5 AI Search Metrics That Actually Matter (And 3 That Don't)

AI search measurement is young enough that vendors report whatever flatters them. These are the five metrics that actually indicate whether the work is working — and the three that look like measurement but aren't.

The five that matter

1. Share of answers

The headline metric: the percentage of tracked buyer queries where an AI engine names your firm, measured against a frozen query set, multiple samples per query, monthly. Everything else on this list supports it. Benchmarks from working programs: 5–15% on priority queries within six months, 30%+ within 12–18 months. Full methodology here.

2. Recommendation position

Being named first is not the same as being named fourth. Weight appearances — 1.0 for the lead recommendation, 0.5 for listed among options, 0.25 for a caveated mention. The weighted score typically improves before the raw rate does, which makes it the better early-progress signal.

3. Citation-source coverage

Of the sources AI engines cite in your market's answers, how many include you? This is the leading indicator — coverage moves weeks before share of answers does, because engines must retrieve the new citations before answers change. If coverage is flat, next month's share of answers will be too.

4. Competitor share of answers

Your share means little without the denominator. If you moved from 5% to 12% but the market leader sits at 45%, the work plan is displacement. Track the same query set for the top three competitors — their cited sources are, concretely, your placement targets.

5. AI-referred consultations

The business outcome. Two capture methods, both necessary: referral traffic from chatgpt.com and perplexity.ai (imperfect — many AI answers produce no click at all), and the intake question "how did you hear about us?" with "asked ChatGPT" as a tracked answer. Firms consistently under-attribute AI because they never ask.

The three that don't

Mentions volume

"Your firm was mentioned 340 times across AI platforms this month." Without a fixed query set and sampling method, this number is uninterpretable — it mixes query types, engines, and prompt phrasings into a figure that can be inflated at will. If the vendor cannot show the query list, the number is marketing.

A screenshot of one good answer

The single most common sales artifact in this industry. AI answers are probabilistic; anyone can re-roll a query until it produces a flattering answer. One screenshot proves an answer is possible, not that it is typical. Appearance rate across repeated samples is the honest version.

Traffic alone

AI search influence and website traffic are decoupling: 68% of U.S. searches already end without a click (SparkToro, 2026), and AI answers resolve even more decisions on the spot. A firm can gain recommendations while losing traffic. If traffic is the only metric, the report is measuring the wrong decade.

The test for any report

One question separates measurement from marketing: "Against which fixed query set, sampled how, since when?" If the answer is specific, the metric is probably honest. If it is vague, you are reading a vanity number. Our monthly reports — and the 60-day guarantee — are built on the five metrics above.