Share of Recommendation: The Metric That Matters in AI Search

Share of voice tells you how loud you are. In AI search, loud is not the contest. The number that decides revenue is your share of recommendation: of every brand an engine names when your buyer asks what to use, what fraction is you. Not how many times you were mentioned. Not how much traffic bounced back to your site. One question, per engine: when the machine hands the buyer a shortlist, is your name on it, and how much of the list is yours?

Volume was the old scoreboard

Share of voice was built for a world of 10 blue links, where being mentioned more meant being found more. It measured presence and let you infer preference. That inference held while a human did the choosing after the click.

The engine broke the chain. It reads everything, then names 3 or 4. Mention volume now tells you how much raw material exists about your brand, not whether the machine selected you from it. In a category with 50 brands, the engine writes 3 into the answer and drops the other 47. A brand can dominate the mention count and still be the one the model reads past on its way to naming a competitor. High share of voice, zero share of recommendation, is the most common blind spot we find. The dashboard is green and the answer names someone else.

Traffic died at the zero-click line

The second habit is measuring traffic, and traffic is now the wrong meter for a different reason: the buyer got their answer inside the engine and never arrived. The recommendation happened. The click did not. If you wait for the session to show up in analytics, you are counting the survivors of a decision that was already made upstream, where you were either named or you were not.

Share of recommendation catches the decision at the point it is made. It reads the answer, not the referral. It asks what the engine told the buyer before any click could exist, which is the only place the choice lives now.

The ladder: retrieved, cited, named, recommended

Recommendation is not one event. It is the top of a ladder, and most brands stall on a lower rung while reporting the whole climb as a win. Each rung is a distinct measurement, and the drop-off between them is where the real diagnosis sits.

Read the ladder and the failure modes separate cleanly. Retrieved-but-not-cited is a trust problem. Cited-but-not-named is an entity resolution problem: the machine read you but could not credit you. Named-but-not-recommended is the hardest and the most common: you made the list, and the list named a leader that was not you. Share of recommendation is the only rung the buyer acts on, so it is the only rung worth the top line of the report.

Measure it per engine or do not bother

The instinct is to average across engines into one tidy figure. Do not. The engines read different grounds and reason over them differently, so a single blended score hides the exact thing you need: where you win and where you vanish. The AI citation economy is brutally concentrated¹: Hexagon found 3% of brands capture 71% of AI recommendations, and that concentration is decided per engine, not in aggregate. You can hold 40% share of recommendation in ChatGPT and 4% in Google AI Mode for the same query set. The average, 22%, describes a brand that exists nowhere.

How to read your share of recommendation

The metric is a fraction, and every part of it has to be defined before the number means anything.

1. Fix the query set to 20 or more real buyer questions. Not your brand name. The comparison questions a buyer asks the engine: "best X for Y," "what should I use for Z." Recommendation only counts on questions where the engine is picking.

2. Run each query per engine. Google AI Mode and ChatGPT at minimum, 2 engines that rarely agree. Log the full set of brands named as recommendations in each answer, not just whether you appeared.

3. Compute the fraction. Your recommendations divided by all brand recommendations in that engine's answers, across the query set. If you are named in 5 of 20 answers and each answer names 4 brands, that fraction is your share of recommendation for that engine.

4. Locate yourself on the ladder for the misses. For every query where your share is 0, mark the rung you stalled on: not retrieved, retrieved-not-cited, cited-not-named, named-not-recommended. The rung tells you what to fix.

5. Track the fraction over time, per engine. The citation sets churn week to week. A one-time reading is a snapshot; the metric earns its keep as a trend, because the concentration means a 5-point shift can move you past 3 competitors.

At Ivanooo, Firoz Azees built the instrument that scores this ladder rung by rung and per engine, because a brand that measures mention volume is measuring the thing that stopped predicting the sale, and a model that hands the buyer a generic answer is naming whoever it trusts most, not whoever posted most. Share of recommendation is downstream of distinctiveness: the engine recommends the brand it can tell apart, which is why a voice a model can average into the category never earns the top of the ladder. The metric is the scoreboard. Distinctiveness is how you move it.

If your current dashboard reports mentions, citations, and share of voice, you are measuring the symptom and not the cause. See where your brand sits on the recommendation ladder, per engine: paste your URL, get the read with the evidence. No call required.