Your organisation probably already has a source of truth.
The problem is that AI does not see only that source.
It can encounter the product page your team updated yesterday, a PDF nobody has touched in two years, an old campaign microsite, a partner listing, structured data, a media article and an AI answer citing any one of them.
Internally, the organisation may know exactly which information is current.
The public web may still tell several different versions of the same story.
As AI systems increasingly discover, compare and synthesise information from across that public estate, enterprises need a way to govern not just the pages they publish, but the individual claims those pages make.
That is the role of a Claim Graph.
What is a Claim Graph?
A Claim Graph creates a governed record of the factual claims a brand makes in public and connects those claims to the places where they appear.
The idea starts with a simple shift:
The atomic unit is the claim, not the document.
Instead of treating an entire webpage, PDF or listing as one large piece of content, AXP ClaimGraph identifies the individual facts within it and maps where those same facts appear elsewhere.
It can then establish the organisation’s official record and compare the wider public information estate against it, across webpages, PDFs, microsites, structured data, controlled profiles, third-party sources and AI answers.
Consider a simple statement:
“Our personal loan starts at 9.2% p.a.”
For a customer, that looks like one sentence.
For the enterprise, it represents several connected pieces of information:
| Element | Example |
|---|---|
| Entity | Personal Loan |
| Attribute | Minimum interest rate |
| Value | 9.2% p.a. |
| Qualifier | Segment = salaried |
| Canonical value | Official approved value |
| Said by | Rate page, PDF, aggregator, AI answer |
Representing information at the claim level makes it possible to ask a much more useful question than:
Do these two pages contain different numbers?
The system can instead ask:
Are these sources actually making different claims about the same fact?
Context decides whether two claims actually conflict
This distinction matters because legitimate variation exists everywhere.
One loan rate may apply to salaried customers and another to self-employed customers.
An insurance premium may vary by age or geography.
A SaaS capability may exist on one plan but not another.
A hotel rate may change by date.
A university fee may vary by campus or intake.
A system that simply compares visible values would generate constant false alarms.
ClaimGraph therefore keeps qualifiers attached to claims so the conditions under which something is true remain part of the comparison. This helps distinguish legitimate variation from an actual contradiction.
That leads to an important principle:
Consistency does not mean making every public source identical. It means making sure every public claim is accurate in its proper context.
Contradictions rarely stay on one page
Most organisations think of outdated information as an old-page problem.
The real public estate is more complicated.
The same fact can diverge in several ways.
Live page vs live page
Two current product pages state different values for the same fact.
Neither appears obsolete.
Section vs section
The primary copy on a page says one thing while an FAQ or another section says something else.
The page is effectively contradicting itself.
Owned website vs microsite
A campaign or product microsite remains live with information that no longer matches the main domain.
Text vs structured data
The visible webpage has been updated, but its structured data still carries the previous value.
Humans and machines can effectively receive different versions of the same fact.
Text vs image or PDF
The HTML is current, while an embedded rate card, brochure or downloadable PDF contains older information.
Owned source vs third party
A partner, aggregator, comparison site, media article or another external source continues to publish a different version.
Correcting the main product page does not necessarily correct the public brand record.
The contradiction may live somewhere else entirely.
AI does not stop at your website
Large organisations accumulate digital properties over time.
Campaign sites remain online. Acquired companies retain domains. PDFs sit on CDNs outside everyday CMS workflows. Regional teams manage separate properties. Partner pages are not necessarily updated when owned content changes.
Some of these assets may receive very little attention internally.
Machines can still find them.
AXP ClaimGraph is designed to discover parts of this wider estate, including off-sitemap URLs, old microsites, acquired domains and documents that teams may no longer actively manage.
The governance question therefore changes from:
“Is our website correct?”
to:
“Where does our brand state this fact anywhere in public?”
Four layers make up the public brand record
The ClaimGraph model brings four information layers into the same system.
1. Owned content
This includes sources the organisation directly publishes:
- Websites
- PDFs
- Uploaded documents
- Video transcripts
- Old microsites
- Acquired domains
2. Controlled profiles
Some brand information exists outside the corporate domain but remains controlled by the organisation.
These can include business profiles, app stores, social profiles and managed listings.
3. Third-party mentions
Then there are sources the organisation does not directly control:
- Media
- Wikipedia
- Review platforms
- Comparison websites
- Aggregators
- Partner pages
These sources may continue repeating an earlier version of a fact even after the brand updates its own properties.
4. AI answers
The final layer is what AI systems are actually telling users.
AXP ClaimGraph can monitor an agreed prompt set across AI environments such as ChatGPT, Perplexity, Gemini and Google AI Overviews and trace the sources behind those answers.
That makes an AI answer both an output to monitor and a discovery mechanism.
If an AI answer repeatedly cites a forgotten PDF, that answer can lead the organisation back to a source it did not realise was still shaping its public record.
Why AI makes public consistency more important
Search engines have always indexed information from multiple sources.
Generative search changes what happens next.
Instead of simply presenting those sources as a list, AI systems can retrieve information from several places and synthesise it into a single response.
Google explains that its AI search experiences can use query fan-out, issuing multiple related searches across subtopics and sources before constructing an answer.
Read Google’s guidance on AI features and query fan-out
That creates a different risk for enterprise brands.
A customer may never see:
The current product page says 9.2%, while an older PDF says 10.4%.
The AI may simply select one.
If the wrong source wins, the contradiction can influence the answer without the organisation knowing it happened.
The enterprise still needs an official record
Finding every version of a claim is only half the problem.
The organisation also needs to establish which value it officially stands behind.
AXP ClaimGraph uses a canonical value as the official version against which public occurrences can be evaluated.
The platform can propose that canonical value, while a human can override the selection and the decision is logged.
That human-governance layer is important.
Technology can discover claims, compare them and surface disagreements.
The enterprise remains responsible for deciding what is true.
The operating model becomes:
Canonical claim → Public occurrences → Legitimate variations → Contradictions
Different inconsistencies create different risks
Claim governance also needs to distinguish between different kinds of problems.
AXP ClaimGraph separates findings into multiple categories.
Contradiction
Two live sources state different values for the same fact.
Outdated
An older source has been superseded by a newer official value.
Expired, still read
Information has passed its validity period, but people or machines are still encountering it.
Duplicate
Near-identical sources remain live, creating ambiguity over which one should be treated as authoritative.
Third-party divergence
A partner, aggregator, media source or another external property differs from the official brand record.
AI-answer divergence
An AI system states something that conflicts with the official claim, with the cited source traced where possible.
The distinction matters because each finding requires a different response.
An expired PDF may need removal or redirection.
A duplicate page may require consolidation.
A partner listing may require outreach.
An incorrect AI answer may first require tracing the source that caused the divergence.
The enterprise needs to know what is wrong, where it is wrong and what kind of problem it is dealing with.
The biggest risk is not the number of contradictions
Imagine an enterprise discovers 600 inconsistencies.
Where should it begin?
A conventional audit might prioritise them by severity.
But severity alone does not show how much an inconsistency is actually affecting the public brand record.
One outdated page may receive almost no readership.
Another may be frequently visited by customers, crawled by search engines and fetched by AI agents.
Both are incorrect.
Their exposure is very different.
AXP ClaimGraph therefore prioritises remediation using factors including severity, exposure and recency.
This shifts the operating question from:
“How many inconsistencies do we have?”
to:
“Which inconsistencies are actually shaping what people and machines learn about our brand?”
That turns claim governance from a content-cleanup exercise into an enterprise brand-governance and information-risk problem.
Measure exposure, not just inconsistency
AXP ClaimGraph uses Contradicted Exposure Share, or CES, to quantify that distinction.
CES looks at the share of total reads across the estate, including:
- Human pageviews
- Search-bot crawls
- AI-agent fetches
that land on documents contradicting the organisation’s official claims.
Contradiction count asks: How much inconsistent information exists?
Contradicted Exposure Share asks: How much inconsistent information is actually being consumed?
A company might have hundreds of outdated documents but very little exposure to them.
Another may have only a handful of contradictions, but those sources account for a substantial share of what customers and machines encounter.
The second situation may deserve much more urgent attention.
This makes CES useful as an exposure and prioritisation metric, rather than simply another count of content errors.
Why the CMS cannot solve the entire problem
Modern CMS platforms already provide important governance capabilities:
- Roles and permissions
- Approval workflows
- Version history
- Audit logs
- Scheduled publishing
Those capabilities remain essential.
But a CMS governs the information inside the CMS.
It does not necessarily know about:
- An old PDF stored elsewhere
- An acquired domain
- A campaign microsite
- A third-party listing
- A partner page
- A media article
- An AI answer citing an obsolete source
The distinction is straightforward.
CMS governance asks: Are we publishing approved information correctly?
Claim governance asks: Does the wider public information estate still agree with what we have approved?
Claim governance does not replace the CMS.
It extends the governance surface beyond it.
And it is not simply another content audit
A traditional content audit remains useful.
But it is usually a snapshot.
The result is often a spreadsheet containing hundreds or thousands of URLs and issues.
AXP ClaimGraph is designed around an ongoing, prioritised remediation queue rather than a one-time inventory. Findings continue to be detected and prioritised according to their impact and exposure.
The operating model moves from:
Audit → Spreadsheet → Cleanup
to:
Detect → Prioritise → Fix → Verify → Continue monitoring
That distinction matters when the underlying information estate is continuously changing.
Why regulated industries feel this problem first
For some enterprises, an incorrect public claim is more than a brand-consistency issue.
An interest rate, premium, dosage, fee or mandatory disclosure can carry direct customer and regulatory consequences.
The ClaimGraph use cases include claims such as:
- Interest rates, fees and foreclosure terms in banking
- Premium tables, benefits and exclusions in insurance
- Dosages, indications and package prices in healthcare and pharma
- Expense ratios and performance disclosures in asset management
- Carpet area, possession dates and RERA numbers in real estate
- Tariff plans and mandatory disclosures in telecom
In these industries, a published claim is not simply another piece of marketing copy.
It is information someone may act on.
The consequences of inconsistency can therefore extend beyond inaccurate AI representation into customer complaints, disputes and compliance exposure.
But the underlying problem is not limited to regulated sectors.
Any enterprise whose customers make decisions based on public facts needs to know whether those facts remain correct wherever they appear.
Claim governance is not about controlling AI
No enterprise can dictate every answer generated by ChatGPT, Gemini, Perplexity or another independent platform.
A Claim Graph is not designed to do that.
What an organisation can control is the quality, consistency and authority of the public information those systems encounter.
It is also not about forcing every source to use identical wording.
Different markets, products and audiences may legitimately require different values.
What matters is that the relevant qualifiers travel with those differences.
And ClaimGraph does not replace the CMS.
The CMS governs publishing.
The Claim Graph governs the public record created by publishing and everything that happens to that information afterwards.
Visibility is only useful if the answer is accurate
Much of the AI-search conversation still begins with one question:
“Does my brand appear?”
That matters.
But visibility is not enough.
A brand can be visible and still be described incorrectly.
It can be cited using an expired source.
It can be recommended using the wrong price.
It can appear inconsistent because an important qualifier has disappeared.
That is why the next layer of AI visibility is not simply more mentions.
It is accuracy, consistency and governance.
We explored the problem itself in Your Brand Is Visible in AI. But Is It Accurate?.
A Claim Graph provides the operating layer behind that idea.
It gives enterprises a way to answer:
What do we officially claim?
Where does that claim appear?
Which variations are legitimate?
Where does the public record disagree?
Which contradictions are actually being read?
And after we fix them, did the correction propagate?
AI visibility needs a truth layer
Within Publive AXP, the three product layers address different parts of the same AI-discovery problem.
AXP Edge focuses on making existing webpages readable to machines.
AI Streams focuses on building presence around the questions audiences are asking.
AXP ClaimGraph focuses on maintaining the governed official record and continuously identifying where the wider public estate disagrees with it.
The third layer matters because solving access does not solve accuracy.
An AI crawler can perfectly read an obsolete PDF.
A search engine can index two conflicting product pages.
An AI assistant can cite a high-authority third-party source that has not been updated in years.
Machine readability makes information easier to consume.
Claim governance helps ensure that the information being consumed is still the information the organisation stands behind.
As AI systems increasingly sit between brands and their audiences, that changes the scope of digital brand governance.
The website is no longer the entire brand record.
The public information estate is.
And enterprises need a way to govern it.
