AI Claims Verification

AI Visibility Claims Verification shows what monitored AI platforms currently say about your organisation and its competitors. It groups related claims, shows how widely each claim appears and lets you check important claims against public evidence.


The Claims inventory is saved. You do not need to rebuild it every time you open the page.


What Claims can help you understand

Use Claims to find:


  • statements AI platforms make about your company, products or services;
  • statements they make about configured competitors;
  • claims repeated across personas, conversations or platforms;
  • claims that are new or have materially changed;
  • differences between AI platforms;
  • claims supported or contradicted by public evidence;
  • claims that are not publicly confirmed;
  • whether an AI answer's own visible citations supported its claim; and
  • cited pages that became unavailable or changed.

Claims organises existing AI Visibility monitoring results. It does not ask the monitored platforms new questions.


Before you begin

You need:


  1. Access to the active workspace.
  2. An AI Visibility project with at least one active persona configuration.
  3. Completed and processed monitoring responses.
  4. Your brand names and competitors configured correctly.
  5. Sufficient workspace credits when creating new claim extractions, retrieving uncached cited pages or running public verification.

Open the Claims inventory

  1. Open AI Visibility Claims.
  2. Select the Project.
  3. If a saved inventory exists, Wheremore loads it automatically.
  4. Use the tabs and filters to review the current claims.

The page shows the publication date of the latest saved inventory. A refresh can run in the background while you continue viewing the previously published inventory.


Find or refresh claims

For a project without an inventory, select Find claims.


For a project with an existing inventory, select Refresh claims.


Wheremore then:


  1. selects the newest fully processed monitoring batch for each active persona configuration;
  2. uses the expected conversations and platforms in those batches;
  3. reuses saved extraction for unchanged answers;
  4. extracts claims only from new or changed answers;
  5. records where every claim appeared;
  6. groups closely related assertions into claim families;
  7. checks the availability of new or expired cited pages; and
  8. publishes the completed inventory as one saved snapshot.

The inventory does not analyse months of historical answers. For each active persona configuration, it uses its newest fully processed monitoring batch. It never fills a missing current answer with an older answer.


You can leave the page while the inventory is being built. Until the new snapshot is published, other users continue seeing the previous saved inventory.


A message such as Cached inventory rebuild complete with 0 extraction requests is a valid result. It means every selected answer was unchanged and its saved extraction could be reused; it does not mean the rebuild was skipped.


Incomplete monitoring

If an expected answer was unavailable or could not be analysed, the page shows Monitoring incomplete.


This does not mean the whole inventory failed. It means the saved snapshot has incomplete coverage. Wheremore does not treat claims missing from those unavailable answers as having disappeared.


Understand the Claims views

Your brand

Shows grouped claims about the company, product or service names configured as your brands.


Competitors

Shows grouped claims about configured competitors. A competitor claim does not imply that you control the competitor's website or should edit its content.


If standalone competitor claims are disabled for the project, the Competitors tab is hidden.


Sources

Shows the domains and pages associated with the current monitored answers, including saved availability information where it could be established.


Some saved sources are background or provider context rather than visible citations supporting an answer. They remain useful source context but are excluded from the answer-source support assessment.


Understand claim families and breadth

Related assertions are placed in a claim family, such as Pricing visibility or Customer segment suitability.


Within a family, each distinct assertion remains visible. Wheremore preserves the original occurrences by persona, conversation and platform.


A breadth line such as Said 4 times across 3 of 12 monitored answers on 2 platforms describes:


  • how many accepted occurrences were found;
  • how many monitored answers contained the claim;
  • the total analysed-answer denominator; and
  • how many platforms carried it.

Breadth measures observed repetition. It does not prove the claim is true.


Filter the inventory

You can filter by:


  • Persona;
  • Conversation;
  • Platform;
  • Verdict;
  • Check state; and
  • Change.

Use Clear filters to return to the complete current inventory.


The filters change what is displayed. They do not rebuild the inventory or alter the saved monitoring scope.


Verdicts and check states

The public-evidence verdict and the operational check state are separate.


Public-evidence verdicts

  • Accurate — reliable public evidence supports the claim.
  • Partly accurate — some material parts are supported and others are not.
  • Incorrect — reliable public evidence contradicts the claim.
  • Sources disagree — reliable sources materially conflict.
  • Not publicly confirmed — the completed check did not find enough reliable public evidence to support a stronger verdict.

Not publicly confirmed does not prove that a claim is false. It means the public check could not substantiate it safely.


Check states

  • Not checked — no completed public check has been saved.
  • Checking now — a background verification job is active.
  • Checked — the saved verdict is current.
  • Needs re-check — the previous verdict is dated or its evidence requires another check.
  • Check didn't finish — a technical or provider condition prevented a reliable verdict. No customer verdict was recorded.

A failed check is not presented as an incorrect claim.


Check claims against public evidence

  1. Select the checkbox beside each eligible claim.
  2. Select Check selected claims.
  3. Leave the page open to watch progress, or return later.
  4. Expand a completed claim to review its finding and evidence.

You can select up to 25 claims in one manual batch.


Existing current verdicts remain available without another public check. A manual check again is an explicit request for a fresh check.


What the public check does

The public check asks whether the grouped assertion is supported by reliable evidence on the public web.


The expanded result can show:


  • the verdict;
  • a plain-English rationale;
  • supporting or contradicting evidence;
  • the source title and link;
  • a validated quotation;
  • pages consulted during the check;
  • where the claim appeared; and
  • technical check details when the user has diagnostic permission.

Pages listed as consulted were opened during the check. Consultation alone does not mean a page supported the claim.


Answer-source support

When Check the answer's own cited pages is enabled, Wheremore performs a second, separate assessment:


Did the original AI answer's visible citations support this claim?


This assessment:


  • uses only visible answer citations;
  • excludes provider-associated background sources;
  • retrieves each cited page once for the representative answer;
  • assesses selected claims from the same answer together;
  • does not run another public-web search; and
  • never changes the public-truth verdict.

Possible results include:


  • the answer's sources support the claim;
  • the answer's sources partly support the claim;
  • the answer's sources do not support the claim; or
  • Answer sources not assessable.

Answer sources not assessable commonly means that the answer included no visible citation for the claim or that no supplied page could be assessed reliably. It is not a public-truth verdict.


An unavailable cited page does not make the claim false.


Changes and evidence health

When enough saved snapshots exist, Claims can identify current changes such as:


  • a new claim;
  • a materially changed assertion about the same fact;
  • a claim that returned;
  • differences between platforms;
  • a changed public verdict;
  • evidence that became unavailable; or
  • evidence whose content changed.

A platform difference describes disagreement within the current inventory. It is not necessarily a change since the previous refresh.


Incomplete coverage prevents missing observations from being treated as proof that a claim disappeared.


An inconclusive health check, such as a timeout or access block, does not by itself invalidate a saved truth verdict.


Claims settings

Select Claims settings after choosing a project.


Inventory

Keep claim inventory updated automatically allows Wheremore to rebuild the saved inventory after new monitoring results become available.


This setting must be enabled if you want a newly completed monitoring batch to trigger a background inventory refresh. Wheremore will not repeatedly rebuild the same settled batch.


Public verification

You can enable:


  • Verify new claims automatically;
  • Verify changed claims automatically; and
  • Re-check saved verdicts after 90 days.

Automatic verification only selects eligible claims. Current saved verdicts can be reused without another public-web check.


Answer sources

Check the answer's own cited pages controls whether selected verification work also performs the separate citation-support assessment.


Settings apply to the selected project, not every project in the workspace.


Credits

Workspace credits can be used for:


  • extracting claims from new or changed answers;
  • organising related claims where AI grouping is required;
  • retrieving uncached or expired cited pages;
  • new public-evidence verification; and
  • answer-source support assessment.

Unchanged answers reuse the extraction cache. Fresh saved truth verdicts can also be reused across equivalent occurrences. These saved-result reuses can avoid new customer credit charges.


The exact cost varies with answer length, number of claims, source retrieval, verification work and whether focused corroboration is needed.


Claims and AI analysis

The Claims page does not create a separate Claims-only AI recommendations report.


Claims remains the source of structured findings. Where supported, verified claim evidence can be supplied to existing AI Visibility conversation, segment and cross-persona analyses. That inclusion should be conditional on Claims verification being enabled and available for the project.


Limitations

  • Claim extraction and semantic grouping use AI and may occasionally split similar claims or group claims too broadly.
  • Breadth reflects the monitored personas, conversations and platforms, not the whole market.
  • A completed public search cannot prove that no evidence exists anywhere.
  • Public pages can change after a verdict is saved.
  • Some websites cannot be read reliably because of access controls or temporary failures.
  • A visible citation can be reachable without supporting the relevant claim.
  • A claim can be accurate even when the original answer cited no supporting source.
  • Platform differences do not reveal the platforms' private retrieval or ranking systems.
  • Verification does not guarantee that a later AI answer will change.

Troubleshooting

Projects do not load

  • Confirm the correct workspace is active.
  • Confirm you are authorised for that workspace.
  • Reload the page after changing workspace.
  • Confirm the project has AI Visibility monitoring configurations.

Find claims is unavailable

  • Select a project.
  • Confirm at least one active persona configuration exists.
  • Confirm a fully processed monitoring batch is available.
  • Check whether another inventory refresh is already queued or running.

The inventory is taking a long time

The work runs in the background. Large answer sets, uncached extraction and source retrieval can take time.


You can leave the page. Do not repeatedly start another refresh while one is active.


A cached rebuild reports zero requests

This is expected when all selected answers are unchanged. Wheremore reused their saved claim extraction and published a new inventory snapshot without another model or page-retrieval request.


Check the publication date and coverage rather than expecting a non-zero request count.


Monitoring incomplete appears

One or more expected current answers were unavailable, failed processing or could not be analysed.


The inventory can still publish with its coverage recorded. Missing claims are not treated as absent.


A previous claim disappeared from the page

Check the active filters first. Then confirm that the newest saved inventory finished publishing.


If monitoring coverage was incomplete, absence and lost-visibility progression should remain blocked for the affected positions.


A checkbox is unavailable

The claim may not be eligible for public verification, a verification cycle may already be active, or the 25-claim manual limit may have been reached.


Checking remains at zero

Allow the background worker time to start. Reloading the page should reconnect to the existing job, not create another one.


If it remains unchanged, check the saved verification job state and worker logs before starting another batch.


Check didn't finish

The verification service did not produce a reliable completed result. This may follow rate limiting, a provider timeout, an invalid response or unusable evidence.


Select the claim and try the check again after the underlying issue has cleared.


If work reached a billing or receipt reconciliation state, Wheremore flags it for review rather than silently charging it under the wrong route. Contact support before repeatedly retrying the same claim.


Answer sources not assessable

The original answer may not contain a visible citation for the claim. Alternatively, the cited content may have been unavailable or unsuitable for a reliable assessment.


This does not change the public-evidence verdict.


A source is marked unavailable

Open the saved link. If the page now returns an error or redirects to the wrong destination, update or replace the cited material where you control it.


An unknown or inconclusive availability result is not the same as a confirmed missing page.


A fresh inventory contains fewer answers than expected

Claims uses the latest fully processed batch for each active persona configuration. Permanent monitoring errors remain missing coverage; Wheremore does not substitute older answers.

Did this answer your question? Thanks for the feedback There was a problem submitting your feedback. Please try again later.