How is citation share different from share of voice?
Citation share counts how often AI answers cite a brand as a source, while earned-media share of voice counts how often press coverage mentions it. The two can move together but may diverge in fast-moving categories. The first measures a source link inside an AI answer, and the second measures mentions in published coverage.
How is citation share calculated?
Citation share is calculated by dividing the number of answers that cite the brand by the total number of answers collected, then multiplying by 100. The same formula can be applied per engine, per prompt type and for each competitor. The figures below are illustrative only.
| Item | Illustrative value |
|---|---|
| Prompts in the set | 40 |
| Engines tested | 5 |
| Runs per prompt per engine | 1 |
| Total answers collected | 200 |
| Answers citing the brand | 30 |
| Citation share | 30 / 200 x 100 = 15% |
What are the steps of the methodology?
The methodology has five steps, and each one should be written down so a later audit can repeat it exactly.
- Define the prompts. Build a set that reflects how buyers ask, tagged by intent such as category, comparison and problem prompts.
- Select the engines. Choose the AI engines that matter to the audience and record the version or mode of each.
- Run and capture. Run each prompt the same way, ideally more than once, and save the full answers and links.
- Extract sources. Parse cited URLs and domains, including inline links, and map each to a brand or publisher.
- Normalize and compare. Adjust for differences in prompt counts and engine behavior, then compare with named competitors.
Illustrative methodology note to publish with a report (example only)
- Prompt count, intent mix and date range
- Engines, versions and settings used
- How a citation was defined and counted
- Number of runs per prompt and how repeats were handled
Why does the methodology matter?
The methodology decides whether a citation share figure is defensible, comparable and reportable. Without a documented method the number is anecdotal, because small changes in prompts or engines can move it. A clear method also lets a team compare one audit cycle with the next. AI engines change often and answers can vary between runs, so a citation share figure is a measured snapshot and no outcome is guaranteed.
What are the common failure modes?
The common failures are ones that make the figure unstable or misleading.
- Not normalizing for different prompt counts across engines
- Source extraction that misses inline URL citations
- Calculating share without category competitor benchmarks
- Changing the method between audits, which breaks the time series
Frequently asked questions
What does citation share methodology mean?
It means the documented method for calculating citation share, covering prompt construction, engine selection, source extraction and normalization.
How many prompts are enough?
There is no fixed number. The set should cover the main buyer questions in the category and be large enough that one unusual answer does not swing the result.
Does a mention count as a citation?
Not always. A citation usually means the engine names or links the brand's page as a source, while a mention is the brand appearing in the text. The method should state which one it counts.
Can citation share be compared across engines?
Only with care. Engines show citations differently, and some show none for certain answers, so comparisons should use normalized figures and say how each engine was handled.
Why report the method with the number?
Because readers cannot judge a percentage without knowing the prompts, engines and dates behind it. Publishing the method makes the figure repeatable.
Part of the 5W GEO Knowledge System · By the 5W Editorial Team · Reviewed by a human editor before publication · Updated October 28, 2026