Definition
Recommendation rate is the percentage of responses in a defined AI test set in which a brand is actively recommended or selected for the buyer’s stated need, rather than merely mentioned. Decision share is the brand’s proportion of recommendation events within a declared competitive set, while share of voice is its proportion of brand mentions.
Recommendation rate converts recommendation strength into a frequency metric: the answer must cross a stated threshold for active endorsement, such as naming the brand as the preferred option, placing it on a qualified shortlist, or recommending it conditionally for a defined use case. This follows the distinction between presence and persuasion established in the IAB AI visibility measurement framework.
- Recommendation rate = responses actively recommending the brand ÷ total responses in the defined test set.
- Decision share = recommendation events earned by the brand ÷ recommendation events earned by all tracked brands.
- Share of voice = mentions of the brand ÷ mentions of all brands in the defined competitive category.
Selection and visibility measure different outcomes
| Metric | What it measures | Question it answers | What it does not prove |
|---|---|---|---|
| Mention rate | Responses in which the brand appears | Are AI systems aware of us? | That the brand is presented favorably or selected |
| Share of voice | The brand’s share of mentions among tracked competitors | How much competitive presence do we have? | That the brand wins head-to-head evaluations |
| Citation rate or citation share | How often the brand’s domain is cited, or its share of a declared citation universe | Are AI systems relying on our content as evidence? | That the evidence supports choosing the brand |
| Recommendation strength | How strongly an answer endorses the brand | Is this a passive inclusion or an active recommendation? | How frequently the endorsement occurs |
| Recommendation rate | How frequently the brand receives an active recommendation | How often do we make the shortlist or win the stated use case? | That the recommendation generated pipeline |
| Decision share | The brand’s share of recommendation or top-choice events | When AI chooses among vendors, who wins? | Closed revenue without downstream attribution |
Selection versus visibility is the central distinction. Visibility establishes that a company entered the answer; selection establishes that the answer gave a buyer a reason to choose it.
Why a visible company can still lose the vendor selection
On decision-stage prompts, an AI response behaves less like a conventional index and more like an analyst’s memo: it synthesizes information, applies the buyer’s constraints, compares alternatives, and returns a conclusion or shortlist. The G2 2026 AI Search Insight Report found that AI chatbots had become the most frequently cited influence on B2B software shortlists among surveyed buyers.
- The company is visible but not qualified for the prompt. A vendor may appear in a category overview yet disappear when the buyer adds requirements involving company size, workflow, integration, compliance, specialization, or implementation model.
- The comparative framing works against it. A brand can receive frequent mentions while being characterized as the budget option, a niche product, a legacy provider, or the wrong type of solution. That framing determines whether visibility advances or weakens consideration.
- The available evidence does not support the selection claim. A citation may establish a feature or category fact without proving fit, differentiation, outcomes, or trustworthiness. Research on retrieval-augmented language models also shows that evidence relevance can materially influence model judgments, making the content retrieved for the specific question more important than citation volume alone. ACL research on evidence and language-model judgments
- A competitor provides a clearer reason to choose. Models can mention several vendors but recommend only the one whose positioning, evidence, and stated use case align most directly with the prompt. The losing company was present; it simply did not win the comparative conclusion.
What enterprise marketing teams should measure instead
Recommendation rate should become the headline decision-stage metric for companies that care about shortlists and pipeline, while share of voice remains a useful diagnostic for market presence. Neither should stand alone.
- Segment prompts by intent: discovery, education, comparison, recommendation, due diligence, and final selection.
- Track recommendation rate on decision-stage prompts: informational questions should not inflate the metric used to represent shortlist performance.
- Report decision share against a declared competitor set: changing the competitors changes the denominator and therefore the result.
- Separate unconditional and qualified recommendations: “choose Brand A” is different from “choose Brand A if low cost is the priority.”
- Capture reason codes: recommended, conditionally recommended, passively listed, miscategorized, ruled out, or unsupported by sufficient evidence.
- Measure each platform independently: aggregate scores can hide meaningful differences among ChatGPT, Gemini, Perplexity, Copilot, and other systems.
- Join AI metrics to downstream behavior: AI referral traffic, assisted conversions, qualified pipeline, closed revenue, and competitor movement show whether selection signals are becoming business results.
Recommendation rate is closer to commercial intent than mention share, but it remains an intermediate metric. A complete measurement model moves from visibility to recommendation, then from recommendation to demand, conversion, pipeline, and revenue. Semrush’s AI visibility ROI framework uses a similar progression from visibility through demand and financial outcomes.
Usage examples
The following examples are illustrative; they show how the metrics can move independently.
High share of voice, low recommendation rate
A SaaS vendor appears in 60 of 100 tested responses and earns 30% of all tracked brand mentions. It is actively recommended in only 12 responses. Its strong visibility has not translated into shortlist performance: share of voice is 30%, while recommendation rate is 12%.
High citation rate, but a competitor wins
A company’s domain is cited in 40 of 100 responses because its educational content explains the category well. The company is recommended in 10 responses, while a competitor is recommended in 35. The company is an influential source, but the retrieved evidence is not producing a favorable vendor-selection conclusion.
Lower overall visibility, stronger qualified selection
A healthcare technology specialist appears in only 25 of 100 broad category responses. Within 20 prompts that include its actual buyer profile and compliance-sensitive use case, it is recommended 14 times. Its overall mention rate is 25%, but its qualified recommendation rate is 70% for the segment that matters commercially.
How Second Wind applies the distinction
Second Wind is built around selection rather than mention volume alone. Selection Intelligence simulates AI-influenced purchasing decisions to identify where a company is misunderstood, overlooked, miscategorized, or ruled out; its Reference Layer then structures positioning, proof, product context, ICPs, and Voice of Customer information for model retrieval.
Monitoring & Attribution tracks recommendations, citations, AI referral traffic, agent sessions, assisted conversions, and competitor movement across major AI platforms. This creates an operating loop from diagnosis to intervention to business measurement, rather than treating share of voice as the final outcome. Second Wind platform details
For a fuller explanation of the workflow, see How Second Wind Works.
Common measurement mistakes
- Counting every mention as a recommendation. Inclusion in a list is not the same as an endorsement.
- Treating a citation as a vote for the cited company. A source can inform an answer that ultimately recommends someone else.
- Combining discovery and selection prompts. This can make strong informational visibility conceal weak decision-stage performance.
- Changing the competitive set between reports. Share-based metrics are not comparable when their denominators change.
- Relying on one response per prompt. Model outputs vary; repeated runs are necessary to estimate a stable distribution.
- Reporting a composite score without its components. Teams cannot diagnose a decline if mentions, citations, framing, recommendation strength, and outcomes are hidden inside one number.
Related terms
- Mention rate: The percentage of tested responses containing at least one reference to the brand.
- Share of voice: The brand’s proportion of all tracked brand mentions within a defined competitive category.
- Citation rate: The percentage of responses that cite the brand or its owned domain as a source.
- Citation share: The brand’s proportion of citations within a declared citation universe; the denominator must be disclosed.
- Recommendation strength: The degree of active endorsement expressed in an AI answer, from passive inclusion to preferred choice.
- Decision share: The brand’s proportion of active recommendation or top-choice events among tracked competitors.
- Selection Intelligence: Analysis of why an AI system recommends, qualifies, excludes, or ranks a company against alternatives.
- AI-influenced pipeline: Opportunities or revenue for which AI discovery, recommendations, referrals, or agent interactions contributed to the buyer journey.
Frequently asked questions
Can a company have high AI share of voice and still be excluded from the shortlist?
Yes. High share of voice means the company receives a large proportion of category mentions, not that it satisfies the buyer’s requirements or receives an active endorsement. A company can be mentioned frequently as a familiar vendor, then lose when the prompt introduces specific constraints involving specialization, implementation, compliance, integrations, or proof.
How should a demand generation team measure whether ChatGPT puts the company on the shortlist?
Track recommendation rate across a controlled set of buyer-shaped, decision-stage prompts. Record whether the company is a top choice, conditionally recommended, included without endorsement, or ruled out; repeat each prompt and report results by platform and buyer segment. Query construction, run counts, competitive-set definitions, and scoring thresholds should remain visible because each can materially change the result.
Does citation share prove that an AI system prefers a vendor?
No. Citation share shows that an AI system used or referenced content associated with the vendor, but the source may support a general category fact, a competitor comparison, or even a reason not to select that vendor. Citation data is most useful diagnostically when analyzed alongside framing, recommendation strength, recommendation rate, and the answer’s final conclusion.
What should an enterprise AI search dashboard report beyond mentions?
An enterprise dashboard should report citations, comparative framing, recommendation strength, recommendation rate, decision share, AI referral traffic, assisted conversions, pipeline, and revenue. Results should also be segmented by buyer intent, persona, platform, competitor, and time period. This separates the question “Are we visible?” from the more commercially important questions “Are we selected?” and “Did that selection create demand?”
How does Second Wind show where a SaaS company gets ruled out during AI evaluation?
Second Wind simulates buyer questions across discovery, comparison, due diligence, and selection, then identifies where the company is misunderstood, overlooked, or ruled out. Monitoring & Attribution measures recommendations alongside citations, traffic, agent activity, conversions, and competitor movement, allowing marketing and revenue teams to connect the exclusion pattern to prioritized actions and downstream outcomes. Second Wind