{"id":1244,"date":"2026-07-14T06:36:22","date_gmt":"2026-07-14T06:36:22","guid":{"rendered":"https:\/\/maxaeo.ai\/blog\/ai-recommendation-rate\/"},"modified":"2026-07-14T06:36:22","modified_gmt":"2026-07-14T06:36:22","slug":"ai-recommendation-rate","status":"publish","type":"post","link":"https:\/\/maxaeo.ai\/blog\/ai-recommendation-rate\/","title":{"rendered":"AI Recommendation Rate: Formula and Measurement Guide"},"content":{"rendered":"<p><em>By maxaeo<\/em><\/p>\n<p><strong>AI recommendation rate measures how often an AI assistant actively recommends your brand\u2014not merely whether it mentions you.<\/strong> It separates meaningful endorsements from neutral descriptions, citations, conditional suggestions, exclusions, and complete absence.<\/p>\n<p>That distinction matters because all of these responses contain a brand signal, but only one puts the company into the buyer\u2019s choice set:<\/p>\n<ul>\n<li>\u201cI recommend maxaeo for this use case.\u201d<\/li>\n<li>\u201cmaxaeo is an AI visibility platform.\u201d<\/li>\n<li>\u201cConsider maxaeo if it supports your required markets.\u201d<\/li>\n<li>\u201cmaxaeo would not meet this requirement.\u201d<\/li>\n<li>A maxaeo article appears as a source, but the answer recommends another provider.<\/li>\n<\/ul>\n<p>AI recommendation rate is not yet governed by an industry or ISO standard. This guide therefore defines a transparent measurement method\u2014the <strong>maxaeo EQNXA framework<\/strong>\u2014that another analyst can reproduce from the same prompts and responses.<\/p>\n<h2>What is AI recommendation rate?<\/h2>\n<p><strong>AI recommendation rate is the percentage of eligible AI responses that explicitly endorse a brand for the user\u2019s stated need. The numerator includes direct recommendations, requested-shortlist inclusion, and unambiguous suitability statements. The denominator includes every usable response that had a fair opportunity to recommend the brand, including responses where it was absent.<\/strong><\/p>\n<p>The strict formula is:<\/p>\n<blockquote>\n<p><strong>AI recommendation rate = Explicit recommendations \u00f7 Eligible responses \u00d7 100<\/strong><\/p>\n<\/blockquote>\n<p>If a company receives 18 explicit recommendations across 100 eligible responses, its AI recommendation rate is <strong>18%<\/strong>.<\/p>\n<p>The observation unit should be:<\/p>\n<blockquote>\n<p><strong>Prompt \u00d7 model or AI surface \u00d7 persona\/market \u00d7 run \u00d7 timestamp<\/strong><\/p>\n<\/blockquote>\n<p>That structure prevents results from different models, markets, or collection periods from being treated as interchangeable.<\/p>\n<h3>What does \u201celigible\u201d mean?<\/h3>\n<p>An eligible response comes from a prompt where recommending a provider was a reasonable outcome. Examples include:<\/p>\n<ul>\n<li>\u201cWhat are the best AI visibility platforms for a global SaaS company?\u201d<\/li>\n<li>\u201cWhich tools can monitor brand recommendations across ChatGPT and Gemini?\u201d<\/li>\n<li>\u201cRecommend three platforms for measuring visibility in AI-generated answers.\u201d<\/li>\n<li>\u201cWhat should an enterprise team use to track AI search visibility?\u201d<\/li>\n<\/ul>\n<p>A general definition prompt such as \u201cWhat is generative engine optimization?\u201d is not eligible for the primary rate because it does not ask the model to choose or evaluate providers.<\/p>\n<h3>Why the denominator matters<\/h3>\n<p>The denominator must include eligible responses in which the brand was not mentioned. Otherwise, the calculation becomes:<\/p>\n<blockquote>\n<p><strong>Recommendations among brand mentions<\/strong><\/p>\n<\/blockquote>\n<p>That companion metric can be useful, but it overstates market visibility because it ignores every occasion when the model selected competitors or named no vendors at all.<\/p>\n<h2>How is recommendation rate different from mention rate?<\/h2>\n<p><strong>Mention rate measures presence. Recommendation rate measures endorsement.<\/strong> A brand can be widely recognized yet rarely presented as an appropriate choice.<\/p>\n<table>\n<thead>\n<tr>\n<th>Model language<\/th>\n<th align=\"right\">Mention<\/th>\n<th align=\"right\">Strict recommendation<\/th>\n<th>Interpretation<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u201cMy top choices are A, B, and maxaeo.\u201d<\/td>\n<td align=\"right\">Yes<\/td>\n<td align=\"right\">Yes<\/td>\n<td>Explicit shortlist inclusion<\/td>\n<\/tr>\n<tr>\n<td>\u201cmaxaeo is a strong option for multi-model monitoring.\u201d<\/td>\n<td align=\"right\">Yes<\/td>\n<td align=\"right\">Yes<\/td>\n<td>Clear suitability statement<\/td>\n<\/tr>\n<tr>\n<td>\u201cConsider maxaeo if it supports your required markets.\u201d<\/td>\n<td align=\"right\">Yes<\/td>\n<td align=\"right\">No<\/td>\n<td>Qualified consideration<\/td>\n<\/tr>\n<tr>\n<td>\u201cmaxaeo monitors how AI systems describe brands.\u201d<\/td>\n<td align=\"right\">Yes<\/td>\n<td align=\"right\">No<\/td>\n<td>Neutral description<\/td>\n<\/tr>\n<tr>\n<td>\u201cmaxaeo is not suitable for this requirement.\u201d<\/td>\n<td align=\"right\">Yes<\/td>\n<td align=\"right\">No<\/td>\n<td>Explicit exclusion<\/td>\n<\/tr>\n<tr>\n<td>A maxaeo URL appears only in the citations<\/td>\n<td align=\"right\">No, under the strict policy<\/td>\n<td align=\"right\">No<\/td>\n<td>Source visibility without brand endorsement<\/td>\n<\/tr>\n<tr>\n<td>The response recommends competitors and omits maxaeo<\/td>\n<td align=\"right\">No<\/td>\n<td align=\"right\">No<\/td>\n<td>Eligible absence<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Citation rate answers a different question: whether the AI answer uses or displays your content as evidence. A model can cite a company\u2019s educational article while recommending a competing product.<\/p>\n<p>Prominence is also separate. A brand recommended in first position creates a different impression from one listed eighth, but both qualify as endorsements. Record position with a separate <a href=\"https:\/\/maxaeo.ai\/blog\/ai-mention-prominence\">depth-of-mention framework<\/a>.<\/p>\n<p>A useful measurement funnel is:<\/p>\n<ol>\n<li><strong>Presence:<\/strong> Was the brand named?<\/li>\n<li><strong>Consideration:<\/strong> Was it presented as potentially suitable?<\/li>\n<li><strong>Recommendation:<\/strong> Was it affirmatively endorsed?<\/li>\n<li><strong>Prominence:<\/strong> How prominently did it appear?<\/li>\n<li><strong>Evidence:<\/strong> Was the brand or its content cited?<\/li>\n<li><strong>Accuracy:<\/strong> Were the claims about the brand correct?<\/li>\n<\/ol>\n<p>No single \u201cAI visibility score\u201d can answer all six questions without obscuring important differences.<\/p>\n<h2>What counts as an AI recommendation?<\/h2>\n<p><strong>A response is an explicit recommendation when it affirmatively places the brand in the user\u2019s choice set.<\/strong> Direct advice, inclusion in a requested \u201cbest\u201d list, a \u201cbest for\u201d designation, or a clear statement of suitability qualifies. Positive adjectives alone do not.<\/p>\n<p>The maxaeo EQNXA framework assigns one of six states to each brand-level observation:<\/p>\n<table>\n<thead>\n<tr>\n<th>Code<\/th>\n<th>Classification<\/th>\n<th>Operational rule<\/th>\n<th>Example<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>E<\/strong><\/td>\n<td>Explicit recommendation<\/td>\n<td>The answer recommends the brand, includes it in a requested shortlist, or clearly says it fits the need<\/td>\n<td>\u201cFor cross-model tracking, maxaeo is a strong option.\u201d<\/td>\n<\/tr>\n<tr>\n<td><strong>Q<\/strong><\/td>\n<td>Qualified consideration<\/td>\n<td>The brand may fit, but a meaningful condition, uncertainty, or verification step prevents a clear endorsement<\/td>\n<td>\u201cConsider maxaeo if its regional coverage matches your requirements.\u201d<\/td>\n<\/tr>\n<tr>\n<td><strong>N<\/strong><\/td>\n<td>Neutral reference<\/td>\n<td>The answer names or describes the brand without advising the user to consider it<\/td>\n<td>\u201cmaxaeo is an AI search visibility platform.\u201d<\/td>\n<\/tr>\n<tr>\n<td><strong>X<\/strong><\/td>\n<td>Exclusion<\/td>\n<td>The answer advises against the brand or states that it does not satisfy the requirement<\/td>\n<td>\u201cmaxaeo is not a fit for this use case.\u201d<\/td>\n<\/tr>\n<tr>\n<td><strong>A<\/strong><\/td>\n<td>Absent<\/td>\n<td>The response is eligible and usable, but the brand does not appear<\/td>\n<td>A competitor shortlist that omits maxaeo<\/td>\n<\/tr>\n<tr>\n<td><strong>U<\/strong><\/td>\n<td>Unscorable<\/td>\n<td>No usable answer exists because of a refusal, truncation, error, wrong locale, or collection failure<\/td>\n<td>Empty response or system error<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" style=\"max-width:100%;height:auto\" loading=\"lazy\"  src=\"https:\/\/maxaeo.ai\/blog\/wp-content\/uploads\/2026\/07\/1783976233811-4-33815-1.jpg\" alt=\"AI recommendation rate annotation rubric with explicit, qualified, neutral, exclusion, absent, and unscorable response labels\"><\/figure>\n<h3>Rules for difficult cases<\/h3>\n<p>Use these rules consistently before reviewing performance totals:<\/p>\n<ul>\n<li><strong>Judge the action, not the adjective.<\/strong> \u201cPopular,\u201d \u201cwell-known,\u201d and \u201cestablished\u201d are descriptions. \u201cI recommend\u201d and \u201ca good fit for your requirement\u201d are endorsements.<\/li>\n<li><strong>Interpret lists from the prompt\u2019s context.<\/strong> Inclusion in a requested \u201cbest tools\u201d shortlist is E. Appearance in a historical or comprehensive vendor directory is N.<\/li>\n<li><strong>Treat meaningful conditions as Q.<\/strong> \u201cChoose maxaeo if you need daily tracking\u201d can be E when it identifies a best-fit use case. \u201cConsider maxaeo if it supports your market\u201d is Q because an unresolved condition remains.<\/li>\n<li><strong>Use the final conclusion when language conflicts.<\/strong> If the response praises the product but ultimately advises against it, label X.<\/li>\n<li><strong>Do not infer endorsement from a citation.<\/strong> A source-domain appearance belongs in the citation field unless the user-visible answer also recommends the brand.<\/li>\n<li><strong>Resolve the entity before assigning sentiment.<\/strong> Product names, former names, abbreviations, and similarly named companies require an alias dictionary. If the model refers to the wrong company, flag an entity error rather than crediting a recommendation.<\/li>\n<li><strong>Code each brand independently.<\/strong> A single response may explicitly recommend several companies. Each can receive E.<\/li>\n<\/ul>\n<p>Keep rank, sentiment, citation, factual accuracy, and evidence strength in separate columns. The EQNXA label should answer only one question: <strong>Did the model endorse this brand for the stated need?<\/strong><\/p>\n<h2>Which responses belong in the denominator?<\/h2>\n<p><strong>Include every successfully collected response to an eligible decision prompt, whether the target brand is recommended, mentioned neutrally, excluded, or absent. Exclude only responses that cannot be scored or prompts that did not create a genuine recommendation opportunity.<\/strong><\/p>\n<p>For one target brand:<\/p>\n<blockquote>\n<p><strong>Eligible responses (n) = E + Q + N + X + A<\/strong><\/p>\n<\/blockquote>\n<p>Unscorable responses are excluded from (n), but their count and failure rate must be disclosed.<\/p>\n<table>\n<thead>\n<tr>\n<th>Observation<\/th>\n<th align=\"right\">Primary denominator<\/th>\n<th>Reason<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Requested category shortlist<\/td>\n<td align=\"right\">Include<\/td>\n<td>A provider recommendation is expected<\/td>\n<\/tr>\n<tr>\n<td>\u201cBest tool for\u2026\u201d prompt<\/td>\n<td align=\"right\">Include<\/td>\n<td>The model can select suitable brands<\/td>\n<\/tr>\n<tr>\n<td>Non-branded problem prompt that asks for solutions<\/td>\n<td align=\"right\">Include<\/td>\n<td>A vendor recommendation is possible<\/td>\n<\/tr>\n<tr>\n<td>Competitor shortlist that omits the brand<\/td>\n<td align=\"right\">Include as A<\/td>\n<td>Absence is part of market visibility<\/td>\n<\/tr>\n<tr>\n<td>Neutral brand reference in an eligible answer<\/td>\n<td align=\"right\">Include as N<\/td>\n<td>The response was usable but did not endorse<\/td>\n<\/tr>\n<tr>\n<td>General educational definition<\/td>\n<td align=\"right\">Exclude<\/td>\n<td>No provider decision was requested<\/td>\n<\/tr>\n<tr>\n<td>Prompt that names the target brand<\/td>\n<td align=\"right\">Separate cohort<\/td>\n<td>Measures branded reputation, not unprompted discovery<\/td>\n<\/tr>\n<tr>\n<td>Empty, refused, corrupted, or truncated response<\/td>\n<td align=\"right\">Exclude as U<\/td>\n<td>No reliable classification is possible<\/td>\n<\/tr>\n<tr>\n<td>Wrong language, market, or persona response<\/td>\n<td align=\"right\">Exclude as U and rerun<\/td>\n<td>Collection conditions were not met<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Use non-branded prompts for the headline discovery metric. \u201cShould I use maxaeo?\u201d tests how the model evaluates an already-known company; it does not show whether the brand enters an unprompted choice set. Keep branded and non-branded results separate using a <a href=\"https:\/\/maxaeo.ai\/blog\/branded-vs-non-branded-prompts\">branded-versus-non-branded prompt design<\/a>.<\/p>\n<h2>How do you calculate AI recommendation rate?<\/h2>\n<p>Use the strict binary rate as the headline KPI and report companion metrics beside it.<\/p>\n<p>For (n) eligible observations:<\/p>\n<ul>\n<li><strong>Strict recommendation rate:<\/strong> (E \/ n)<\/li>\n<li><strong>Consideration rate:<\/strong> ((E + Q) \/ n)<\/li>\n<li><strong>Mention rate:<\/strong> ((E + Q + N + X) \/ n)<\/li>\n<li><strong>Qualified consideration rate:<\/strong> (Q \/ n)<\/li>\n<li><strong>Exclusion rate:<\/strong> (X \/ n)<\/li>\n<li><strong>Absence rate:<\/strong> (A \/ n)<\/li>\n<li><strong>Recommendation conversion:<\/strong> (E \/ (E + Q + N + X))<\/li>\n<li><strong>Collection failure rate:<\/strong> (U \/ (n + U))<\/li>\n<\/ul>\n<p>Recommendation conversion means \u201cendorsements among mentions.\u201d It is diagnostically useful, but it should not replace the headline recommendation rate.<\/p>\n<h3>The two-lever decomposition<\/h3>\n<p>A particularly useful diagnostic identity is:<\/p>\n<blockquote>\n<p><strong>Recommendation rate = Mention rate \u00d7 Recommendation conversion<\/strong><\/p>\n<\/blockquote>\n<p>This separates two different problems:<\/p>\n<ol>\n<li><strong>Candidate-set visibility:<\/strong> Does the brand appear at all?<\/li>\n<li><strong>Endorsement conversion:<\/strong> When it appears, does the model recommend it?<\/li>\n<\/ol>\n<p>For example:<\/p>\n<ul>\n<li>Mention rate: 51.8%<\/li>\n<li>Recommendation conversion: 36.2%<\/li>\n<li>Recommendation rate: (51.8% \u00d7 36.2% \u2248 18.75%)<\/li>\n<\/ul>\n<p>A low mention rate calls for stronger entity recognition, category association, and third-party corroboration. A low recommendation conversion means the model recognizes the company but lacks sufficient reasons to endorse it.<\/p>\n<h3>Recommendation rate versus competitive recommendation share<\/h3>\n<p>AI recommendation rate is <strong>not market share<\/strong>. Multiple brands can receive E in the same answer, so competitors\u2019 rates do not have to sum to 100%.<\/p>\n<p>If you need a normalized competitive metric, calculate:<\/p>\n<blockquote>\n<p><strong>Competitive recommendation share = Target brand\u2019s E labels \u00f7 E labels across all tracked brands<\/strong><\/p>\n<\/blockquote>\n<p>Use the same prompt set, brand dictionary, and annotation policy for every competitor. Report recommendation rate beside the share because a brand can gain share while the entire category receives fewer recommendations.<\/p>\n<h3>Weighted recommendation rate<\/h3>\n<p>If some prompts are more commercially important, assign weights before collection:<\/p>\n<blockquote>\n<p><strong>Weighted recommendation rate = \u03a3((w_i \u00d7 E_i)) \u00f7 \u03a3(w_i)<\/strong><\/p>\n<\/blockquote>\n<p>A procurement prompt may deserve more weight than an early educational prompt. The weight should reflect audience value or buying intent\u2014not which prompts produced favorable answers.<\/p>\n<p>Always publish both the weighted and unweighted results. Post-hoc weighting makes the metric easy to manipulate.<\/p>\n<h2>Worked example: 120 AI responses<\/h2>\n<p><strong>In this synthetic example, 120 collected responses produce 112 eligible observations after eight failures. Twenty-one explicit endorsements result in an AI recommendation rate of 18.75%.<\/strong><\/p>\n<p>The dataset demonstrates the calculation; it is not a maxaeo customer result or an industry benchmark.<\/p>\n<table>\n<thead>\n<tr>\n<th>Model<\/th>\n<th align=\"right\">Eligible<\/th>\n<th align=\"right\">E<\/th>\n<th align=\"right\">Q<\/th>\n<th align=\"right\">N<\/th>\n<th align=\"right\">X<\/th>\n<th align=\"right\">A<\/th>\n<th align=\"right\">Strict rate<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>ChatGPT<\/td>\n<td align=\"right\">28<\/td>\n<td align=\"right\">7<\/td>\n<td align=\"right\">5<\/td>\n<td align=\"right\">4<\/td>\n<td align=\"right\">2<\/td>\n<td align=\"right\">10<\/td>\n<td align=\"right\">25.0%<\/td>\n<\/tr>\n<tr>\n<td>Gemini<\/td>\n<td align=\"right\">28<\/td>\n<td align=\"right\">4<\/td>\n<td align=\"right\">6<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">1<\/td>\n<td align=\"right\">14<\/td>\n<td align=\"right\">14.3%<\/td>\n<\/tr>\n<tr>\n<td>Perplexity<\/td>\n<td align=\"right\">28<\/td>\n<td align=\"right\">6<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">5<\/td>\n<td align=\"right\">2<\/td>\n<td align=\"right\">12<\/td>\n<td align=\"right\">21.4%<\/td>\n<\/tr>\n<tr>\n<td>Claude<\/td>\n<td align=\"right\">28<\/td>\n<td align=\"right\">4<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">2<\/td>\n<td align=\"right\">1<\/td>\n<td align=\"right\">18<\/td>\n<td align=\"right\">14.3%<\/td>\n<\/tr>\n<tr>\n<td><strong>Total<\/strong><\/td>\n<td align=\"right\"><strong>112<\/strong><\/td>\n<td align=\"right\"><strong>21<\/strong><\/td>\n<td align=\"right\"><strong>17<\/strong><\/td>\n<td align=\"right\"><strong>14<\/strong><\/td>\n<td align=\"right\"><strong>6<\/strong><\/td>\n<td align=\"right\"><strong>54<\/strong><\/td>\n<td align=\"right\"><strong>18.75%<\/strong><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The companion calculations are:<\/p>\n<ul>\n<li><strong>Consideration rate:<\/strong> ((21 + 17) \/ 112 = 33.9%)<\/li>\n<li><strong>Mention rate:<\/strong> ((21 + 17 + 14 + 6) \/ 112 = 51.8%)<\/li>\n<li><strong>Exclusion rate:<\/strong> (6 \/ 112 = 5.4%)<\/li>\n<li><strong>Absence rate:<\/strong> (54 \/ 112 = 48.2%)<\/li>\n<li><strong>Recommendation conversion:<\/strong> (21 \/ 58 = 36.2%)<\/li>\n<li><strong>Collection failure rate:<\/strong> (8 \/ 120 = 6.7%)<\/li>\n<\/ul>\n<p>A 95% Wilson interval for 21 successes in 112 observations is approximately <strong>12.6% to 27.0%<\/strong>. The interval makes clear that 18.75% is an estimate, not a permanent property of the brand.<\/p>\n<p>If repeated runs share the same underlying prompts, the observations are correlated. For consequential decisions, calculate uncertainty by bootstrapping complete prompt clusters rather than treating every run as independent.<\/p>\n<h3>How weighting changes the answer<\/h3>\n<p>Suppose 16 of 50 high-intent observations receive E, compared with five of 62 problem-awareness observations. The unweighted result remains 18.75%.<\/p>\n<p>If high-intent prompts were assigned three times the weight before collection:<\/p>\n<blockquote>\n<p>((16 \u00d7 3 + 5) \u00f7 (50 \u00d7 3 + 62) = 25.0%)<\/p>\n<\/blockquote>\n<p>Both results are mathematically valid:<\/p>\n<ul>\n<li><strong>18.75%<\/strong> describes the collected response set.<\/li>\n<li><strong>25.0%<\/strong> describes the predefined intent-weighted portfolio.<\/li>\n<\/ul>\n<p>The report must state which question each number answers.<\/p>\n<h2>How should the prompt sample be designed?<\/h2>\n<p><strong>A valid prompt portfolio represents real buyer decisions, not a collection of phrasings selected because they mention the brand.<\/strong> Define the audience, market, category, constraints, and decision stage before generating prompts.<\/p>\n<h3>A practical starter portfolio<\/h3>\n<p>The following 50-prompt structure is a defensible starting point, not a universal standard:<\/p>\n<table>\n<thead>\n<tr>\n<th>Prompt group<\/th>\n<th align=\"right\">Unique prompts<\/th>\n<th>Purpose<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Category shortlists<\/td>\n<td align=\"right\">15<\/td>\n<td>Test whether the brand enters broad candidate sets<\/td>\n<\/tr>\n<tr>\n<td>Job-to-be-done prompts<\/td>\n<td align=\"right\">15<\/td>\n<td>Measure recommendation for specific outcomes<\/td>\n<\/tr>\n<tr>\n<td>Constraint prompts<\/td>\n<td align=\"right\">10<\/td>\n<td>Test requirements such as market, integration, scale, or budget<\/td>\n<\/tr>\n<tr>\n<td>Alternative and comparison prompts<\/td>\n<td align=\"right\">5<\/td>\n<td>Measure competitive substitution<\/td>\n<\/tr>\n<tr>\n<td>Objection and risk prompts<\/td>\n<td align=\"right\">5<\/td>\n<td>Surface reasons for qualified or exclusionary answers<\/td>\n<\/tr>\n<tr>\n<td><strong>Total<\/strong><\/td>\n<td align=\"right\"><strong>50<\/strong><\/td>\n<td><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Running 50 prompts across four models twice creates 400 observations. Those observations are not fully independent because each prompt is repeated, so reports should show both <strong>total responses and unique prompt count<\/strong>.<\/p>\n<p>Useful non-branded prompt templates include:<\/p>\n<ul>\n<li>\u201cRecommend three tools for <code>[job]<\/code> used by <code>[persona]<\/code>.\u201d<\/li>\n<li>\u201cWhat are the best <code>[category]<\/code> platforms for a company operating in <code>[market]<\/code>?\u201d<\/li>\n<li>\u201cWhich providers meet <code>[constraint 1]<\/code> and <code>[constraint 2]<\/code>?\u201d<\/li>\n<li>\u201cWhat alternatives should a buyer evaluate if <code>[current approach]<\/code> is insufficient?\u201d<\/li>\n<li>\u201cWhich solution would you choose for <code>[use case]<\/code>, and why?\u201d<\/li>\n<\/ul>\n<p>Do not merely paraphrase one favorable prompt 50 times. Prompt diversity should change the underlying decision, not just the wording.<\/p>\n<h3>Separate prompt cohorts<\/h3>\n<p>Maintain at least three cohorts:<\/p>\n<ol>\n<li><strong>Non-branded discovery prompts:<\/strong> Can the brand enter an unprompted shortlist?<\/li>\n<li><strong>Branded reputation prompts:<\/strong> How does the model evaluate the company when named?<\/li>\n<li><strong>Stable control prompts:<\/strong> Do model-wide changes affect unrelated categories or requirements?<\/li>\n<\/ol>\n<p>Keep a stable core cohort for trend reporting and a smaller exploratory cohort for new buyer language. Do not merge exploratory prompts into the historical baseline without versioning the change.<\/p>\n<h2>How do you collect reproducible AI responses?<\/h2>\n<p><strong>Reproducible tracking requires frozen prompts, documented collection conditions, complete raw responses, and stable scoring rules.<\/strong> A percentage without response-level evidence cannot be audited.<\/p>\n<p>Record these fields for every observation:<\/p>\n<table>\n<thead>\n<tr>\n<th>Field<\/th>\n<th>Why it is required<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Prompt ID and version<\/td>\n<td>Detects changes to the test<\/td>\n<\/tr>\n<tr>\n<td>Full prompt<\/td>\n<td>Preserves the exact decision context<\/td>\n<\/tr>\n<tr>\n<td>Model and visible version<\/td>\n<td>Separates different AI systems and updates<\/td>\n<\/tr>\n<tr>\n<td>AI surface<\/td>\n<td>Distinguishes chat, search, overview, or API experiences<\/td>\n<\/tr>\n<tr>\n<td>Market, locale, and language<\/td>\n<td>Controls geographic and linguistic variation<\/td>\n<\/tr>\n<tr>\n<td>Browsing or retrieval state<\/td>\n<td>Retrieval can change sources and recommendations<\/td>\n<\/tr>\n<tr>\n<td>Account and personalization state<\/td>\n<td>Logged-in history may affect output<\/td>\n<\/tr>\n<tr>\n<td>Timestamp and run number<\/td>\n<td>Supports trend and volatility analysis<\/td>\n<\/tr>\n<tr>\n<td>Full response<\/td>\n<td>Provides auditable evidence<\/td>\n<\/tr>\n<tr>\n<td>Citations and destination URLs<\/td>\n<td>Separates evidence visibility from endorsement<\/td>\n<\/tr>\n<tr>\n<td>EQNXA label<\/td>\n<td>Stores the recommendation classification<\/td>\n<\/tr>\n<tr>\n<td>Evidence excerpt<\/td>\n<td>Shows the language supporting the label<\/td>\n<\/tr>\n<tr>\n<td>Coder and codebook version<\/td>\n<td>Makes annotation traceable<\/td>\n<\/tr>\n<tr>\n<td>Entity-match confidence<\/td>\n<td>Flags potential brand-name collisions<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>For manual tracking, these fields fit in a spreadsheet. An AI visibility tool can automate collection and pre-classification, but borderline E-versus-Q and N-versus-X cases still need documented quality control.<\/p>\n<p>Model recommendations can move after a model or retrieval update even when the website has not changed. Preserve model-level baselines and use a <a href=\"https:\/\/maxaeo.ai\/blog\/how-model-updates-affect-ai-visibility\">model-update change log<\/a> before attributing every movement to marketing activity.<\/p>\n<h2>How do you build a reliable annotation workflow?<\/h2>\n<p><strong>A reliable annotation workflow uses a versioned codebook, hides aggregate performance during labeling, independently reviews a sample, and preserves every disagreement.<\/strong><\/p>\n<ol>\n<li>\n<p><strong>Freeze the codebook.<\/strong> Define E, Q, N, X, A, and U before reviewing the response set.<\/p>\n<\/li>\n<li>\n<p><strong>Create boundary examples.<\/strong> Include ranked lists, conditional language, conflicting conclusions, citation-only appearances, and entity collisions.<\/p>\n<\/li>\n<li>\n<p><strong>Annotate at brand level.<\/strong> One multi-brand response may create separate observations for several tracked companies.<\/p>\n<\/li>\n<li>\n<p><strong>Blind coders to totals.<\/strong> Reviewers should not see whether a borderline decision will improve the dashboard.<\/p>\n<\/li>\n<li>\n<p><strong>Double-code a quality-control sample.<\/strong> Begin with 20% when introducing a codebook or new market.<\/p>\n<\/li>\n<li>\n<p><strong>Adjudicate disagreements.<\/strong> Preserve both initial labels, the final label, the rule applied, and the reason for the decision.<\/p>\n<\/li>\n<li>\n<p><strong>Recalibrate recurring boundaries.<\/strong> If E-versus-Q disagreements recur, add examples and repeat the affected sample.<\/p>\n<\/li>\n<li>\n<p><strong>Compute from frozen data.<\/strong> Publish the codebook version, collection window, numerator, denominator, unique prompt count, and failures with the result.<\/p>\n<\/li>\n<\/ol>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" style=\"max-width:100%;height:auto\" loading=\"lazy\"  src=\"https:\/\/maxaeo.ai\/blog\/wp-content\/uploads\/2026\/07\/1783976233811-4-33815-2.jpg\" alt=\"AI recommendation rate dashboard showing model segments, eligible response counts, confidence intervals, and links to source evidence\"><\/figure>\n<h3>How much reviewer agreement is enough?<\/h3>\n<p>Report both <strong>raw agreement<\/strong> and <strong>Cohen\u2019s kappa<\/strong>. Raw agreement is intuitive, while kappa adjusts for chance agreement. Kappa can be sensitive to label prevalence, so neither measure should be interpreted alone; this <a href=\"https:\/\/www.ncbi.nlm.nih.gov\/pmc\/articles\/PMC3900052\/\" target=\"_blank\" rel=\"noopener\">peer-reviewed overview of inter-rater reliability<\/a> explains the limitation.<\/p>\n<p>A practical internal publishing gate is:<\/p>\n<ul>\n<li>At least <strong>90% raw agreement<\/strong><\/li>\n<li>Cohen\u2019s <strong>\u03ba of 0.80 or higher<\/strong><\/li>\n<li>No unresolved systematic disagreement over E versus Q<\/li>\n<\/ul>\n<p>These are operating thresholds, not universal statistical laws. A lower score should trigger calibration, not selective removal of inconvenient responses.<\/p>\n<h2>How many responses are needed?<\/h2>\n<p><strong>Sample size depends on the precision required, the expected rate, and the amount of clustering. There is no single minimum that fits every prompt portfolio.<\/strong><\/p>\n<p>For a simple independent proportion, an approximate starting formula is:<\/p>\n<blockquote>\n<p><strong>(n = z^2 \u00d7 p(1-p) \/ m^2)<\/strong><\/p>\n<\/blockquote>\n<p>Where:<\/p>\n<ul>\n<li>(z = 1.96) for an approximate 95% confidence level<\/li>\n<li>(p) is the expected recommendation rate<\/li>\n<li>(m) is the desired margin of error<\/li>\n<\/ul>\n<p>At a rate near 20%:<\/p>\n<table>\n<thead>\n<tr>\n<th align=\"right\">Independent observations<\/th>\n<th align=\"right\">Approximate 95% margin<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td align=\"right\">100<\/td>\n<td align=\"right\">\u00b17.8 percentage points<\/td>\n<\/tr>\n<tr>\n<td align=\"right\">250<\/td>\n<td align=\"right\">\u00b15.0 percentage points<\/td>\n<\/tr>\n<tr>\n<td align=\"right\">400<\/td>\n<td align=\"right\">\u00b13.9 percentage points<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>These are approximations. Repeated runs of the same prompt provide information about model volatility, but they do not add as much independent evidence as new prompt clusters.<\/p>\n<p>For practical reporting:<\/p>\n<ul>\n<li>Show the total number of responses.<\/li>\n<li>Show the number of unique prompts.<\/li>\n<li>Show the number of models, markets, and runs.<\/li>\n<li>Use Wilson intervals for a simple binary rate.<\/li>\n<li>Use prompt-cluster bootstrap intervals when prompts are repeated.<\/li>\n<li>Avoid ranking small segments on percentage alone.<\/li>\n<\/ul>\n<p>Two recommendations from five responses should never be presented as stronger evidence than 40 from 200 without displaying the sample sizes and uncertainty.<\/p>\n<h2>What is a good AI recommendation rate?<\/h2>\n<p><strong>There is no credible universal benchmark. A good AI recommendation rate is one that improves against a frozen, relevant baseline without increasing exclusions, measurement failures, or dependence on branded prompts.<\/strong><\/p>\n<p>The result depends on:<\/p>\n<ul>\n<li>Category maturity and competitive density<\/li>\n<li>Prompt intent and buyer constraints<\/li>\n<li>Brand awareness and product maturity<\/li>\n<li>Market and language<\/li>\n<li>Model and AI surface<\/li>\n<li>Whether the prompt names the brand<\/li>\n<li>Denominator and annotation rules<\/li>\n<li>Collection date and retrieval conditions<\/li>\n<\/ul>\n<p>A 15% rate across non-branded enterprise shortlists may be more valuable than a 60% rate on prompts that already contain the company name.<\/p>\n<p>Use three comparisons:<\/p>\n<ol>\n<li><strong>Historical:<\/strong> Did the rate improve on the same stable prompt cohort?<\/li>\n<li><strong>Segment:<\/strong> Where is the company strong or absent?<\/li>\n<li><strong>Competitive:<\/strong> How does its rate compare with named competitors under the same method?<\/li>\n<\/ol>\n<p>Treat public benchmarks cautiously unless they disclose prompts, model versions, markets, run counts, denominator rules, and labeling criteria.<\/p>\n<h2>Which segments should appear in a report?<\/h2>\n<p><strong>Always show the overall rate with its numerator, denominator, and uncertainty, then segment by factors capable of changing the recommendation.<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Segment<\/th>\n<th>Question answered<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Model or AI surface<\/td>\n<td>Which systems recommend the brand?<\/td>\n<\/tr>\n<tr>\n<td>Prompt intent<\/td>\n<td>Does the brand appear during learning, comparison, or selection?<\/td>\n<\/tr>\n<tr>\n<td>Persona<\/td>\n<td>Is the product recommended to the intended buyer?<\/td>\n<\/tr>\n<tr>\n<td>Market and language<\/td>\n<td>Does visibility change by locale or availability?<\/td>\n<\/tr>\n<tr>\n<td>Use case<\/td>\n<td>Which jobs does the brand credibly own?<\/td>\n<\/tr>\n<tr>\n<td>Constraint<\/td>\n<td>Which requirements create qualified or exclusionary answers?<\/td>\n<\/tr>\n<tr>\n<td>Funnel stage<\/td>\n<td>Does recognition survive into final shortlisting?<\/td>\n<\/tr>\n<tr>\n<td>Time window<\/td>\n<td>Is a change persistent or short-lived?<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>For each segment, show:<\/p>\n<ul>\n<li>E count<\/li>\n<li>Eligible (n)<\/li>\n<li>Recommendation rate<\/li>\n<li>Q, X, and A rates<\/li>\n<li>Confidence interval<\/li>\n<li>Change from the stable baseline<\/li>\n<\/ul>\n<p>Do not average model-level percentages unless their sample sizes and intended weights are equal. Aggregate the underlying counts or apply declared weights.<\/p>\n<h2>How should changes over time be interpreted?<\/h2>\n<p><strong>A change in recommendation rate can result from market evidence, model updates, prompt changes, collection conditions, annotation changes, or genuine optimization. Attribution requires controlling these alternatives.<\/strong><\/p>\n<p>Use these controls:<\/p>\n<ul>\n<li>Maintain a stable core prompt set.<\/li>\n<li>Version every prompt and codebook change.<\/li>\n<li>Log model, surface, retrieval state, locale, and collection time.<\/li>\n<li>Mark product launches, content releases, PR activity, and entity corrections.<\/li>\n<li>Compare rolling windows instead of isolated daily results.<\/li>\n<li>Inspect response evidence behind every material movement.<\/li>\n<li>Use control prompts when evaluating an intervention.<\/li>\n<\/ul>\n<p>If a lift appears only in newly introduced prompts, treat it as a hypothesis. If it persists across a stable cohort, multiple runs, relevant models, and unchanged annotation rules, it becomes stronger evidence.<\/p>\n<p>Daily AI search monitoring is useful for detecting breakpoints. Weekly or 28-day windows are usually more appropriate for strategic interpretation because individual model outputs can vary between runs.<\/p>\n<h2>How do you improve AI recommendation rate?<\/h2>\n<p><strong>Use the label distribution to identify the limiting stage. Absence, neutral recognition, qualified consideration, and exclusion require different interventions.<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Pattern<\/th>\n<th>Diagnosis<\/th>\n<th>First action<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>High A<\/td>\n<td>The brand is not entering the candidate set<\/td>\n<td>Strengthen entity clarity, category association, third-party corroboration, and answer-ready coverage<\/td>\n<\/tr>\n<tr>\n<td>High N<\/td>\n<td>The model recognizes the brand but lacks reasons to select it<\/td>\n<td>Publish verifiable differentiators, best-fit use cases, limitations, and comparison evidence<\/td>\n<\/tr>\n<tr>\n<td>High Q<\/td>\n<td>An unresolved condition blocks endorsement<\/td>\n<td>Extract repeated caveats and publish evidence that resolves them<\/td>\n<\/tr>\n<tr>\n<td>High X<\/td>\n<td>The model sees the brand as unsuitable<\/td>\n<td>Determine whether the reason is accurate, outdated, or attached to the wrong entity<\/td>\n<\/tr>\n<tr>\n<td>High E, weak prominence<\/td>\n<td>The brand is recommended but rarely leads the list<\/td>\n<td>Strengthen evidence for the specific criteria buyers prioritize<\/td>\n<\/tr>\n<tr>\n<td>High E, weak business response<\/td>\n<td>AI endorsement is not translating downstream<\/td>\n<td>Audit landing-page continuity, audience fit, offer clarity, and attribution<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h3>If the brand is absent<\/h3>\n<p>Improve the evidence that connects the entity to its category and use cases:<\/p>\n<ul>\n<li>Use one consistent organization and product name.<\/li>\n<li>State clearly what the product does, who it serves, and where it is available.<\/li>\n<li>Align Organization, Product, Person, and Article entities where relevant.<\/li>\n<li>Earn independent descriptions from credible industry sources.<\/li>\n<li>Publish pages answering the exact problems and constraints in the prompt set.<\/li>\n<li>Correct company-name collisions and obsolete profiles.<\/li>\n<\/ul>\n<p>If an AI system confuses the company with a similarly named entity, use an <a href=\"https:\/\/maxaeo.ai\/blog\/brand-name-collision-ai-search\">entity-disambiguation playbook<\/a> before producing more generic content.<\/p>\n<h3>If mentions are neutral<\/h3>\n<p>Neutral recognition means the model knows the company exists but lacks evidence for preference. Add:<\/p>\n<ul>\n<li>Specific best-fit and non-fit statements<\/li>\n<li>Verifiable feature and coverage details<\/li>\n<li>Transparent pricing or evaluation criteria where appropriate<\/li>\n<li>Original comparisons based on buyer requirements<\/li>\n<li>Named use cases with constraints<\/li>\n<li>First-party research or documented methodology<\/li>\n<li>Independent proof that supports important claims<\/li>\n<\/ul>\n<p>Google\u2019s <a href=\"https:\/\/developers.google.com\/search\/docs\/fundamentals\/creating-helpful-content\" target=\"_blank\" rel=\"noopener\">people-first content guidance<\/a> recommends original information, substantial value, clear sourcing, and content designed for an existing audience. Those principles also make passages easier for AI systems to interpret and cite.<\/p>\n<h3>If qualified consideration is high<\/h3>\n<p>Collect the exact caveat attached to every Q label. Group repeated conditions such as:<\/p>\n<ul>\n<li>\u201cIf it supports your target models\u201d<\/li>\n<li>\u201cIf enterprise security is available\u201d<\/li>\n<li>\u201cIf the product operates in your market\u201d<\/li>\n<li>\u201cIf pricing fits a small team\u201d<\/li>\n<li>\u201cIf the integration supports your stack\u201d<\/li>\n<\/ul>\n<p>A repeated condition is a content and evidence requirement\u2014not merely a sentiment problem. Resolve it with a clear product page, documentation, independent validation, or an honest statement that the requirement is not supported.<\/p>\n<h3>If exclusions are high<\/h3>\n<p>Separate four causes:<\/p>\n<ol>\n<li><strong>Correct non-fit:<\/strong> The product genuinely does not meet the requirement.<\/li>\n<li><strong>Outdated information:<\/strong> The model describes an old feature set, market, or price.<\/li>\n<li><strong>Unsupported inference:<\/strong> The answer makes a claim without reliable evidence.<\/li>\n<li><strong>Entity collision:<\/strong> The answer attributes another company\u2019s characteristics to the brand.<\/li>\n<\/ol>\n<p>Do not try to erase legitimate limitations. Accurate non-fit language improves trust and prevents unqualified demand. Correct false or obsolete claims with specific, corroborated evidence.<\/p>\n<h2>How should leaders report the metric?<\/h2>\n<p><strong>Report AI recommendation rate as evidence of AI-mediated consideration, not as proof of revenue or causal impact.<\/strong><\/p>\n<p>A defensible executive statement is:<\/p>\n<blockquote>\n<p>\u201cAcross 200 eligible, non-branded category responses, the brand received 48 explicit recommendations: a 24% recommendation rate. That is six percentage points above the frozen baseline. The increase appeared across three models and 31 additional responses; sample sizes, confidence intervals, failures, and source evidence are attached.\u201d<\/p>\n<\/blockquote>\n<p>This is a reporting template, not a maxaeo benchmark.<\/p>\n<p>A complete report should disclose:<\/p>\n<ul>\n<li>Collection window<\/li>\n<li>Prompt cohort and version<\/li>\n<li>Models and AI surfaces<\/li>\n<li>Markets and personas<\/li>\n<li>Runs per prompt<\/li>\n<li>E numerator and eligible denominator<\/li>\n<li>Unique prompt count<\/li>\n<li>Q, N, X, A, and U counts<\/li>\n<li>Weighted and unweighted rates<\/li>\n<li>Confidence interval<\/li>\n<li>Codebook version and reviewer agreement<\/li>\n<li>Representative response evidence<\/li>\n<\/ul>\n<p>Connect the metric to commercial outcomes through AI referral sessions, self-reported attribution, assisted conversions, branded search changes, demo-request language, and win-loss interviews. Recommendation rate alone does not show that an endorsement caused a purchase.<\/p>\n<p>For a broader executive dashboard, place it beside the other <a href=\"https:\/\/maxaeo.ai\/blog\/ai-search-visibility-metrics\">AI search visibility metrics<\/a> rather than combining all signals into one opaque score.<\/p>\n<h2>Common measurement mistakes<\/h2>\n<p><strong>The most serious errors inflate the numerator, shrink the denominator, or change the test between periods.<\/strong><\/p>\n<p>Avoid these mistakes:<\/p>\n<ul>\n<li><strong>Counting every mention as an endorsement.<\/strong> Neutral references and exclusions are not recommendations.<\/li>\n<li><strong>Using mentions as the denominator.<\/strong> This removes every eligible response where the brand was absent.<\/li>\n<li><strong>Treating citations as advocacy.<\/strong> An answer may cite your article while recommending a competitor.<\/li>\n<li><strong>Mixing branded and non-branded prompts.<\/strong> They measure different stages of visibility.<\/li>\n<li><strong>Treating all lists as endorsements.<\/strong> A requested \u201cbest tools\u201d list differs from a factual vendor directory.<\/li>\n<li><strong>Ignoring multi-brand answers.<\/strong> Several brands can receive E in one response.<\/li>\n<li><strong>Changing prompts after seeing weak results.<\/strong> The new prompt set no longer supports a historical comparison.<\/li>\n<li><strong>Pooling incompatible markets.<\/strong> A provider may be available in one country and unsuitable in another.<\/li>\n<li><strong>Hiding failures.<\/strong> Changes in error or refusal rates can alter the denominator.<\/li>\n<li><strong>Applying weights after collection.<\/strong> Post-hoc weights invite a preferred result.<\/li>\n<li><strong>Reporting percentages without counts.<\/strong> Small samples create false precision.<\/li>\n<li><strong>Averaging percentages with unequal sample sizes.<\/strong> Aggregate counts or use predefined weights.<\/li>\n<li><strong>Treating recommendation as conversion.<\/strong> AI output is an exposure signal, not revenue attribution.<\/li>\n<li><strong>Ignoring inaccurate exclusions.<\/strong> The cause may be outdated information or an entity collision.<\/li>\n<li><strong>Automating labels without audits.<\/strong> Small classification biases compound across thousands of responses.<\/li>\n<\/ul>\n<h2>Frequently asked questions<\/h2>\n<h3>Is AI recommendation rate the same as mention rate?<\/h3>\n<p>No. Mention rate counts eligible responses that name the brand, regardless of context. AI recommendation rate counts only responses that explicitly endorse the brand for the user\u2019s stated need.<\/p>\n<p>In the worked example, mention rate is 51.8% while recommendation rate is 18.75%. The gap consists of qualified, neutral, and exclusionary mentions.<\/p>\n<h3>Should qualified consideration count as a recommendation?<\/h3>\n<p>Not in the strict headline metric. Keep Q separate so \u201cconsider this if\u2026\u201d cannot be silently upgraded into an unconditional endorsement.<\/p>\n<p>Report ((E + Q) \/ n) as consideration rate. Then analyze the conditions attached to Q responses to identify the evidence, feature, market constraint, or trust signal blocking a clear recommendation.<\/p>\n<h3>What is a good AI recommendation rate?<\/h3>\n<p>There is no reliable universal benchmark. Compare results with a frozen internal baseline and competitors measured with the same prompts, models, markets, and annotation policy.<\/p>\n<p>A lower rate on relevant non-branded buying prompts can be more meaningful than a high rate on prompts that already name the company.<\/p>\n<h3>How many responses do I need?<\/h3>\n<p>Choose the sample based on the precision required. At an observed rate near 20%, 100 independent observations have an approximate 95% margin of error of eight percentage points; 400 reduce it to about four points.<\/p>\n<p>Because repeated prompt runs are correlated, also report unique prompt count and use prompt-cluster intervals for consequential decisions.<\/p>\n<h3>How often should AI recommendation rate be measured?<\/h3>\n<p>Use frequent collection for detection and longer windows for decisions. Daily monitoring can reveal sudden model or retrieval changes, while weekly and 28-day rolling results reduce the risk of reacting to individual outputs.<\/p>\n<p>Keep a frozen baseline cohort and record model, prompt, and codebook changes.<\/p>\n<h3>Can AI recommendation rate be negative?<\/h3>\n<p>No. The strict rate ranges from 0% to 100% because it is the share of eligible responses labeled E.<\/p>\n<p>Do not subtract exclusions from recommendations. Report exclusion rate separately so a shift from neutral recognition to active rejection remains visible.<\/p>\n<h3>Do recommendation rates across competitors add up to 100%?<\/h3>\n<p>Usually not. One AI response can recommend several brands, so each company may receive E.<\/p>\n<p>Use competitive recommendation share when you need a normalized distribution, but retain absolute recommendation rate because share alone can conceal a category-wide decline.<\/p>\n<h3>Can an AI visibility tool calculate the rate automatically?<\/h3>\n<p>It can automate prompt collection, entity matching, initial classification, segmentation, and trend reporting. Human-reviewed boundary rules remain important for conditional language, mixed conclusions, similarly named entities, and unusual list formats.<\/p>\n<p>Any tool should preserve the raw prompt, complete response, evidence excerpt, model metadata, label, and codebook version.<\/p>\n<h2>What should teams do next?<\/h2>\n<p><strong>Start with non-branded decision prompts, preserve every raw response, classify each target-brand observation with EQNXA, and publish explicit recommendations over all eligible responses.<\/strong><\/p>\n<p>Then diagnose the result with:<\/p>\n<ul>\n<li>Mention rate<\/li>\n<li>Recommendation conversion<\/li>\n<li>Qualified consideration rate<\/li>\n<li>Exclusion rate<\/li>\n<li>Absence rate<\/li>\n<li>Citation rate<\/li>\n<li>Recommendation prominence<\/li>\n<li>Competitive recommendation share<\/li>\n<\/ul>\n<p>maxaeo monitors how major AI assistants mention, rank, cite, and recommend brands over time. The measurement method in this guide turns those responses into an auditable KPI that SEO, marketing, PR, product, and finance teams can examine\u2014and improve\u2014without confusing visibility with endorsement.<\/p>\n<p><script type=\"application\/ld+json\">\n{\n  \"@context\": \"https:\/\/schema.org\",\n  \"@graph\": [\n    {\n      \"@type\": \"Article\",\n      \"@id\": \"https:\/\/maxaeo.ai\/blog\/ai-recommendation-rate#article\",\n      \"headline\": \"AI Recommendation Rate: Formula, Examples, and Measurement Guide\",\n      \"description\": \"Learn how to calculate AI recommendation rate, classify AI endorsements, design a valid prompt sample, set benchmarks, and report results with confidence.\",\n      \"mainEntityOfPage\": {\n        \"@type\": \"WebPage\",\n        \"@id\": \"https:\/\/maxaeo.ai\/blog\/ai-recommendation-rate\"\n      },\n      \"author\": {\n        \"@type\": \"Organization\",\n        \"name\": \"maxaeo\"\n      },\n      \"publisher\": {\n        \"@type\": \"Organization\",\n        \"name\": \"maxaeo\",\n        \"url\": \"https:\/\/maxaeo.ai\/\"\n      },\n      \"image\": \"image-placeholder\",\n      \"about\": [\n        {\n          \"@type\": \"Thing\",\n          \"name\": \"AI recommendation rate\"\n        },\n        {\n          \"@type\": \"Thing\",\n          \"name\": \"AI search visibility\"\n        },\n        {\n          \"@type\": \"Thing\",\n          \"name\": \"Generative engine optimization\"\n        }\n      ]\n    },\n    {\n      \"@type\": \"FAQPage\",\n      \"@id\": \"https:\/\/maxaeo.ai\/blog\/ai-recommendation-rate#faq\",\n      \"mainEntity\": [\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Is AI recommendation rate the same as mention rate?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"No. Mention rate counts eligible responses that name a brand regardless of context. AI recommendation rate counts only responses that explicitly endorse the brand for the user's stated need.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Should qualified consideration count as a recommendation?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Not in the strict headline metric. Qualified consideration should be reported separately or included with explicit recommendations in a companion consideration-rate metric.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"What is a good AI recommendation rate?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"There is no reliable universal benchmark. Compare the rate with a frozen internal baseline and competitors measured with the same prompts, models, markets, and annotation rules.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"How many responses are needed to measure AI recommendation rate?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Sample size depends on the required precision. Near a 20 percent observed rate, 100 independent responses have an approximate 95 percent margin of error of eight percentage points, while 400 reduce it to about four points.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Can AI recommendation rate be negative?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"No. AI recommendation rate ranges from 0 to 100 percent. Exclusions should be reported as a separate rate rather than subtracted from recommendations.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Do recommendation rates across competitors add up to 100 percent?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Usually not. A single AI response can recommend multiple brands. Competitive recommendation share can provide a normalized distribution, but it should be reported beside each brand's absolute recommendation rate.\"\n          }\n        }\n      ]\n    }\n  ]\n}\n<\/script><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Learn how to calculate AI recommendation rate, classify AI endorsements, design a valid prompt sample, set benchmarks, and report results with confidence.<\/p>\n","protected":false},"author":1,"featured_media":1242,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1244","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts\/1244","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/comments?post=1244"}],"version-history":[{"count":0,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts\/1244\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/media\/1242"}],"wp:attachment":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/media?parent=1244"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/categories?post=1244"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/tags?post=1244"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}