proof and inconsistent across third-party sources, an AI-generated description has weak material to work with. If the organisation is precise about who the product is for, what it does, why it differs and what proves those claims, there is a clearer account to retrieve and corroborate.

That does not mean one product page controls what an AI system says. It means positioning is distributed.

A claim becomes stronger when the product page, customer story, research, executive explanation, review and independent coverage point in the same direction.

The same logic creates a renewal opportunity. If a strong customer story or research finding already supports a product claim but the product or advice page never uses it, the evidence is being under-activated.

Evidence should travel beyond its original asset

The strategic unit is wider than a webpage — and smaller than an asset.

The useful unit is often the evidence itself: the statistic, customer outcome, expert explanation, practitioner quote, methodology, example or observation that supports a claim.

Research and experience
        ↓
Source-worthy evidence
        ↓
Evidence routed into the answers that need it
        ↓
Owned publication and activation
        ↓
Third-party validation and repetition
        ↓
Stronger public information environment
        ↓
Improved human and AI understanding

This is different from traditional repurposing. Repurposing asks how one asset can become several formats. Evidence routing asks where a piece of knowledge can materially improve an answer the organisation already needs to give.

A customer story can remain a customer story while one of its proof points strengthens a use-case page. A research report can remain intact while a finding updates an evergreen guide. A video can remain a video while its strongest expert answer becomes attributable written evidence elsewhere.

This is why PR, customer marketing, analyst relations, research, product marketing and editorial work cannot be treated as entirely separate visibility systems. They contribute different forms of evidence to the same reputation.

Measure three layers together

AI-mediated discovery creates an attribution problem because a buyer can use an AI assistant and never click an AI referral link. GA4 will only see the part of the journey that reaches the site in a measurable way.

A useful measurement model therefore has three layers.

1. AI representation

Sample relevant AI systems across agreed buyer questions and decision scenarios.

Look for recurring patterns rather than pretending every answer is a stable ranking:

Does the organisation appear?
How is it described?
Which topics or categories is it associated with?
Which sources or evidence appear?
How are competitors represented?
Which important claims are missing, vague or wrong?

2. Observable behaviour

Track what analytics can genuinely show: LLM referrals, landing pages, conversions and assisted journeys where the data is reliable.

Useful, but incomplete.

3. Qualitative buyer reconstruction

Ask buyers how they actually researched the problem.

Questions can be as simple as:

How did you search?
Did you use an AI assistant?
What did you ask it?
Which brands, claims or comparisons came up?
What made you investigate further?

That is not just attribution. It is customer-signal capture about how the organisation is being represented during AI-mediated discovery.

If those answers are recorded consistently in discovery calls, win/loss work or CRM fields, they can reveal buyer questions, competitor narratives and evidence gaps that referral analytics cannot see.

Turn visibility gaps into editorial routes

A weak AI answer does not always mean the page needs optimisation.

It does not always mean the organisation needs new content either.

A useful diagnosis produces three main editorial routes.

Route 1: renew

The right page exists and the organisation already knows enough, but the answer is stale, incomplete, poorly structured or too generic.

Action: refresh and republish the existing answer.

Route 2: route

The organisation already has stronger evidence, but it is stranded in another article, report, customer story, transcript, video or team.

Action: connect that evidence to the page or answer that needs it, with provenance and context.

Route 3: report

The organisation cannot answer the question strongly enough because the evidence itself does not yet exist.

Action: use research, customer access, events, webinars, expert interviews or other signal gathering to acquire the missing knowledge.

This creates a more useful loop:

Buyer question or decision criterion
        ↓
AI representation / visibility gap
        ↓
Inspect current answer + existing evidence
        ↓
Renew / route / report
        ↓
Stronger answer and evidence network
        ↓
Owned publication + third-party repetition
        ↓
Re-measure representation and buyer response
        ↓
New signals

When a visibility gap reveals a real knowledge gap, research-led thought leadership can become the intervention: investigate the question, generate original evidence, develop a defensible interpretation and connect it into a narrative the organisation can build over time.

But new evidence should not be commissioned merely to satisfy every missing prompt. GEO provides demand intelligence and a feedback surface; editorial judgement still decides which questions are worth investigating and which positions the organisation has earned the right to take.

The layers have distinct jobs:

AI Visibility & Buyability diagnoses where the organisation is visible, credible and retrievable — and classifies weak answers into renewal, evidence-routing or genuine knowledge gaps.
GEO-informed thought leadership applies that diagnosis upstream to thought-leadership commissioning: what should be renewed, what evidence should travel and what is important enough to investigate.
Research-led thought leadership creates original evidence and interpretation when an important question cannot be answered strongly enough from existing knowledge.
Signal gathering deliberately acquires missing knowledge and evidence through customer interviews, sales and support conversations, events, webinars, research, expert access and other interactions.
The Evidence Engine captures, verifies, connects and maintains that evidence so it can compound rather than disappear into isolated assets.
Narrative Architecture connects evidence and arguments into a coherent position rather than a collection of standalone assets.
Editorial Activation expresses the strongest evidence and positions in the formats and channels most useful to buyers, stakeholders and AI systems — including renewal of existing high-value pages.

Then AI visibility can be measured again. What changed becomes new evidence for the next cycle.

Content renewal is activation, not housekeeping

Updating an existing page is often treated as maintenance work: refresh the date, add a paragraph, check the links.

That undersells what a good renewal workflow can do.

If GEO intelligence shows that an important audience question is weakly answered, and an existing research programme or customer interview contains stronger evidence, updating the existing page is a form of strategic activation. The organisation is moving knowledge to the point where it can actually influence understanding.

That makes a useful content-renewal brief more specific than “refresh this page”. It should capture:

the buyer or GEO question;
the existing target page;
the current weakness;
the evidence that already exists;
the provenance of that evidence;
the proposed editorial change;
whether the page should be refreshed, expanded, connected or left alone;
the owner or publishing dependency;
what should be re-measured afterwards.

This matters particularly in organisations using conventional CMS and approval workflows. The intelligence layer may be AI-assisted and fast while publishing remains manual. The answer is not to pretend that bottleneck does not exist. It is to make the handoff compact, prioritised and batchable.

Five evidence-led updates to existing pages may create more value than five new articles.

The four pillars

Visibility — people and AI systems can find the organisation’s expertise.

Consistency — the underlying position is recognisable across the places buyers and machines encounter it.

Relationships — customers, journalists, analysts, reviewers and communities reinforce or challenge the reputation.

Credibility — claims are supported by evidence, provenance, methodology and demonstrated experience.

The pillars are connected. Visibility without credibility creates exposure without trust. Consistency without evidence creates repetition without authority. Relationships without a clear position create mentions that do not accumulate into meaning.

Buyability

Visibility is not the commercial outcome.

A buyer still has to feel confident enough to choose, recommend and defend the decision. That requires evidence that answers the practical questions surrounding a purchase: fit, outcomes, comparable customers, risk, implementation, cost, security, compliance and the organisation’s ability to deliver what it claims.

Customer proof, clear frameworks, research and third-party validation are therefore not merely marketing assets. They are decision-enablement assets.

AI-mediated research may change where some of that confidence begins to form. It does not remove the need for proof.

What this means in practice

Do not start by asking how many prompts you should track.

Start by asking what the organisation needs to be understood for, which buyer questions determine that understanding and what evidence currently supports the answer.

Then:

test how the organisation is represented;
inspect the current published answer and the wider evidence estate;
classify the problem as technical discoverability, content renewal, evidence routing or a genuine knowledge gap;
renew existing pages where the knowledge is already strong enough;
route useful evidence from customer stories, research, video and other sources into the answers that need it;
gather or create genuinely missing evidence where necessary;
connect the evidence into clear, reusable positions;
publish and activate it across owned and credible third-party channels;
measure representation, observable behaviour and buyer feedback together;
feed what you learn back into the next cycle.

That is the role of Editorial Intelligence in GEO, AI search, answer-engine optimisation or whatever terminology ultimately becomes standard.

The aim is not to become exceptionally good at manipulating individual AI outputs. It is to build an organisation that is easier to understand because its knowledge is clearer, its evidence is stronger, its useful archive is active and its public account of itself is more coherent.

Apply the model: GEO-informed thought leadership · AI Visibility Audit · Research-led thought leadership · Signal gathering · Evidence Engine · Narrative Architecture

This framework forms part of Editorial Intelligence OS — the practical operating system built to make the discipline repeatable.