How SeenAndCited Works

AI assistants now answer questions that used to begin as searches. A business can be described, compared, recommended or ignored inside those answers without ever knowing it happened. Most tools in this space report what happened and stop there.

SeenAndCited is built as an AI Visibility Operating System. That phrase means something specific here: one connected workflow that runs from understanding the business, through measurement and inspectable evidence, to a defined improvement problem, a recommended action, committed work, verification of what actually changed, and then subsequent measurement.

SeenAndCited brings together the parts of SEO, AEO and GEO that materially affect AI Visibility. What is AI Visibility? explains those terms and where conventional SEO tooling ends. This page explains the operating methodology itself.

What is happening?

Performance summarizes the current measured picture and meaningful changes in it.

What evidence supports it?

Monitoring lets you inspect the questions, AI engines, answers and cited sources behind that summary.

Why might it be happening?

Readiness and representation evidence examine whether important information can be reached, understood and reused, and whether what AI systems say lines up with confirmed business information.

What could improve?

Where evidence is strong enough, an Opportunity is identified — an underlying improvement problem, not a single failed question.

What should we do?

The AI Action Plan turns supported Opportunities into prioritized recommended work.

Did it actually change?

Where possible, the implementation itself is verified before any later measurement is used to describe an outcome.

What happened afterwards?

Completed work waits for eligible subsequent measurement before any outcome is stated.

Business Identity

Before the system can decide whether an AI answer refers to a business, it has to know what counts as that business. Business Identity holds the confirmed forms of the business name, accepted aliases, relevant product or brand names, and the owned web address.

This matters because a loose approximation is not proof. A confirmed name — or a variant a user has explicitly accepted — can count as a match; an unconfirmed approximation does not. Without an explicit identity, textual mentions, owned citations and competitor evidence cannot be interpreted reliably.

Website analysis

SeenAndCited analyzes the customer’s public website to understand what pages exist, what the business says about itself, which services, products and locations are relevant, how pages are structured, and whether a suitable existing page could support an improvement.

Website intelligence is refreshed when it is needed. Page-level implementation recommendations are made against current analysis rather than a stale picture of the site.

Confirmed information as approved business truth

Business information is not only used to understand positioning and generate relevant questions. It is also the reference point when SeenAndCited later evaluates how AI systems and relevant external sources describe the business — so the strength of each fact matters.

Confirmed

Information the customer has locked or explicitly confirmed. Only sufficiently strong confirmed facts can support a conflict or material-omission finding.

Well supported

Information supported by strong evidence but not explicitly confirmed. It informs context and can be promoted once confirmed.

Inferred or uncertain

Discovered or inferred information. It is never treated as approved truth, and a difference from it is not reported as a conflict.

Prompt Portfolio

AI visibility has to be measured against questions. Each site therefore has a Prompt Portfolio: the set of questions SeenAndCited repeatedly measures. Prompts are measurement instruments, not improvement problems — a question that repeatedly fails to produce a citation is evidence, and the problem it points to is represented separately.

Monitoring

Monitoring is the inspectable evidence layer beneath Performance. A monitored result can show the question, the AI engines measured, whether a usable answer was obtained, whether the business appeared, whether it was mentioned, whether it was recommended, whether it was cited, which sources were cited, where competitors appeared, and when the measurement happened. A mention, a recommendation and a citation are different outcomes and are never treated as interchangeable.

Surface Answers is the view into the actual stored answers behind those results — the answer text itself, per question and per engine. It is a window onto the same measurement, not a separate product.

See how SeenAndCited measures AI visibility, including usable, unavailable and unmeasured results

AI engine coverage

Monitoring can measure across multiple AI engines. Which engines are measured for a given site depends on that site’s current measurement configuration and plan; the workflow itself is the same in every case, while capacity, sampling depth and engine coverage vary. Positioning AI maintains its own separate engine coverage because it uses a different measurement approach.

Measurement shows what AI systems say. Readiness examines whether the underlying information is in a condition that supports being found, understood and reused. SeenAndCited organizes this around four questions rather than a checklist of technical tests, and a check that could not be completed is recorded as unknown rather than converted into a failure.

Can important information be accessed and discovered?

Indexing directives, search and AI crawler access, sitemap and content-discovery health, important-page discovery and internal-link evidence. Important information cannot contribute if automated systems cannot reach or find it.

Can systems understand what the page is about?

Heading and semantic structure, internal relationships between pages, structured data and its validity, and duplicate-content ambiguity where relevant. These are structural conditions that make content and entity relationships clearer or more ambiguous to automated systems.

Does the content actually answer the buyer’s question?

Question-to-page relationships, whether a substantive answer exists in ordinary prose as well as FAQ content, and whether material parts of the question are unanswered. Having content about a subject is not the same as clearly answering the question.

Is the answer clear enough to isolate and reuse?

Whether the relevant answer can be clearly identified and understood in context, or is buried, fragmented, mixed with unrelated topics or missing the heading context that would frame it.

Appearing in an answer is not the same as being described correctly. Where the evidence supports it, SeenAndCited compares relevant statements in measured AI answers against sufficiently confirmed business information, and records what the comparison shows.

Consistent

The statement lines up with confirmed information.

Conflicts

The statement contradicts confirmed information.

Cannot be substantiated

A claim SeenAndCited has no confirmed information to support.

Materially incomplete

Something important is missing, where the evidence supports that conclusion.

Not enough evidence

The comparison cannot be made safely, so nothing is asserted.

Citations as evidence, not causation

Where existing evidence associates a citation with a particular statement in a measured answer, SeenAndCited keeps that association: the statement that was made, the citation associated with it, and whether the cited source is owned, a competitor or another external source where that is known. Where no citation can be associated with the statement, that is recorded as such.

Check important external representations

Where SeenAndCited already has evidence that an external source matters to this business’s AI visibility — for example through a measured citation or a verified Authority source — it can assess whether an attributable external page is consistent with sufficiently confirmed business information. That can surface a confirmed contradiction, an unsupported external claim, materially incomplete information where justified, or evidence that cannot be verified.

SeenAndCited does not scan the web for listings, does not assume every external page should contain every business fact, does not edit external services automatically, and does not claim an external source caused an AI answer.

Performance is the interpreted view: what is currently happening, what changed, and the evidence and context behind it. The primary observed measures are Citation Rate, Share of Voice, and AI Referrals where first-party tracking supports attribution. Coverage sits beneath them as measurement context.

Citation Rate

How often the business was confirmed as cited across eligible measured questions.

Share of Voice

The business’s confirmed citation presence compared with explicitly configured competitors — not general market share.

AI Referrals

Visits arriving from a recognizable AI assistant referrer, where first-party tracking is installed. Many AI-influenced visits arrive without a recognizable referrer, so this is a floor rather than a total.

Coverage

How much of the planned measurement was successfully usable. Coverage is measurement completeness and context, not performance.

See how Citation Rate, Share of Voice and measurement eligibility work

Observed measurement and assessed diagnostics are kept visibly distinct. Specialist scores are labelled as assessments and never presented as equivalent to observed Performance.

Specialist areas examine different dimensions of the same picture. They are complementary views inside one workflow, not separate dashboards: competitor evidence, Authority evidence, citation and source evidence, readiness evidence and representation evidence all feed the same prioritization process, and where any of them produces actionable work, that work is handed to the central AI Action Plan and follows the same Task and outcome lifecycle.

Authority

Examines external-source and citation-environment evidence that may support or limit AI visibility: which independent sources reference the business, how that changes over time, and where cited sources cluster.

AI Readiness

Assesses whether important information can be accessed, discovered, understood, answered and reused — access and indexing conditions, discovery and internal links, semantic and structured-data clarity, answerability and answer clarity. High Readiness does not automatically mean high AI visibility, and a check that could not be completed is recorded as unknown rather than turned into a failure score.

AI Understanding

Assesses how complete and coherent the business information available to SeenAndCited is, and holds the representation evidence — how measured AI answers describe the business compared with confirmed information. It is not a claim about what any particular AI system knows about the business.

Positioning AI

Examines how supported AI surfaces describe and position the business in relevant market contexts — how a buyer would see the business framed. It is distinct from Citation Rate, from brand mention, from Share of Voice, from sentiment scoring and from SEO ranking, and a Positioning observation does not automatically become an Opportunity.

An Opportunity is the underlying improvement problem, supported by evidence. It is not the monitored prompt, not a generic recommendation, and not a Task.

Where a group of related questions repeatedly fails to produce a citation, the problem is represented once — for example “For ‘airport transfers in Havant’ — Improve AI visibility” — with the individual measurements held beneath it as evidence. That keeps the plan from becoming a long list of near-duplicate, question-level recommendations.

Each Opportunity keeps a stable identity as it is refined, so its evidence, committed work and measured outcomes stay attached to the same problem over time. Absence evidence is expected to repeat across distinct measurement days before it is treated as established; confirmed positive evidence can be verified from the cited source itself.

Opportunities can arise from any of the evidence the workflow holds: measured AI gaps, competitor differences, Authority and source evidence, access or discovery problems, answerability and answer clarity, structural or structured-data issues, duplicate ambiguity, representation findings and verified external inconsistencies.

The AI Action Plan is the prioritized operational view of what SeenAndCited recommends working on next. Each item connects the improvement problem, the recommended response and the supporting evidence, and can be read either by focus or by delivery stage.

Content Growth

Work that improves or creates content so an AI answer has something citable to use.

Authority Building

Work concerned with external sources and the wider citation environment.

Competitor Defense

Work responding to evidence involving explicitly configured competitors.

Other evidence-driven work

Readiness and representation work appears when the underlying evidence supports it, rather than on request.

Users can ask for more actions and choose a focus: Balanced, Content Growth, Authority Building or Competitor Defense. The important principle is that focus changes which supported actions are selected. It does not lower the evidence standard. Prioritization reflects how strong and how relevant the available evidence is, rather than counting warnings. If only three valid actions exist, SeenAndCited returns three rather than inventing filler to reach a requested number.

Some plan items are not visibility problems at all but setup requirements — such as confirming Business Identity. These are shown separately as System Actions and do not enter the measured-outcome lifecycle.

Knowing that visibility is weak is not the same as knowing what to do. For content-related Opportunities SeenAndCited recommends the specific intervention:

• whether to improve an existing page or create a new one;
• whether a direct-answer or FAQ section is the right addition;
• whether an eligible structured-data enhancement applies;
• which page is the best target;
• where on that page the change should go;
• why that intervention was recommended.

Because these are page-level recommendations, the website analysis behind them is refreshed when it is out of date before a page is targeted with confidence.

Where a change is published through a supported integration, a successful response from that system is not treated as proof that the page is live. SeenAndCited can check whether the resulting public page is reachable and whether the expected content is present, where the available evidence allows.

When a user decides to act on an Opportunity, a Task is created. The Opportunity is the underlying problem; the Task is one committed attempt to improve it. At the point work is committed, the relevant measurement position is recorded as the baseline for that work episode.

Where possible, SeenAndCited therefore verifies the implementation itself rather than relying on the task being marked complete — for example whether an indexing restriction was removed, whether crawler access changed, whether an internal link is actually present, whether the answer on the page improved, whether structured data is now valid, whether a published page is publicly live, or whether a correction to an external source is visible. Where verification is not possible, that is recorded as unknown rather than assumed.

Where no eligible evidence existed to form a baseline, that is recorded honestly and limits what can later be concluded about the work.

Three things are kept deliberately separate.

Work completed

The implementation task has been performed.

Implementation verified

Where possible, SeenAndCited checked whether the intended underlying condition actually changed.

Later AI measurement

SeenAndCited measured the relevant AI environment again and compared the later observations with the earlier baseline.

A completed Task moves the Opportunity to waiting for results: the intervention has happened and the result is not known yet. SeenAndCited then waits for eligible subsequent measurement — a settling period, and evidence on more than one distinct measurement day — before comparing it against the baseline recorded for that work episode.

Improved

Subsequent eligible evidence is better than the baseline.

No clear change

Enough comparable evidence exists, but it does not show sufficient improvement.

Weaker

Subsequent eligible evidence is weaker than the baseline.

Not enough evidence

A defensible before-and-after comparison cannot yet be made.

Where a representation issue was recorded, later measured answers can show whether that representation corrected, remains, changed but is still incomplete, or cannot yet be compared safely. That is a later observation about the AI environment, not a statement about what caused it.

A problem may need more than one attempt. If an issue remains actionable after an earlier work episode, new work can be created under the same Opportunity while the previous Task and its recorded outcome are preserved, so the history of attempts stays intact.

The methodology above is the same on Professional and Agency plans. Agency adds a delivery and client-management layer on top of it — multiple client sites, client management, client-facing reporting and white-label presentation where implemented — not a different measurement model.

Client portal

A simplified live view an agency can give a client: current Performance, measurement context, work in progress, work waiting for results, measured Outcomes, relevant specialist context, and a Positioning summary where applicable.

Report

A period-specific snapshot rather than a live view. It summarizes observed Performance, measurement context, the canonical work for the period, measured Outcomes and material specialist findings.

Reports use the same definitions and the same no-data rules as the product: an unmeasured citation rate is reported as not measured, never quietly rendered as 0%. Reports do not independently invent recommendations — they report the same canonical work and evidence the product holds.

SeenAndCited deliberately does not claim that:

• every AI response is measurable;
• an unavailable or unusable measurement means the business performed poorly;
• no available measurement means zero visibility;
• a completed Task means the work succeeded;
• a subsequent improvement proves causal attribution, simply because the work happened first;
• a citation associated with a statement caused the AI system to make it;
• a claim SeenAndCited cannot substantiate is therefore false;
• absence of evidence proves absence;
• fixing a technical or content condition will produce a citation or a recommendation;
• SeenAndCited knows how any AI system ranks or retrieves sources internally;
• every external source should contain every business fact;
• an assessed diagnostic is the same as observed AI-engine behavior;
• one content format guarantees better visibility across every AI engine, or that individual engines require a particular house style;
• every AI-influenced visit can be attributed.

AI systems are variable. Answers change between runs, availability differs between engines, and no measurement captures every possible answer. Measurements are therefore evidence over time rather than a complete picture, and this methodology is designed to make those limits visible rather than hide them behind a single score.