CBOUND Observatory · Research Index 001

What don’t we know about AI search yet?

Businesses and people are being found, interpreted, cited and recommended by AI systems. Many of the rules behind those decisions remain difficult to observe.

Much of today’s conversation focuses on optimisation. CBOUND is collecting the questions that need answering first. Some have partial answers. Some can be tested. Others remain genuinely unresolved.

FOUND
IDENTIFIED
UNDERSTOOD
EVIDENCED
CITED
RECOMMENDED
Research index

30 questions we’re not pretending to have answered.

Ten areas of uncertainty across discovery, identity, understanding, evidence, contradiction, time, citation, recommendation, correction and governance.

01 · Discovery

Can the system find the entity at all?

FEATURE QUESTION

What determines whether an AI system finds an entity at all?

Search visibility, source accessibility, structured information and third-party references may all contribute. The question is how those signals combine when an AI system retrieves information.

02 · Identity

Has the system found the right person or business?

Three questions currently being tracked in this area.
FEATURE QUESTION

How does AI decide that scattered information belongs to the same person or business?

Names, organisations, places, dates and relationships can connect fragments. Namesakes and incomplete profiles make that much harder.

03 · Understanding

Finding the entity is not the same as understanding it.

The purpose is to expose the uncertainty clearly before turning it into advice.
FEATURE QUESTION

What happens when AI finds the right business but misunderstands what it does?

A correct entity match can still produce an incorrect description. Business clarity is therefore a separate problem from discoverability.

04 · Evidence

What is the system treating as support?

Three questions currently being tracked in this area.
FEATURE QUESTION

When does repetition start looking like evidence?

The same claim can appear across many pages without becoming independently verified. Repetition, corroboration and source independence are not the same thing.

05 · Contradiction

What happens when legitimate sources disagree?

Three questions currently being tracked in this area.
FEATURE QUESTION

What happens when a website, LinkedIn profile and directory describe the same business differently?

Contradictions can reflect errors, different dates or genuine changes. Which source does an AI system privilege, and does it recognise the disagreement?

06 · Time

Does the system understand that the web changes?

Three questions currently being tracked in this area.
FEATURE QUESTION

Can AI distinguish a closed business from one that simply stopped updating its website?

Silence is ambiguous. An old website can belong to a dormant, closed or simply low-maintenance business.

07 · Citation

Why does one source appear while another disappears?

Three questions currently being tracked in this area. The purpose is to expose the uncertainty clearly before turning it into advice.
FEATURE QUESTION

Why does an AI system cite one source and ignore another?

Citation can depend on retrieval, accessibility, answer construction and the system being used.

08 · Recommendation

What changes between being known and being suggested?

Three questions currently being tracked in this area.
FEATURE QUESTION

What makes an AI move from understanding a business to recommending it?

An entity can be found and described correctly without being selected when a user asks for a recommendation.

09 · Correction

How does machine understanding change after the web is corrected?

Three questions currently being tracked in this area.
FEATURE QUESTION

How long does an incorrect business fact survive after the original source is corrected?

Correction at the source is only the first step. Secondary sources and refresh cycles may allow an old claim to persist.

10 · Governance

What happens when machine interpretation becomes consequential?

The purpose is to expose the uncertainty clearly before turning it into advice.
FEATURE QUESTION

When should observed information become associated, corroborated or verified?

These states should not be treated as interchangeable. Each requires a different level of evidence.

How CBOUND works with uncertainty

Observation before optimisation.

The index separates what has appeared in practical work from what can be tested and what remains genuinely unresolved.

01

Observe

Record the behaviour without assuming the cause.

02

Test

Compare systems, sources and controlled changes where a question can be examined.

03

Keep it open

Where the evidence is insufficient, preserve the question rather than manufacture an answer.

Living research document

What are we missing?

AI search is developing while businesses, publishers and users are still learning how it behaves. If you have encountered a contradiction, unexplained result or question that deserves investigation, it may belong in this index.

First published 22 September 2026 · Last reviewed 22 September 2026 · CBOUND Digital Visibility Observatory