Mention vs citation vs recommendation: classify outcomes
Mention, owned citation, other citation, and recommendation are separate, nonexclusive labels applied to each eligible observed cell under frozen text, source-role, host, and endorsement rules. Preserve ambiguous links, unavailable surfaces, and failed collections instead of forcing them to no. This companion ledger reports counts and coverage only; it is not a citation rate or evidence that the identity patch caused an outcome. Site traffic is a separate outcome and is not required.
What is the scope of this fixed-cell classification?
Use it after a dated, deterministically accepted change when the original prompt and context panel remains frozen. The unit is one prompt-surface cell, and the four outcome labels can coexist.
The method describes what the preserved answer and displayed source roles show. It does not infer retrieval, provider confidence, recommendation logic, or the reason an outcome changed.
Which prompts, contexts, and rules stay frozen?
SYNTHETIC EXAMPLE — Harborlight, Harborlight Systems, Inc., Harborlight Atlas, every URL, prompt, surface, assertion, patch, and outcome below are fictional; this is not RankEcho data, customer data, provider evidence, or a benchmark.
The synthetic shipment is recorded at 2026-08-24T16:00:00Z; the outcome window is 2026-08-31T15:00:00Z. Six prompts crossed with two contexts create 12 planned cells.
The mention alias rule, qualifying citation role, URL normalization, owned-host rule, other-citation rule, recommendation language, locale, session, and exposed-model label remain unchanged across the panel.
How are mention, owned citation, other citation, and recommendation separated?
Mention requires a declared target alias in answer text. Owned citation requires a qualifying UI-marked source URL on the target host. Other citation requires a qualifying source on another host. Recommendation requires explicit language selecting or endorsing the target for the prompt's job.
A citation can appear without a text mention, a mention can appear without a citation, and a recommendation can appear without an owned citation. Related links are not silently promoted to citations; unresolved roles remain ambiguous.
What does the worked synthetic outcome ledger show?
Ten of 12 planned cells contain eligible observed answers. C06 is unavailable and C12 fails before an answer is preserved, so their four labels remain not observed rather than absent. C04 has qualifying owned and other citations, while its Atlas mention and recommendation remain ambiguous because the referent cannot be resolved.
| Cell | Prompt and surface | Collection | Mention | Owned citation | Other citation | Recommendation |
|---|---|---|---|---|---|---|
| C01 | Q01 · CTX-A | eligible-observed | present | present | absent | present |
| C02 | Q01 · CTX-B | eligible-observed | present | absent | absent | present |
| C03 | Q02 · CTX-A | eligible-observed | present | present | present | absent |
| C04 | Q02 · CTX-B | eligible-observed | ambiguous | present | present | ambiguous |
| C05 | Q03 · CTX-A | eligible-observed | present | absent | present | absent |
| C06 | Q03 · CTX-B | unavailable | not observed | not observed | not observed | not observed |
| C07 | Q04 · CTX-A | eligible-observed | present | absent | absent | present |
| C08 | Q04 · CTX-B | eligible-observed | present | absent | absent | absent |
| C09 | Q05 · CTX-A | eligible-observed | absent | absent | present | absent |
| C10 | Q05 · CTX-B | eligible-observed | present | present | absent | absent |
| C11 | Q06 · CTX-A | eligible-observed | present | absent | absent | absent |
| C12 | Q06 · CTX-B | failed | not observed | not observed | not observed | not observed |
How do the counts and coverage reconcile?
10 observed of 12 planned. Among those ten eligible observed cells, mention is present in 8, owned citation in 4, other citation in 4, and explicit recommendation in 3. These are overlapping counts, so they must not be added together.
C03, C04, C05, and C09 contain a qualifying other-domain citation. C04's target alias and endorsement label remain ambiguous while both displayed citation roles stay classifiable.
| Outcome | Present | Absent | Ambiguous |
|---|---|---|---|
| Mention | 8 | 1 | 1 |
| Owned citation | 4 | 6 | 0 |
| Other citation | 4 | 6 | 0 |
| Recommendation | 3 | 6 | 1 |
What does the M/C/R cross-tab preserve?
The three-bit signature records target mention, target-owned citation, and explicit recommendation in that order. Among nine resolved signatures it contains one 000 cell, zero 010 cells, three 100 cells, two 101 cells, two 110 cells, and one 111 cell; C04 remains indeterminate. Other citations remain a separate context field.
The cross-tab is a count ledger, not a percentage, score, hierarchy, conversion funnel, or rate.
| M/C/R signature | Observed cells | Meaning |
|---|---|---|
| 000 | 1 | No target mention, owned citation, or explicit recommendation |
| 010 | 0 | Owned target citation without a target-name mention or recommendation |
| 100 | 3 | Target mention only among the three target outcomes |
| 101 | 2 | Target mention and explicit recommendation without an owned citation |
| 110 | 2 | Target mention and owned citation without an explicit recommendation |
| 111 | 1 | Target mention, owned citation, and explicit recommendation |
| indeterminate | 1 | At least one target M/C/R label remains ambiguous |
How are ambiguous, unavailable, failed, and zero-eligible states reported?
Ambiguous applies only to the affected label; the rest of an eligible cell remains classifiable. Unavailable and failed cells have no outcome labels and never enter an absent count.
If no planned cell produces an eligible observed answer, the outcome display is N/A. It must not render four zero counts as though 0 were an observed result.
How is this different from adjacent measurements?
Citation rate uses a positive rule and eligible-cell denominator. Answer volatility compares feature signatures across repeats. Fixed-prompt retest compares matched periods. Citation persistence follows exact-URL continuity across ordered windows. This page supplies the per-cell outcome labels those methods may consume; it calculates none of their rates.
What interpretation is supported after the patch?
The frozen synthetic window contains the stated label counts after the accepted deployment. It does not establish that the identity patch caused any mention, source link, recommendation, absence, or ambiguity.
Provider changes, source changes, prompt context, ordinary answer variation, and other concurrent conditions remain plausible. No traffic, ranking, referral, conversion, or revenue conclusion follows from this ledger.
What can RankEcho retain and what remains a companion method?
RankEcho’s current appearance signal can be positive from either a target-owned URL or target-name text, and its automated brand stance labels are heuristic review leads. Neither is this companion ledger’s manually adjudicated mention, owned-citation, other-citation, or explicit-recommendation label. RankEcho does not currently compute this four-label count table or M/C/R cross-tab as a supported product metric.
Attach the cell ledger to the proof report, and use Proof Loop only under its documented product units. Do not relabel these manual counts as another product measure.
What deterministic validation completes Prove?
All 12 coordinates and evidence pointers must be unique. Ten observed cells must supply four non-null labels, the unavailable and failed cells must supply none, recommendation-present cells must also mention the target, and the counts and cross-tab must recompute exactly.
Completion depends on the declared collection record, not a favorable direction or site-traffic threshold.
| Check | Pass condition | Evidence |
|---|---|---|
| Panel | Six prompts crossed with two contexts produce 12 unique coordinates | Frozen prompt and context records |
| Collection | Ten observed, one unavailable, and one failed state remain separate | Complete cell ledger |
| Labels | Every observed cell has four labels; non-observed cells have none | M/C/O/R fields |
| Recommendation | Every present recommendation has a resolved present mention | Per-cell rule check |
| Counts | M 8/1/1, C 4/6/0, O 4/6/0, and R 3/6/1 recompute | Counts-only projection |
| Cross-tab | Nine resolved signatures plus one indeterminate signature reconcile | M/C/R cross-tab |
| Zero eligible | No observed cells renders N/A rather than zero outcomes | Null-display check |
How does the complete record return to Find and Fix?
Send the full ledger, including C04 ambiguity, C06 unavailability, C12 failure, and every absent label, back to Find. Select another Fix only when a new directly evidenced owned-publication defect survives competing explanations.
Who maintains this guide?
Written and maintained by Abiot Y. Derbie. Published 2026-09-07; sources checked 2026-09-07; last updated 2026-09-07. The outcome ledger is synthetic.
Recheck source-interface roles and frozen classification rules before applying the method. Send material corrections through the contact page.
Sources reviewed
Provider eligibility and measurement claims below were checked against primary documentation. These records do not establish a universal selection formula, causation, or a guaranteed ranking, impression, recommendation, or citation.
4 claim-level source records
| Claim reviewed | Official source | Review record |
|---|---|---|
| OpenAI says ChatGPT Search responses may include citations that open their sources and may provide a Sources panel with cited sources and other relevant links. | OpenAI Help: searching the web with ChatGPT | Checked 2026-09-07 · Current consumer ChatGPT Search guidance · Primary-source product guidance; it does not define this guide's mention or recommendation labels, expose a source-selection cause, or make all links equivalent citations. · Confidence: High |
| Google's Gemini Apps help explains that a response can provide Sources and related links when available and that not every response includes sources. | Google Gemini Apps Help: sources and related links | Checked 2026-09-07 · Current Gemini Apps source guidance · Consumer-interface guidance; related links and displayed sources are not interchangeable, and the page does not define this manual mention, citation, or recommendation taxonomy. · Confidence: High |
| Bing's public-preview AI Performance report exposes citation counts, cited pages, trends, and sampled grounding-query phrases while warning that citation count does not indicate placement, ranking, authority, or page importance. | Bing Webmaster Blog: AI Performance | Checked 2026-09-07 · Public preview announced February 10, 2026 · Primary-source documentation; Bing's aggregate citation reporting does not define this manual cell-level outcome taxonomy or its eligible population. · Confidence: High |
| OpenAI's Web Search API documentation distinguishes inline url_citation annotations from the broader sources list returned for a search call. | OpenAI API: web search tool | Checked 2026-09-07 · Current OpenAI Web Search API documentation · Direct API-schema support for separating cited URLs from broader retrieved sources; it does not define target mention, URL ownership, recommendation, cross-provider interface roles, or this manual ledger. · Confidence: High |
Frequently asked questions
No. A mention is declared target text in the answer; a citation is a qualifying displayed source link under the frozen interface-role and URL rules.
No. A recommendation requires explicit selection or endorsement language for the prompt's job. A source link alone does not supply that language.
Yes. Mention, owned citation, other citation, and recommendation are nonexclusive fields and must be recorded separately.
No. They contain no eligible answer to classify and retain not-observed labels rather than entering absent counts.
No. It reports overlapping label counts and observed coverage only. Rate calculation requires a separately declared numerator and eligible denominator.
No. It is a later observation under the frozen rules and does not isolate the patch from provider, source, context, or run-to-run changes.