Insights/Guide
Share of Answers: The Metric That Replaces Rankings
Share of answers measures the percentage of buyer queries where AI engines name your firm. Here is the full methodology: query selection, sampling, scoring, and reporting.
Share of answers is the percentage of tracked buyer queries where an AI engine names your business in its response. If you track 100 queries and your firm appears in 23 of the answers, your share of answers is 23%. It is the AI-search equivalent of rank tracking — and for recommendation-driven businesses, it is the only number that captures whether AI search is working.
Why rankings stopped being enough
Rank tracking assumes a results page where position determines clicks. AI answers break both assumptions: there are no positions, and there are often no clicks — 68% of U.S. Google searches ended without a click in early 2026, per SparkToro. An AI answer names two to five businesses in prose. You are in it or you are not. Share of answers measures exactly that.
How to build the query set
The metric is only as good as the queries. Three rules:
- Use buyer language, not industry language. Prospects ask "best CPA for a small business in Phoenix," not "top-rated certified public accounting firm." Pull phrasing from intake calls, not keyword tools.
- Cover the intent spread. Direct recommendation queries ("best X in city"), problem queries ("what kind of lawyer do I need for a contract dispute"), and comparison queries ("X firm vs Y firm").
- Fix the set. 100–300 queries, frozen at baseline. Changing the queries between measurements destroys comparability — the most common way this metric gets gamed.
How to sample correctly
AI answers are probabilistic: the same query can produce different answers across sessions. Reliable measurement requires:
- Multiple runs per query — at least three samples per engine per cycle, scored on appearance rate.
- Clean sessions — no logged-in history or memory, which personalizes answers.
- Every engine that matters — ChatGPT, Perplexity, and Google AI at minimum. Citation studies show these engines share as little as 11% of cited domains, so visibility on one says nothing about another.
- A fixed schedule — same cadence, same method, every cycle.
How to score an answer
Not every mention is equal. A workable scoring rubric:
| Appearance | Score |
|---|---|
| Named as the recommendation, or first among options | 1.0 |
| Named in a list of options | 0.5 |
| Mentioned but not recommended (or with caveats) | 0.25 |
| Absent | 0 |
Report both the raw appearance rate and the weighted score. The weighted score is what improves first — engines often add a firm to lists before they lead with it.
What to report monthly
A useful share-of-answers report contains four things: the headline share versus baseline and versus last month; movement by engine (Perplexity typically moves first, ChatGPT last); the competitor share for the same queries; and the specific queries won and lost, with the source citations that drove the change. That last item turns the report into a work plan — every lost query has a traceable cause.
Why this metric is honest
Share of answers resists the two failure modes of marketing measurement. It cannot be inflated by vanity wins — a screenshot of one good answer means nothing against a 300-query baseline. And it cannot hide failure — if the number does not move, the work is not working. That is why we tie our 60-day guarantee to it.
Frequently asked questions
What is a good share of answers?
Most firms start near zero. Benchmarks from agencies doing this work suggest 5–15% on priority queries after six months of disciplined effort, and 30%+ within 12–18 months — enough to be a default recommendation in a market.
Why not just track AI referral traffic instead?
Referral traffic only counts users who click through, and most AI answers produce no click. Share of answers measures influence at the recommendation moment, whether or not a click follows.
How often should share of answers be measured?
Monthly for reporting, on a fixed schedule with identical queries. More frequent sampling is useful for spotting volatility, but month-over-month against a stable query set is the comparison that matters.