# Publication Eligibility Changes Are Not Citation-Rate Changes

A Machine Relations Index methods note separating publication-state transitions from true citation-rate changes in longitudinal AI citation reporting.

Canonical URL: https://machinerelations.ai/research/publication-eligibility-not-citation-change
Published: 2026-09-16
Research type: Index Analysis

## Source Body

A source entering a published Machine Relations Index leaderboard does not, by itself, prove that answer engines started citing that source today. It may mean the segment crossed the release evidence floor. To read a longitudinal change correctly, separate eligibility status, additional observations, source-role metadata, and true within-cohort citation-rate movement before naming a winner or decline. ([MRI public index](https://machinerelations.ai/data/machine-relations-index.json))

## Why publication eligibility is not the same as citation change

**Publication eligibility is a release-state rule; citation change is an observed-rate claim.** The September 16, 2026 Machine Relations Index reports source-segment citation rates only after a segment clears the evidence floor of at least 10 observed runs across at least seven distinct run dates. Segments below that line are marked `collecting` rather than scored, so an entry can appear in a published leaderboard because the denominator became eligible, not because engines changed behavior on the release date. ([MRI public index](https://machinerelations.ai/data/machine-relations-index.json))

The current release makes the interpretation risk visible. The September 16 public index reports 157 total strata: 151 collectable, 6 not collectable, 82 published, and 69 collecting. Its source ledger covers 15,540 observed answer runs across six engines from May 10 through September 16. Those totals are release context. They are not 157 equally sampled experiments, and they do not prove that any particular collecting row crossed into published status today. ([MRI release manifest](https://machinerelations.ai/data/mri-release-manifest.json)) ([MRI public index](https://machinerelations.ai/data/machine-relations-index.json))

The safe question is therefore not “who entered the leaderboard?” The safe question is “what changed between two released snapshots after holding the cohort, denominator, role metadata, and evidence status apart?”

## The four-transition table for longitudinal MRI reporting

**A transition table should classify the kind of change before it interprets the size of change.** A domain appearing in a new visible row can come from at least four different mechanisms, and only one of them is a true within-cohort citation-rate movement.

| Transition type | What changed between released snapshots | What stayed required for interpretation | Correct statement | Incorrect statement |
| --- | --- | --- | --- | --- |
| 1. Collecting-to-published eligibility | A segment moved from not publishable to publishable after meeting the release floor | The analyst still needs the first published numerator, denominator, date coverage, and release ID | “This segment is now published; prior releases did not publish a comparable rate.” | “The source improved from zero to its new rate.” |
| 2. Additional observations | More answer runs or dates were added to an already published segment | Same cohort definition, source domain, formula, and release boundary | “The rate was recomputed on a larger observed denominator.” | “More observations alone prove a directional citation change.” |
| 3. Source-role change | The domain’s interpretive metadata or grouping changed | Domain-level numerator and denominator must remain visible | “The role label changed; the underlying rate may or may not have moved.” | “The source became stronger because its role bucket changed.” |
| 4. Genuine within-cohort rate change | The same domain in the same cohort has a different cited-run share across two released snapshots | Same segment identity, formula, source domain, and comparable release snapshots | “The observed rate changed from A/B to C/D within this defined cohort.” | “A new leaderboard row proves citation behavior changed.” |

This structure deliberately keeps the methods note separate from three companion questions. It does not decide which category denominator to use, which is covered in the [AI citation leaderboard denominator note](https://machinerelations.ai/research/ai-citation-leaderboard-category-denominator). It does not test role-taxonomy sensitivity, which is covered in the [source-role classification note](https://machinerelations.ai/research/source-role-classification-sensitivity-ai-citation-rankings). It does not model interval uncertainty or run-date dependence, which is covered in the [run-date clustering note](https://machinerelations.ai/research/citation-rate-uncertainty-run-date-clustering).

## A worked hypothetical: one new published row, four possible readings

**The same visible leaderboard event can require four different interpretations.** The following example is hypothetical. It is not a claim about an observed September 16 transition, because that would require comparing two released snapshots and proving the exact row changed state between them.

| Snapshot comparison field | Hypothetical release N | Hypothetical release N+1 | Transition class | Interpretation |
| --- | ---: | ---: | --- | --- |
| Evidence status for `example.com / AI Visibility & GEO / best_x` | Collecting | Published | Collecting-to-published eligibility | The row became interpretable under the release floor. No prior published rate exists for this exact row. |
| Observed answer runs | 8 | 14 | Additional observations | The denominator crossed the floor and grew. The new rate is the first published rate, not a before/after improvement. |
| Distinct run dates | 5 | 7 | Collecting-to-published eligibility | Date coverage met the minimum. The date threshold explains publication state, not source performance. |
| Source-role label | Editorial publication | Editorial publication | No role change | Role metadata did not explain the transition in this hypothetical. |
| Runs citing domain | Unpublished | 3 | First published measurement | The safe statement is 3 of 14 in the first eligible release, not “up from zero.” |
| Citation rate | Unpublished | 21.43% | First published measurement | A published rate exists only in release N+1; no comparable published rate exists in release N. |

The tempting story is that `example.com` “surged onto the leaderboard.” The reproducible story is narrower: the row was collecting in release N, then became published in release N+1 with 3 cited runs out of 14 observed runs across seven dates. That is a first published measurement, not evidence of a change from 0% to 21.43%.

A true within-cohort rate-change claim would need a different setup. For example, if release N had already published 3 of 14 and release N+1 published 5 of 20 for the same domain, category, question type, formula, and source-role context, then the analyst could describe a movement from 21.43% to 25.00% for that defined cohort. Even then, the claim would remain descriptive unless a separate design justified causal language.

## How to compare two released MRI snapshots without overclaiming

**A release-to-release comparison should begin with row identity, not the displayed rank.** Machine Relations reporting should freeze both releases, match the same domain and segment identity, then classify the transition before reading the rate.

Use this sequence:

1. **Freeze both releases.** Record release date, release ID, methodology version, artifact hash, data-through date, and public index URL.
2. **Match the same row identity.** Hold domain, subject category, question type, source-segment definition, and formula constant.
3. **Read evidence status first.** If the earlier row was collecting, the later published row is a first published measurement, not a measured increase from zero.
4. **Check denominator and date coverage.** Record observed answer runs and distinct dates before reporting any rate.
5. **Check source-role metadata.** If the role label or grouping changed, report that as metadata change before interpreting a ranking change.
6. **Compare rates only after the row is comparable.** Use numerator over denominator for both snapshots, and state whether the comparison is descriptive rather than causal.
7. **Keep non-comparable rows out of the rate-change table.** Put them in an eligibility-transition table instead.

This sequence follows the same documentation logic used by statistical and metadata standards: define the unit, denominator, missingness state, and provenance before interpreting an estimate. NIST treats measurement context as part of technical interpretation; AAPOR separates sampled units and reporting definitions; the Census Bureau publishes design and data-quality documentation with estimates; FDA missing-data guidance separates observed outcomes from incomplete evidence. ([NIST/SEMATECH e-Handbook](https://www.itl.nist.gov/div898/handbook/)) ([AAPOR best practices](https://aapor.org/standards-and-ethics/best-practices/)) ([U.S. Census Bureau ACS methodology](https://www.census.gov/programs-surveys/acs/methodology/design-and-methodology.html)) ([FDA missing-data guidance](https://www.fda.gov/regulatory-information/search-fda-guidance-documents/missing-data-clinical-trials))

For machine-readable public data, the same principle appears in metadata standards. OECD's statistical quality framework, World Bank metadata guidance, ISO 8000 data-quality vocabulary, W3C DCAT, Dublin Core metadata terms, Schema.org Article, and Google Search Central structured-data guidance all preserve scope and provenance fields so machines can interpret a resource rather than strip it down to a headline. ([OECD quality framework](https://www.oecd.org/sdd/qualityframeworkforoecdstatisticalactivities.htm)) ([World Bank metadata guidance](https://datahelpdesk.worldbank.org/knowledgebase/articles/889386-developing-data-metadata)) ([ISO 8000-2 vocabulary](https://www.iso.org/standard/63555.html)) ([W3C DCAT 3](https://www.w3.org/TR/vocab-dcat-3/)) ([DCMI Metadata Terms](https://www.dublincore.org/specifications/dublin-core/dcmi-terms/)) ([Schema.org Article](https://schema.org/Article)) ([Google Search Central structured data](https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data))

## What a longitudinal leaderboard should show

**A longitudinal MRI leaderboard should include an eligibility-transition column next to any rate-change column.** Without that column, a newly visible row can be misread as a performance movement when it is actually a publication-state movement.

| Reporting field | Required value | Why it matters |
| --- | --- | --- |
| Earlier release | Date, release ID, artifact hash, and data-through date | Prevents mixing a valid release with an interim collection. |
| Later release | Date, release ID, artifact hash, and data-through date | Shows exactly which public snapshot is being compared. |
| Row identity | Domain, category, question type, and source segment | Keeps the cohort fixed. |
| Earlier evidence status | Published, collecting, or not collectable | Separates missing evidence from observed rate. |
| Later evidence status | Published, collecting, or not collectable | Shows whether the row became interpretable. |
| Transition class | Eligibility, additional observations, role metadata, or within-cohort rate change | Names the mechanism before ranking the result. |
| Numerator and denominator | Runs citing domain / observed answer runs for both snapshots when available | Makes rate movement reproducible. |
| Source-role label | Earlier and later role values | Prevents metadata shifts from being read as citation behavior. |
| Interpretation limit | Descriptive, not causal, unless a separate design supports causation | Preserves uncertainty and avoids conversion claims. |

This table is intentionally conservative. It makes the public Machine Relations Index more useful, not less. A reader who wants to know whether a domain “started being cited” needs to know whether the row had no published prior rate, had a published prior rate with a lower numerator, had a different denominator, or had a changed label. Those are different stories.

## What the September 16 release can motivate but not prove

**The September 16 MRI release motivates the eligibility problem; it does not by itself prove a specific transition.** The release reports 82 published strata out of 157 total, 151 collectable strata, 69 collecting strata, and a publication floor of 10 observed runs across seven distinct dates. It also reports 15,540 observed runs across six engines for the May 10 through September 16 window. ([MRI release manifest](https://machinerelations.ai/data/mri-release-manifest.json)) ([MRI public index](https://machinerelations.ai/data/machine-relations-index.json))

Those counts are enough to explain why readers will encounter new and changing leaderboard membership. They are not enough to say any named domain crossed from collecting to published today. That claim needs two released snapshots and row-level comparison. Until that comparison is made, the responsible wording is: “This release includes published and collecting strata under the September 16 evidence floor,” not “this source newly gained citation authority.”

That boundary is part of the [Machine Relations](https://machinerelations.ai/glossary/machine-relations) measurement discipline. AI visibility reporting is strongest when it refuses to turn release-state bookkeeping into a ranking narrative. A leaderboard row is useful only when its status, denominator, source role, release identity, and uncertainty travel with it.

## FAQ

### Does a new published leaderboard row mean answer engines started citing the domain today?

No. A newly published row may mean the segment crossed the evidence floor for publication. To claim engines started citing the domain, compare two released snapshots for the same row and show the earlier eligible numerator and denominator.

### How should MRI reports label a row that was collecting in the earlier release?

Label it as an eligibility transition or first published measurement. Do not write that it rose from zero unless the earlier release published an eligible zero for the same row.

### What is the difference between additional observations and a citation-rate change?

Additional observations mean the denominator or date coverage grew. A citation-rate change means the cited-run share changed inside a comparable published cohort. More observations can produce a new rate, but they do not automatically prove citation behavior changed.

### Can source-role changes make a leaderboard look different?

Yes. A role label or grouping change can alter interpretation even when the domain-level numerator and denominator stay fixed. Report role metadata separately from citation-rate movement.

### What should a longitudinal AI citation table include?

It should include both release identities, row identity, evidence status in each release, transition class, numerator, denominator, source-role label, and an interpretation limit. Rows that were not published in the earlier release belong in an eligibility table, not a rate-change table.

## Sources

1. Machine Relations Research. "MRI release manifest." September 16, 2026. https://machinerelations.ai/data/mri-release-manifest.json
2. Machine Relations Research. "Machine Relations Index public data." September 16, 2026. https://machinerelations.ai/data/machine-relations-index.json
3. Machine Relations Research. "Can an overall AI citation leaderboard pick the best source for a buying problem?" September 14, 2026. https://machinerelations.ai/research/ai-citation-leaderboard-category-denominator
4. Machine Relations Research. "Source-Role Classification Sensitivity in AI Citation Rankings." September 13, 2026. https://machinerelations.ai/research/source-role-classification-sensitivity-ai-citation-rankings
5. Machine Relations Research. "Citation-Rate Uncertainty Needs Run-Date Clustering, Not Just More Events." September 15, 2026. https://machinerelations.ai/research/citation-rate-uncertainty-run-date-clustering
6. NIST/SEMATECH. "e-Handbook of Statistical Methods." https://www.itl.nist.gov/div898/handbook/
7. AAPOR. "Best Practices." https://aapor.org/standards-and-ethics/best-practices/
8. U.S. Census Bureau. "ACS Design and Methodology." https://www.census.gov/programs-surveys/acs/methodology/design-and-methodology.html
9. U.S. Food and Drug Administration. "Missing Data in Clinical Trials." https://www.fda.gov/regulatory-information/search-fda-guidance-documents/missing-data-clinical-trials
10. OECD. "Quality Framework for OECD Statistical Activities." https://www.oecd.org/sdd/qualityframeworkforoecdstatisticalactivities.htm
11. World Bank Data Help Desk. "Developing Data Metadata." https://datahelpdesk.worldbank.org/knowledgebase/articles/889386-developing-data-metadata
12. ISO. "ISO 8000-2 Data quality vocabulary." https://www.iso.org/standard/63555.html
13. W3C. "Data Catalog Vocabulary (DCAT) - Version 3." https://www.w3.org/TR/vocab-dcat-3/
14. Dublin Core Metadata Initiative. "DCMI Metadata Terms." https://www.dublincore.org/specifications/dublin-core/dcmi-terms/
15. Schema.org. "Article." https://schema.org/Article
16. Google Search Central. "Intro to structured data markup in Google Search." https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data

## Attribution

This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research and standards initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about

## Machine-readable related links

### Related concepts

- [Machine Relations](https://machinerelations.ai/glossary/machine-relations)
- [Machine Relations Index](https://machinerelations.ai/glossary/machine-relations-index)
- [Citation Rate](https://machinerelations.ai/glossary/citation-rate)
- [Source Role](https://machinerelations.ai/glossary/source-role)

### Supporting research

- [Why ‘Collecting’ Is Not a Zero Citation Rate](https://machinerelations.ai/research/collecting-is-not-zero-citation-rate)
- [Can an overall AI citation leaderboard pick the best source for a buying problem?](https://machinerelations.ai/research/ai-citation-leaderboard-category-denominator)
- [Source-Role Classification Sensitivity in AI Citation Rankings](https://machinerelations.ai/research/source-role-classification-sensitivity-ai-citation-rankings)
- [Citation-Rate Uncertainty Needs Run-Date Clustering, Not Just More Events](https://machinerelations.ai/research/citation-rate-uncertainty-run-date-clustering)

### Framework context

- [Machine Relations Index](https://machinerelations.ai/index)
- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)

## Attribution

This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research and standards initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about

## Machine-readable related links

### Related concepts

- [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index)
- [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations)
- [Citation Rate](https://machinerelations.ai/glossary/citation-rate)
- [MRI Score](https://machinerelations.ai/glossary/mri-score)

### Supporting research

- [Citation-Rate Uncertainty Needs Run-Date Clustering, Not Just More Events](https://machinerelations.ai/research/citation-rate-uncertainty-run-date-clustering)
- [Forbes Answer-Engine Citation Authority: 4.04% Citation Rate Reveals Breadth-Over-Depth Editorial Strategy](https://machinerelations.ai/research/forbes-answer-engine-citation-authority-mri)
- [Can an overall AI citation leaderboard pick the best source for a buying problem?](https://machinerelations.ai/research/ai-citation-leaderboard-category-denominator)
- [Semrush AI Visibility Index: 126M Prompts, Citation Gap](https://machinerelations.ai/research/semrush-ai-visibility-index-126-million-prompts-citation-authority-2026)

### Framework context

- [Machine Relations Index](https://machinerelations.ai/index)
- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)