# Claim and source review worksheet

Check the evidence behind one page before approving a factual change. Record each claim, the exact support and the allowed wording, then retain the checks needed to publish and maintain the correction.

Worksheet version 1.0 · 22 September 2026. Complete and repeat records as needed. Blank fields request your evidence; this manual file does not verify sources, publish changes or collect provider answers.

Complete guide: https://rankecho.io/resources/fix/fact-sourcing

## 1 Define the page and claim

Complete one page record, then repeat the claim and source fields for each material assertion. Preserve the original wording so a reviewer can see exactly what changes.

**Target and saved version** — Record the canonical page URL, page owner, review date, saved before version and the Find observation that prompted this review.

[Enter your response]

**Claim identity and location** — Assign a claim ID. Copy the current statement and its visible locator. List repetitions in metadata, FAQs, tables, accessibility text, product feeds or structured data where applicable.

[Enter your response]

**Claim type and review rule** — Distinguish product facts, research, derived calculations, estimates and opinions. State why this claim matters, what source would support it and any unresolved ownership or scope.

[Enter your response]

## 2 Match the claim to its evidence

Open the underlying record. A page about the same topic is insufficient if it does not support this exact claim. Repeat this section for each source and retain conflicting records.

**Source record and delivery** — Record source owner, title, original and final URL, response status, effective date or version, checked time and saved copy or retrievable record. Mark unavailable evidence as unresolved.

[Enter your response]

**Exact support and conditions** — Record the passage, section or data location. Compare subject, quantity, unit, population, date, geography, product version and exceptions. For calculations, retain inputs, formula, exclusions and rounding.

[Enter your response]

**Evidence judgment** — State what the record supports and what it does not. Identify inference, outdated or conflicting evidence and copied claims. Choose adequate, partial, contradicted or unresolved support with a reason.

[Enter your response]

## 3 Write the supported correction

Qualify, split, narrow or remove the claim as the evidence requires. Keep conditions beside the statement. These Northstar S1 examples are fictional; their manuals and bulletin are not live sources.

**Decision and allowed wording** — Record keep, qualify, add a supported exception, narrow, remove or hold. Write the exact proposed text and join every material clause to its source and locator.

[Enter your response]

**Public representations** — List each affected representation and its proposed correction. Check that metadata, FAQ text, accessibility labels and markup make no broader claim than the visible page.

[Enter your response]

**Reviewer and scope** — Name the reviewer, decision date, unresolved questions and exact approved change. Approval of wording does not mark deployment or live checks complete.

[Enter your response]

| Fictional claim | Action | Supported boundary |
| --- | --- | --- |
| F01 interval | Qualify | 12 months under standard conditions |
| F02 shock exception | Add | Bulletin conditions remain attached |
| F03 most accurate | Remove | No replacement superlative |
| F04 every system | Narrow | Modbus TCP and OPC UA only |

## 4 Validate and maintain the correction

Separate the proposed checks from their executed results. A supported content correction does not prove a ranking, citation, visit or sale.

**Acceptance evidence** — Record planned and actual results for claim coverage, exact support, source-link delivery, visible-copy parity and removal of unsupported wording. Retain status, time and evidence; unexecuted checks remain pending.

[Enter your response]

**Deployment and rollback** — Record approval, release/version and actual deployment time when known. Identify the saved supported version, rollback owner and triggers such as a missing qualifier, wrong source or unrelated change.

[Enter your response]

**Maintenance and later observations** — Assign source-change review triggers and the next owner/date. Keep registered prompt observations, missing cells and later traffic separate from content acceptance and causal claims.

[Enter your response]

## Complete guide

Fact sourcing is a claim-by-claim repair: inventory the material statements on one existing page, attach the most direct authoritative source available, record exactly what that source supports, and retain, narrow, rewrite, or remove each claim. Keep visible copy, links, dates, and structured data aligned. Acceptance is complete when every retained material claim passes that support check; a later AI citation is not required. Site traffic is a separate outcome and is not required to complete this fix.

Original guide update: 2026-09-07. Worksheet prepared 22 September 2026. The original guide and its evidence records retain their dates. The worksheet is an original manual form, not a provider-approved measurement standard.

## What is the scope of a fact-sourcing fix?

Use this fix after a citation-gap analysis identifies one page whose material facts are missing support, exceed their sources, use stale versions, or hide the relevant evidence from readers. Bound the work to that page, its visible claims, its structured data, and a dated claim-to-source record.

Sourcing can make a claim inspectable and correct. It does not reveal an answer system's selection formula, make markup an AI citation signal, or guarantee crawling, indexing, ranking, a mention, a recommendation, a citation, a referral, or traffic.

- [Return to the citation-gap diagnosis](https://rankecho.io/resources/find/ai-citation-gap-analysis)

## Which prerequisites must be attached before editing?

Capture the before state and define what counts as a material claim before changing the page.

- Canonical page URL, saved before version, visible text, metadata, structured data, publication dates, and every current outbound source link.
- The exact claim or field identified by the Find record, its page location, and the observed source comparison that motivated inspection.
- Claim types and source-selection rules: first-party product facts, public records, standards, research findings, derived calculations, estimates, and opinions are not interchangeable.
- A source owner, current version or effective date, checked date, support scope, correction path, deployment check, and rollback plan.
- The matched prompt-and-surface cells reserved for later proof, without making their outcome an acceptance condition for this edit.

## How should sources match different claim types?

Choose the source closest to the fact and preserve the boundary between a source statement and your inference.

| Claim type | Preferred evidence | Required boundary |
| --- | --- | --- |
| First-party product fact | Current product documentation, changelog, or directly inspected behavior | State plan, version, date, and inspection scope |
| Law, policy, or public record | The issuing authority's current text or dataset | State jurisdiction, effective date, and exceptions |
| Technical standard | The standard or steward's specification | State version and distinguish requirements from implementation advice |
| Research result | Original paper, report, and accessible data or method | State population, collection date, denominator, and limits |
| Derived calculation | Visible inputs plus the declared formula | Show arithmetic, exclusions, rounding, and owner |
| Opinion or recommendation | Named author and disclosed reasoning | Do not relabel judgment as measured fact |


## What does the worked synthetic claim-to-source artifact contain?

SYNTHETIC EXAMPLE — fictional brand, surfaces, sources, and observations; not RankEcho data, customer data, or a provider result.

Northstar S1 is fictional. Its source names are plain-text placeholders, not live evidence. The bounded after block keeps the supported interval and protocol facts, adds the documented exception, and removes the unsupported superlative.

| Claim ID and before text | Fictional source and exact scope | Action | Bounded after text |
| --- | --- | --- | --- |
| F01 · Calibration interval: 12 months | Synthetic Manual v3 §4.2 · only under standard conditions | Qualify and cite | Under standard conditions, the fictional S1 manual specifies a 12-month interval. |
| F02 · No shock exception stated | Synthetic Bulletin B-2026-04 §2 · recalibrate after shock | Add exception | Recalibrate after a shock under the fictional bulletin's stated conditions. |
| F03 · Most accurate sensor available | No market, measure, period, or independent comparison | Remove | Claim removed; no replacement superlative. |
| F04 · Works with every control system | Synthetic Manual v3 §7 · Modbus TCP and OPC UA only | Narrow and cite | The fictional S1 supports Modbus TCP and OPC UA. |


## How do you rewrite a claim to match its source?

Open the source, locate the exact supporting passage or data, and test subject, predicate, quantity, population, time, geography, version, and exceptions. Quote only when the wording matters; otherwise paraphrase without broadening. Label inferences and derived calculations as such.

If the source supports only part of a sentence, split the sentence. If the source is secondary, volatile, inaccessible, or contradicted by a more direct current record, find a better source or narrow the claim. If no adequate source exists, remove the material factual assertion.


## How must visible copy and structured data stay aligned?

Readers must be able to see the supported claim, its relevant qualifier, and the link or source context. Metadata, FAQ text, comparison tables, accessibility labels, and JSON-LD must not introduce a broader version of the claim than the rendered page.

Use structured data only when it represents the visible content and applicable type. Valid markup can support eligibility for applicable rich results; it is not a special AI citation mechanism and does not guarantee display.


## Which deterministic validation checks must pass?

Complete the fix when the page and its evidence record pass these checks. Do not wait for organic crawler traffic, a citation, a referral, or a RankEcho outcome threshold.

| Required check | Pass condition | Evidence to retain |
| --- | --- | --- |
| Claim coverage | Every retained material factual claim has one or more adequate sources | Claim identifier, page location, source URL, and action |
| Exact support | Subject, quantity, scope, date, version, and exceptions do not exceed the source | Supporting passage or data location and checked date |
| Source delivery | Every visible source link resolves to the intended authoritative record | Final URL, status, title, and checked time |
| Presentation parity | HTML, metadata, FAQ, accessibility text, and structured data make no broader claim | Rendered checks and structured-data diff |
| Correction safety | Removed and narrowed claims disappear from all public surfaces and can be restored from the saved before version | Search results, route render, diff, and rollback version |


## Which false positives and risks should stop the fix?

Stop when the proposed source merely repeats the claim, describes another plan or date, supports only a neighboring concept, or is cited because a competitor page uses it. A link is not support unless the opened record matches the exact assertion.

Do not add citations decoratively, hide qualifications in footnotes, publish fabricated statistics, create schema-only claims, or turn provider guidance into a universal selection rule. Preserve uncertainty, conflicting primary records, and source limitations where they affect meaning.


## What is the rollback and correction plan?

Save the before version and reverse the deployment if a source link is wrong, a material qualifier disappears, markup exceeds visible copy, the page breaks, or an unrelated claim changes. Restore the last supported version, then repeat the same checks.

When a source changes later, update the claim, checked date, and visible evidence together. Do not refresh a page date without rechecking the material sources that justify it.


## How does the fix hand off to proof?

Record the deployment time and exact semantic delta, then collect the pre-registered prompt-and-surface cells under the same protocol. Keep the content acceptance result separate from later answer outcomes.

Report no movement, adverse movement, and unavailable or failed cells. A changed manual cell-level owned-link rate after deployment is descriptive evidence under the matched design, not proof that fact sourcing caused the change.

- [Run the fixed-prompt retest](https://rankecho.io/resources/prove/fixed-prompt-retest)
- [Calculate the manual cell-level owned-link rate](https://rankecho.io/resources/prove/citation-rate)

## Who maintains this fact-sourcing guide?

Written and maintained by Abiot Y. Derbie. Published 2026-09-07; sources checked 2026-09-06; last updated 2026-09-07. The worked artifact is synthetic and does not represent a customer page.

Recheck source ownership, effective dates, product versions, visible-copy parity, and every material claim before applying the method. Send corrections through the contact page.

- [Send a correction](https://rankecho.io/contact)
- [Apply the evidence standard](https://rankecho.io/product/evidence-standard)

## Does adding authoritative sources guarantee an AI citation?

No. It makes the page's claims inspectable and bounded. Crawling, indexing, retrieval, selection, citation, referral, and traffic remain separate later observations.

## Should every sentence have an outbound link?

No. Inventory material factual claims and source them according to risk, volatility, specificity, and reader need. First-party experience, derived analysis, and opinion should be labeled rather than decorated with unrelated links.

## Can structured data contain facts that are absent from the page?

No. Google requires structured data to represent visible page content accurately. Keep the claim and its qualifiers aligned across HTML, metadata, FAQ text, accessibility text, and JSON-LD.

## What if a source supports only part of a claim?

Split or narrow the claim to the supported subject, quantity, population, date, version, and exceptions. Remove the unsupported remainder or find a more direct source.

## Must I wait for traffic or a citation to accept the fix?

No. Accept the content fix when every retained material claim and every public representation passes the declared support checks. Treat later outcomes as learning signals.

## Preserved dated evidence

- [Google helpful-content guidance](https://developers.google.com/search/docs/fundamentals/creating-helpful-content) — checked 2026-09-06. Google recommends helpful, reliable, people-first content and encourages clear authorship, appropriate production-method disclosure, and a people-first purpose while demonstrating first-hand expertise and supporting trust. Current people-first content and authorship guidance Primary-source documentation check; these editorial questions guide claim support and transparency but are not represented as a disclosed ranking formula or citation guarantee. Confidence: High

- [Google structured-data general guidelines](https://developers.google.com/search/docs/appearance/structured-data/sd-policies) — checked 2026-09-06. Google requires structured data to represent visible page content accurately and says valid markup makes a page eligible for supported rich results rather than guaranteeing that a feature will appear. Guidelines last updated July 10, 2026 Primary-source documentation check; visible-copy and markup parity is an acceptance check, not evidence that an answer system will select or cite the page. Confidence: High

- [Google guide to generative AI Search optimization](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) — checked 2026-09-06. Google says its generative AI Search features use core Search systems and require no special AI file, content chunking, writing style, or AI-specific structured data. Search eligibility does not guarantee display in a generative AI feature. Google Search guidance updated July 10, 2026 Primary-source documentation check; ordinary Google Search guidance is not generalized into a formula for other answer products or a guaranteed citation outcome. Confidence: High

- [OpenAI Help: searching the web with ChatGPT](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt) — checked 2026-09-06. OpenAI says ChatGPT responses that use web search may include citations and may expose cited sources and other relevant links in a Sources view. It also warns that search results and citations can be incomplete, outdated, or incorrect and should be checked against the opened source. Current ChatGPT search source guidance Primary-source documentation check; the observed visible citation is retained as evidence for that response, while source support and answer accuracy are checked separately. Confidence: High

## Related reading

- [Fix resource selector](https://rankecho.io/resources/fix)
- [Fixed-prompt retest](https://rankecho.io/resources/prove/fixed-prompt-retest)
- [Measurement definitions](https://rankecho.io/methodology)

## Product next step

After the evidence review, use the sample fix to inspect a bounded change and evaluate the Fix Engine for the next step. This manual worksheet does not verify sources or publish changes. https://rankecho.io/product/fix-engine