Share of Citation

Machine Relations defines Share of Citation as a specified brand or domain's qualifying citations as a percentage of all qualifying citations counted in the same sampled answers. It records how much of the cited-source pool the subject occupied within a declared measurement window. It does not establish claim support, endorsement, recommendation, or business impact.

Share of Citation Defined #

Machine Relations defines Share of Citation as a share of the cited-source pool:

Share of Citation = (qualifying citations to the specified brand or domain / all qualifying citations in the same sampled answers) x 100

An arithmetic illustration: if a sample of 100 eligible answers yields 320 qualifying citations and 45 of them resolve to the specified domain, its observed Share of Citation is 14.1% for that sample. This is arithmetic, not a benchmark.

The measurement contract must declare:

• the query set;
• the engine or engine set;
• the observation window;
• the number of runs per query and engine;
• the unit counted as an eligible answer;
• the unit counted as a qualifying citation, including whether repeated links to one domain inside a single answer are counted once or each time;
• the identity rule used to associate a citation with a brand or domain; and
• any weighting used to combine engine-level results.

Share of Citation is not Citation Rate #

Share of Citation and Citation Rate are computed from the same sample and answer different questions, so they are not interchangeable and a report names which one it carries.

• Share of Citation divides the subject's qualifying citations by all qualifying citations observed. Its denominator is the cited-source pool, so it is a competitive position: it moves when other sources gain or lose citations even if the subject's own count does not change.
• Citation Rate divides eligible answer runs containing at least one qualifying citation to the subject by all eligible answer runs. Its denominator is the sample, so it is absolute coverage: it moves only with the subject's own presence.

One sample produces both numbers and they will differ. In the illustration above, if 19 of the 100 eligible answers cite the specified domain at least once, the same sample reports 14.1% Share of Citation and 19% Citation Rate. Neither figure is a correction of the other, and a percentage published without naming its denominator cannot be compared with another.

Jaxon Parrott introduced Share of Citation in Machine Relations measurement work.

Microsoft added a platform-native Citation Share metric to Bing Webmaster Tools in June 2026 (Microsoft Bing).

---

What the Metric Observes #

Share of Citation answers one declared question: of all the citations engines issued in this sample, what fraction resolved to the specified brand or domain?

The metric does not classify mention-only appearances, claim support, sentiment, recommendations, source-selection explanations, traffic, consideration, revenue, or other downstream outcomes.

A citation is an observable attribution event. It is not proof that the linked page supports the answer's claim. Machine Relations treats citation presence and Answer-Source Fidelity as separate measurements: Share of Citation records whether a citation appeared; Answer-Source Fidelity evaluates whether the cited page supports the associated claim.

Share of Citation and Adjacent Metrics #

Machine Relations keeps these observations separate:

• Share of Citation divides the subject's qualifying citations by all qualifying citations in the sample; it does not grade claim support or recommendation.
• Citation Rate divides eligible answer runs citing the subject by all eligible answer runs; it reads absolute coverage where Share of Citation reads competitive position.
• AI Share of Voice compares declared brand-observation counts across a competitive set and uses a different denominator.
• Brand Mention Rate counts eligible answers naming the brand, linked or unlinked; a mention is not necessarily a citation.
• Citation Velocity records change in citation observations between comparable windows; it does not identify the cause.
• Citation Gap records declared queries where a comparison domain is cited and the focal domain is not; it does not prescribe an intervention.
• Answer-Source Fidelity grades claim-source pairs for evidentiary support; it does not measure citation frequency.

Organic rank and citation presence are different observations.

A Moz analysis reported that 88% of citations in Google's AI Mode did not match the organic top 10 for the same query (Moz, 2026).

An Ahrefs analysis covering 863,000 keywords reported that 38% of pages cited in AI Overviews also appeared in the organic top 10, down from 76% in its prior comparison (Search Engine Journal, 2026).

Those studies describe their samples; Machine Relations does not treat them as a universal relationship or a causal mechanism.

---

Measurement Protocol #

The following is the Machine Relations protocol for producing a reproducible Share of Citation observation.

1. Freeze the Query Set #

Record the exact query text and a query-set version before collection. Branded and non-branded queries may both be measured, but the protocol labels and reports them separately as different sampling conditions.

2. Declare the Engines #

Record the engine name, model or product surface when available, collection method, and collection date. Do not present a result from one engine as a cross-engine result.

3. Repeat and Preserve Runs #

Choose and disclose the number of runs per query-engine pair. Preserve the raw answer and its citations so another reviewer can reproduce the numerator.

Repeated runs represent output variation in this protocol. The chosen run count is a protocol parameter, not a universal adequacy threshold.

4. Apply One Citation Rule #

Before collection, define what qualifies as a citation and how redirects, subdomains, syndicated URLs, and brand-domain mappings are handled. Apply the same rule to every answer in the sample.

5. Publish the Denominator #

Report the numerator and denominator beside the percentage so the total qualifying-citation count remains visible, and name the eligible-answer count that produced it.

6. Report Per-Engine Results #

Calculate each engine separately:

Per-engine Share of Citation = (qualifying citations to the specified brand or domain on that engine / all qualifying citations in that engine's sampled answers) x 100

A blended result is accompanied by its weighting rule and the component engine observations.

A 2026 Foglift study found low overlap among top-cited domains across the engines in its sample, with 61.7% appearing in only one engine's rankings (Foglift, 2026).

That finding is sample-specific and does not define an expected overlap rate.

7. Compare Like With Like #

Trend comparisons require the same query-set version, engine definition, citation rule, and sampling design. When any of those inputs changes, publish a new series or mark the discontinuity rather than presenting it as ordinary movement.

---

Platform-Native Measurement #

Microsoft added Citation Share to Bing Webmaster Tools in June 2026. Its announcement describes the percentage of citations a site receives for a grounding query across Microsoft Copilot, Bing AI, and select partner experiences. The product does not expose competitor domains or attach a quality score to the citation (Microsoft Bing).

That platform-native metric describes Microsoft's declared surfaces. It must not be relabeled as a cross-engine Share of Citation result.

Interpretation Rules #

Machine Relations treats Share of Citation as a descriptive measurement, not a causal diagnosis.

• Zero means no qualifying citation in the declared sample resolved to the specified brand or domain. It does not mean the brand is absent from every AI answer.
• A higher share means the subject occupied more of the cited-source pool in that sample. It does not prove that the source was more accurate, more authoritative, or more persuasive, and it can rise because competing sources lost citations rather than because the subject gained any.
• A change between windows is an observed difference. The metric alone does not identify what caused it.
• A competitor comparison is valid only when every subject uses the same query set, engines, dates, and citation rule.
• A citation does not prove claim support. That requires a separate claim-source evaluation.
• A citation does not prove recommendation. Recommendation is a separate answer-level classification.

Machine Relations publishes no universal good, bad, leader, or maturity benchmark for Share of Citation. Any threshold used in an internal program is labeled as a program rule, not an observed industry standard.

---

Where It Fits in Machine Relations #

Share of Citation belongs to the measurement layer of the MR Stack because it records an output observation. The stack placement is a Machine Relations taxonomy, not evidence that any upstream layer caused the measured share.

Entity records, publishing activity, earned coverage, content structure, and distribution may be investigated as hypotheses when the share changes. Share of Citation alone cannot rank those hypotheses, assign causality, prescribe an intervention, or predict the result of one.

Common Measurement Errors #

1. Do not count mentions as citations. A linked or otherwise attributable source event satisfies the declared citation rule.
2. Do not hide the denominator. Publish the eligible-answer count with the percentage.
3. Do not combine engines without preserving per-engine results. A blend is an additional view, not a replacement for its components.
4. Do not change the query set between windows without marking the break. A changed sampling frame creates a different series.
5. Do not count several links in one answer as several answer-level citation events. The answer-level numerator is binary for each specified brand or domain.
6. Do not treat citation presence as evidentiary support. Grade the cited page against the associated claim separately.
7. Do not treat citation presence as recommendation. Classify recommendation separately from attribution.
8. Do not infer a cause from movement. The metric reports what changed, not why.
9. Do not generalize a platform-native result. Bing Citation Share does not represent engines outside Microsoft's declared surfaces.

FAQ #

Who coined Share of Citation? #

Jaxon Parrott introduced Share of Citation in Machine Relations measurement work. Microsoft later added its Citation Share metric to Bing Webmaster Tools.

Is Share of Citation the same as share of voice? #

No. Under the Machine Relations definitions, Share of Citation divides the subject's qualifying citations by all qualifying citations in the sampled answers. Share-of-voice measures count brand mentions against other brands' mentions, a different declared observation and denominator.

What is a good Share of Citation? #

Machine Relations publishes no universal benchmark. Compare shares only across subjects measured under the same protocol or across comparable windows in the same series, and never against a figure whose denominator was answers rather than citations.

How often should it be measured? #

Use a declared cadence and keep the sampling contract stable. Machine Relations does not claim a universally correct interval.

Does Bing Webmaster Tools report Citation Share? #

Yes. Microsoft added Citation Share to Bing Webmaster Tools in June 2026 for its declared Copilot, Bing AI, and partner surfaces.

Does a citation mean the cited page supports the answer? #

No. Citation presence and claim support are separate observations. Use Answer-Source Fidelity or another declared claim-source grading method to test support.

Does a citation mean the engine recommends the brand? #

No. An answer can cite a source without recommending the source's brand. Recommendation requires a separate classification.

Can the measurement be automated? #

Collection and calculation can be automated when the system preserves raw answers, citations, run metadata, and the declared counting rule. Query design and protocol changes must remain explicit in the resulting series.