After two years of running GEO programs alongside our research, the same three numbers come up in every defensibility conversation. These are the benchmarks we hand to teams who need to defend GEO spend internally.
Benchmark 1: Tracked prompt count
The number of prompts you instrument continuously. Below the floor, you are guessing; above the ceiling, you are paying for noise.
| Stage | Floor | Working range | Ceiling |
|---|---|---|---|
| Pilot (first 90 days) | 50 | 100–200 | 300 |
| Operational | 200 | 400–800 | 1,200 |
| Multi-brand / agency | 800 | 1,500–3,000 | 5,000 |
Programs that report tracking 10,000+ prompts are almost always tracking variants of the same prompt; the underlying coverage is far smaller and the team is fooling itself with the number.
Benchmark 2: Citation rate
The percentage of tracked prompts where any owned domain is cited at least once across the engines you track. The single most-quoted GEO health number.
| Tier | Citation rate | Read |
|---|---|---|
| Latent | < 5% | The category does not see you. Foundation work needed before any optimisation pays. |
| Developing | 5–15% | Citations exist but are concentrated. Coverage is the bottleneck, not quality. |
| Operational | 15–35% | Most healthy programs sit here for two to three years. |
| Dominant | > 35% | Category-leading. Worth defending, not extending; extension at this level usually trades off concentration risk. |
Most programs in the operational tier overweight chasing 50%+. The marginal cost of moving from 30% to 40% is greater than from 10% to 30%, and the win is rarely commercially worth it.
Benchmark 3: Answer share
Of the prompts where you are cited at all, the percentage where your domain is in the top-3 cited sources for that prompt. The “depth” partner to citation rate’s “breadth”.
A working program targets 40% answer share once citation rate is above 15%. Below that, you are getting cited but only at the margins, and the citation rarely converts to traffic or recall.
How to read the numbers together
Plot citation rate on one axis and answer share on the other. Four quadrants:
- Low rate, low share: GEO has not begun. Foundation work.
- Low rate, high share: A handful of pages dominate a small number of prompts. Concentration risk; broaden the base before extending.
- High rate, low share: Broad presence but shallow. Quality work (making the cited pages better) beats more pages.
- High rate, high share: The state to maintain. Now spend on defending it from competitors who saw the playbook.
How we benchmark
We re-run the underlying numbers each quarter as part of the citation share study. The methodology is documented at /methodology; the engines, prompt set, and dedup rules are the same as the study.
Adjacent reading
- For tactical patterns that move the numbers see publisher playbook.
- For the live data behind the bands above see citation share study.
