{"id":1553,"date":"2026-07-22T02:47:06","date_gmt":"2026-07-22T02:47:06","guid":{"rendered":"https:\/\/maxaeo.ai\/blog\/ai-visibility-rfp-template\/"},"modified":"2026-07-22T02:47:06","modified_gmt":"2026-07-22T02:47:06","slug":"ai-visibility-rfp-template","status":"publish","type":"post","link":"https:\/\/maxaeo.ai\/blog\/ai-visibility-rfp-template\/","title":{"rendered":"AI Visibility RFP Template: 100-Point Vendor Scorecard"},"content":{"rendered":"<p><em>By the maxaeo Editorial Team<\/em><\/p>\n<p>The best <strong>AI visibility RFP template<\/strong> does not ask only which engines a vendor supports. It requires proof that every reported metric can be traced from a prompt to a captured answer, source record, calculation, and export.<\/p>\n<p>Use this template to compare vendors on five evidence requirements:<\/p>\n<ol>\n<li><strong>Coverage:<\/strong> The exact engines, answer surfaces, markets, and collection methods monitored.<\/li>\n<li><strong>Traceability:<\/strong> Complete prompts, answers, citations, timestamps, and run metadata.<\/li>\n<li><strong>Reproducibility:<\/strong> Stable prompt cohorts, repeated runs, failure records, and comparable periods.<\/li>\n<li><strong>Transparency:<\/strong> Documented formulas, denominators, classifications, and exclusions.<\/li>\n<li><strong>Commercial fit:<\/strong> Security, integrations, service limits, data ownership, and total cost.<\/li>\n<\/ol>\n<p>The template includes pass-or-fail qualification gates, a copy-ready questionnaire, a 100-point scorecard, a 1,440-observation proof-of-concept protocol, and a worked vendor evaluation.<\/p>\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\/1784634900907-15-922-1.jpg\" alt=\"AI visibility RFP template scorecard with weighted requirements and evidence fields\"><\/figure>\n<h2>What is an AI visibility RFP template?<\/h2>\n<p>An AI visibility RFP template is a procurement questionnaire and scoring model used to compare platforms that monitor how generative engines mention, recommend, rank, cite, and describe brands. It converts vendor claims into testable requirements for collection coverage, raw-answer evidence, reproducibility, calculations, security, exports, service, and total cost.<\/p>\n<p>Use one when selecting an AI visibility tool, renewing an existing platform, consolidating point solutions, or choosing software for an agency portfolio.<\/p>\n<h2>What should an AI visibility RFP include?<\/h2>\n<p>A complete RFP should include <strong>scope, mandatory qualification gates, scored requirements, evidence instructions, a controlled proof of concept, security questions, commercial terms, and a decision rule established before vendor responses arrive<\/strong>.<\/p>\n<p>At minimum, define:<\/p>\n<ul>\n<li>Required engines and answer surfaces.<\/li>\n<li>Markets, languages, devices, and account states.<\/li>\n<li>Initial and expected prompt volumes.<\/li>\n<li>Run frequency and historical-retention requirements.<\/li>\n<li>Required raw-answer and citation evidence.<\/li>\n<li>Metric definitions and acceptable exclusions.<\/li>\n<li>Taxonomy, reporting, API, and warehouse requirements.<\/li>\n<li>User roles, workspaces, and agency separation.<\/li>\n<li>Security, privacy, retention, and deletion controls.<\/li>\n<li>Pricing units, overages, onboarding, and support.<\/li>\n<li>Proof-of-concept dates and acceptance criteria.<\/li>\n<li>The scoring threshold and disqualification rules.<\/li>\n<\/ul>\n<h2>Copy-ready RFP scope sheet<\/h2>\n<p>Complete this sheet before sending the RFP. Vendors should respond against the same scope so that coverage and pricing remain comparable.<\/p>\n<table>\n<thead>\n<tr>\n<th>Scope field<\/th>\n<th>Buyer requirement<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Business objective<\/td>\n<td>[Example: measure brand discovery, recommendation, citation, and description accuracy across priority buying journeys]<\/td>\n<\/tr>\n<tr>\n<td>Required engines<\/td>\n<td>[List engines and specific answer surfaces]<\/td>\n<\/tr>\n<tr>\n<td>Required markets<\/td>\n<td>[Countries, regions, or cities]<\/td>\n<\/tr>\n<tr>\n<td>Required languages<\/td>\n<td>[Languages and locale variants]<\/td>\n<\/tr>\n<tr>\n<td>Prompt volume<\/td>\n<td>[Initial prompts and expected 12-month volume]<\/td>\n<\/tr>\n<tr>\n<td>Run frequency<\/td>\n<td>[Daily, weekly, monthly, or custom]<\/td>\n<\/tr>\n<tr>\n<td>Repeated runs<\/td>\n<td>[Number of runs required per prompt and period]<\/td>\n<\/tr>\n<tr>\n<td>Competitor capacity<\/td>\n<td>[Tracked brands per project or prompt cohort]<\/td>\n<\/tr>\n<tr>\n<td>Users and workspaces<\/td>\n<td>[Seats, roles, teams, business units, or clients]<\/td>\n<\/tr>\n<tr>\n<td>Historical retention<\/td>\n<td>[Raw answers, citations, metadata, and aggregates]<\/td>\n<\/tr>\n<tr>\n<td>Required integrations<\/td>\n<td>[CSV, API, warehouse, BI platform, webhooks]<\/td>\n<\/tr>\n<tr>\n<td>Security requirements<\/td>\n<td>[SSO, provisioning, audit logs, DPA, data location]<\/td>\n<\/tr>\n<tr>\n<td>Implementation deadline<\/td>\n<td>[Target production date]<\/td>\n<\/tr>\n<tr>\n<td>Proof-of-concept period<\/td>\n<td>[Dates and participating vendors]<\/td>\n<\/tr>\n<tr>\n<td>Contract term<\/td>\n<td>[Initial term, renewal options, and notice period]<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h3>Required vendor response format<\/h3>\n<p>For every requirement, instruct vendors to provide:<\/p>\n<ol>\n<li><strong>Status:<\/strong> Supported, partially supported, not supported, or planned.<\/li>\n<li><strong>Limitations:<\/strong> Engine, market, plan, volume, retention, or workflow restrictions.<\/li>\n<li><strong>Evidence:<\/strong> Product record, export, documentation, contract language, or live demonstration.<\/li>\n<li><strong>Availability:<\/strong> Generally available, beta, services-assisted, or estimated delivery date.<\/li>\n<li><strong>Cost:<\/strong> Included, paid add-on, usage-based, or professional services.<\/li>\n<li><strong>Owner:<\/strong> Vendor contact accountable for the response.<\/li>\n<\/ol>\n<p>A roadmap statement is not evidence of a current capability. Score \u201cplanned\u201d as absent unless the buyer has approved a contractual delivery condition.<\/p>\n<h2>Use the TRACE framework to test vendor evidence<\/h2>\n<p>The maxaeo TRACE framework prevents procurement teams from scoring isolated dashboard features without examining the evidence chain behind them.<\/p>\n<table>\n<thead>\n<tr>\n<th>TRACE component<\/th>\n<th>What it means<\/th>\n<th>Buyer test<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>T \u2014 Taxonomy<\/strong><\/td>\n<td>Prompts are organized by stable products, markets, personas, intents, and priorities<\/td>\n<td>Change a segment and inspect its version history<\/td>\n<\/tr>\n<tr>\n<td><strong>R \u2014 Run controls<\/strong><\/td>\n<td>Prompt wording, surface, location, language, account state, and schedule are recorded<\/td>\n<td>Repeat the same controlled prompt set<\/td>\n<\/tr>\n<tr>\n<td><strong>A \u2014 Answer evidence<\/strong><\/td>\n<td>Full responses, citations, classifications, and failures remain inspectable<\/td>\n<td>Trace a chart value to its original answer<\/td>\n<\/tr>\n<tr>\n<td><strong>C \u2014 Calculation disclosure<\/strong><\/td>\n<td>Formulas, denominators, weights, and exclusions are documented<\/td>\n<td>Recalculate a metric from exported records<\/td>\n<\/tr>\n<tr>\n<td><strong>E \u2014 Exportability<\/strong><\/td>\n<td>Data can be independently analyzed outside the vendor interface<\/td>\n<td>Join prompt, run, answer, brand, and citation exports<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>A vendor should not receive a high trust score unless the complete chain works. For example, a correct formula cannot compensate for missing answers, and an answer archive cannot compensate for changing prompt cohorts that are not versioned.<\/p>\n<p>This evidence-first approach also underpins a rigorous <a href=\"https:\/\/maxaeo.ai\/blog\/how-accurate-are-ai-visibility-tools\">data-quality audit of AI visibility tools<\/a>.<\/p>\n<h2>Which requirements should be mandatory qualification gates?<\/h2>\n<p>Qualification gates should cover deficiencies that make every downstream metric difficult to verify. Apply these gates before considering price or weighted scores.<\/p>\n<table>\n<thead>\n<tr>\n<th>Gate<\/th>\n<th>Pass criterion<\/th>\n<th>Evidence required<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Raw-answer access<\/td>\n<td>Every successful observation links to the complete submitted prompt and response<\/td>\n<td>Live record plus answer-level export<\/td>\n<\/tr>\n<tr>\n<td>Run metadata<\/td>\n<td>Records include engine, surface, timestamp, market, language, account state, and collection method<\/td>\n<td>Product record and export schema<\/td>\n<\/tr>\n<tr>\n<td>Citation traceability<\/td>\n<td>Exact destination URLs, labels, and citation order are stored when the surface exposes citations<\/td>\n<td>Raw answer and citation export<\/td>\n<\/tr>\n<tr>\n<td>Failure visibility<\/td>\n<td>Errors, blocks, refusals, timeouts, and incomplete runs are retained rather than silently replaced<\/td>\n<td>Failed-run sample and status definitions<\/td>\n<\/tr>\n<tr>\n<td>Metric transparency<\/td>\n<td>Formulas, denominators, exclusions, deduplication, weighting, and rounding are documented<\/td>\n<td>Metric dictionary and worked calculation<\/td>\n<\/tr>\n<tr>\n<td>Historical evidence<\/td>\n<td>Raw answers and citations remain accessible for the contracted retention period<\/td>\n<td>Retention policy and historical sample<\/td>\n<\/tr>\n<tr>\n<td>Machine-readable export<\/td>\n<td>Answer-level observations can be exported with stable identifiers<\/td>\n<td>CSV, JSON, API, or warehouse sample<\/td>\n<\/tr>\n<tr>\n<td>Taxonomy control<\/td>\n<td>Prompts support buyer-defined segments and versioned changes<\/td>\n<td>Demonstration using the buyer\u2019s taxonomy<\/td>\n<\/tr>\n<tr>\n<td>Proof-of-concept access<\/td>\n<td>The vendor permits controlled testing with buyer-supplied prompts<\/td>\n<td>Written POC agreement<\/td>\n<\/tr>\n<tr>\n<td>Governance documentation<\/td>\n<td>Security controls, subprocessors, data use, retention, and deletion procedures are documented<\/td>\n<td>Security pack, DPA, and contract terms<\/td>\n<\/tr>\n<tr>\n<td>Data portability<\/td>\n<td>Buyer-created prompts, tags, annotations, and exports can be retrieved at contract end<\/td>\n<td>Contract language and exit procedure<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>A vendor that fails a mandatory gate should not advance because of a high feature score. If raw answers are unavailable, the buyer cannot independently defend the vendor\u2019s visibility, sentiment, or recommendation metrics.<\/p>\n<h2>Copy-ready 100-point vendor scorecard<\/h2>\n<p>Score each category from 0 to 4:<\/p>\n<ul>\n<li><strong>0 \u2014 Absent:<\/strong> Not supported.<\/li>\n<li><strong>1 \u2014 Claimed:<\/strong> Described without inspectable evidence or available only on a roadmap.<\/li>\n<li><strong>2 \u2014 Partial:<\/strong> Available with material limitations or manual work.<\/li>\n<li><strong>3 \u2014 Meets requirement:<\/strong> Works as requested and has been demonstrated.<\/li>\n<li><strong>4 \u2014 Exceeds requirement:<\/strong> Adds useful controls, depth, or automation beyond the requirement.<\/li>\n<\/ul>\n<p>Calculate each weighted result as:<\/p>\n<p><strong>Weighted result = rating \u00f7 4 \u00d7 category weight<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Evaluation category<\/th>\n<th align=\"right\">Weight<\/th>\n<th>Requirement buyers should test<\/th>\n<th>Required evidence<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Engine and surface coverage<\/td>\n<td align=\"right\">14<\/td>\n<td>Monitors required surfaces with documented collection methods, markets, languages, and run conditions<\/td>\n<td>Records from every contracted surface<\/td>\n<\/tr>\n<tr>\n<td>Raw-answer evidence<\/td>\n<td align=\"right\">12<\/td>\n<td>Preserves complete prompts, answers, classifications, and run metadata<\/td>\n<td>Answer-level record and export<\/td>\n<\/tr>\n<tr>\n<td>Reproducibility and data quality<\/td>\n<td align=\"right\">14<\/td>\n<td>Supports repeated runs, versioned prompts, stable cohorts, failure tracking, and comparable periods<\/td>\n<td>Controlled re-run history<\/td>\n<\/tr>\n<tr>\n<td>Citation and source capture<\/td>\n<td align=\"right\">12<\/td>\n<td>Stores exact URLs, citation context, order, and historical changes<\/td>\n<td>Citation records joined to answers<\/td>\n<\/tr>\n<tr>\n<td>Taxonomy and segmentation<\/td>\n<td align=\"right\">9<\/td>\n<td>Filters by product, persona, intent, market, funnel stage, business unit, or client<\/td>\n<td>Demonstration using buyer taxonomy<\/td>\n<\/tr>\n<tr>\n<td>Metric transparency<\/td>\n<td align=\"right\">12<\/td>\n<td>Discloses formulas, denominators, weights, exclusions, and classification rules<\/td>\n<td>Metric dictionary and recalculation<\/td>\n<\/tr>\n<tr>\n<td>History and change management<\/td>\n<td align=\"right\">5<\/td>\n<td>Retains evidence and logs changes to prompts, taxonomies, collection, and metrics<\/td>\n<td>Historical sample and change log<\/td>\n<\/tr>\n<tr>\n<td>Exports and integrations<\/td>\n<td align=\"right\">8<\/td>\n<td>Delivers normalized, joinable data to analyst and reporting workflows<\/td>\n<td>CSV\/API sample and schema<\/td>\n<\/tr>\n<tr>\n<td>Governance, security, and privacy<\/td>\n<td align=\"right\">9<\/td>\n<td>Provides access controls, auditability, data-handling terms, and deletion controls<\/td>\n<td>Security pack, DPA, and subprocessors<\/td>\n<\/tr>\n<tr>\n<td>Service and commercial fit<\/td>\n<td align=\"right\">5<\/td>\n<td>Defines implementation, support, limits, overages, ownership, and exit terms<\/td>\n<td>Order form, SLA, and implementation plan<\/td>\n<\/tr>\n<tr>\n<td><strong>Total<\/strong><\/td>\n<td align=\"right\"><strong>100<\/strong><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Set the shortlist threshold before opening responses. A threshold of 75 points may be a useful starting rule, but mandatory gates should always take precedence over the numerical total.<\/p>\n<h2>Copy-ready vendor requirements questionnaire<\/h2>\n<h3>Engine coverage and collection<\/h3>\n<p><strong>COV-01 \u2014 Surface-level coverage<\/strong><\/p>\n<p>List every supported engine and answer surface separately. For each, provide:<\/p>\n<ul>\n<li>Consumer interface, official API, third-party API, browser automation, or other collection method.<\/li>\n<li>Supported countries and languages.<\/li>\n<li>Logged-in or logged-out account state.<\/li>\n<li>Desktop, mobile, or other device context.<\/li>\n<li>Standard and maximum run frequency.<\/li>\n<li>Citation availability.<\/li>\n<li>Model or mode metadata captured.<\/li>\n<\/ul>\n<p><strong>Acceptance evidence:<\/strong> One current answer record and export row from every proposed surface.<\/p>\n<p><strong>COV-02 \u2014 Interface equivalence<\/strong><\/p>\n<p>Explain whether any API or proxy dataset is presented as representative of a consumer interface. Provide the validation method, comparison date, sample size, and known differences.<\/p>\n<p><strong>Acceptance evidence:<\/strong> A written equivalence study or a clear statement that the datasets are reported separately.<\/p>\n<p>An API response should not be described as what users saw in a consumer product unless the vendor has demonstrated equivalence. Model configuration, retrieval systems, system instructions, location, and account state can all affect the response.<\/p>\n<p><strong>COV-03 \u2014 Collection-change management<\/strong><\/p>\n<p>Describe how customers are notified when an engine changes its interface, retrieval method, model, citation format, or access restrictions.<\/p>\n<p><strong>Acceptance evidence:<\/strong> A sample customer notice or collection-method change log.<\/p>\n<p><strong>COV-04 \u2014 Run failures and retries<\/strong><\/p>\n<p>Define every run status, including success, refusal, block, timeout, parse failure, incomplete answer, and retry.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Exported failed records showing original attempts and retries as separate events.<\/p>\n<p><strong>COV-05 \u2014 Index and retrieval context<\/strong><\/p>\n<p>Explain what the vendor can and cannot infer about the search or retrieval systems behind each answer surface.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Methodology documentation that distinguishes observed citations from inferred source systems.<\/p>\n<p>Buyers can use maxaeo\u2019s <a href=\"https:\/\/maxaeo.ai\/blog\/which-search-engines-power-ai-answers\">mapping of the search indexes behind AI answer engines<\/a> to identify questions that should be asked surface by surface.<\/p>\n<h3>Raw answers, brands, and citations<\/h3>\n<p><strong>EVD-01 \u2014 Complete answer retention<\/strong><\/p>\n<p>Store the exact prompt and full answer without replacing either with a summary.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Product drill-down and export containing matching prompt and response text.<\/p>\n<p><strong>EVD-02 \u2014 Brand classification<\/strong><\/p>\n<p>For every detected brand, identify whether it was:<\/p>\n<ul>\n<li>Mentioned.<\/li>\n<li>Explicitly recommended.<\/li>\n<li>Included in a ranked or unranked shortlist.<\/li>\n<li>Compared with another brand.<\/li>\n<li>Criticized or associated with a risk.<\/li>\n<li>Cited as a source rather than discussed as a vendor.<\/li>\n<\/ul>\n<p><strong>Acceptance evidence:<\/strong> Classification rules, confidence values, and examples of manual correction.<\/p>\n<p><strong>EVD-03 \u2014 Entity resolution<\/strong><\/p>\n<p>Explain how the system handles acronyms, former company names, product names, parent companies, spelling variants, and similarly named organizations.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Test results for five buyer-supplied entity edge cases.<\/p>\n<p><strong>EVD-04 \u2014 Citation records<\/strong><\/p>\n<p>Store the exact destination URL, visible label, citation position, referring answer, and capture time. Domain-only totals do not pass.<\/p>\n<p><strong>Acceptance evidence:<\/strong> A citation export joined to raw-answer identifiers.<\/p>\n<p>A useful workflow connects the cited URL to the answer claim and the page-level action an owner can take. The <a href=\"https:\/\/maxaeo.ai\/blog\/geo-citation-tracking\">GEO citation-tracking workflow<\/a> shows how to turn source records into specific fixes.<\/p>\n<p><strong>EVD-05 \u2014 Historical source changes<\/strong><\/p>\n<p>Explain how redirects, deleted pages, URL parameters, canonicalization, syndicated content, and citation-order changes are represented over time.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Historical examples retaining both the captured URL and any subsequently resolved destination.<\/p>\n<h3>Measurement and reproducibility<\/h3>\n<p><strong>MET-01 \u2014 Prompt versioning<\/strong><\/p>\n<p>Assign stable identifiers to prompts and preserve every wording, taxonomy, weight, and status change.<\/p>\n<p><strong>Acceptance evidence:<\/strong> A prompt history showing who changed what and when.<\/p>\n<p><strong>MET-02 \u2014 Repeated observations<\/strong><\/p>\n<p>Support multiple runs of the same prompt without overwriting earlier answers.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Run-level history for one prompt repeated three times.<\/p>\n<p><strong>MET-03 \u2014 Formula disclosure<\/strong><\/p>\n<p>Document every formula, denominator, eligibility rule, weight, exclusion, deduplication step, and rounding rule.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Metric dictionary and a calculation reconstructed from ten answer records.<\/p>\n<p><strong>MET-04 \u2014 Missingness and volatility<\/strong><\/p>\n<p>Report collection success, answer variation, and citation variation separately from visibility movement.<\/p>\n<p><strong>Acceptance evidence:<\/strong> A view or export that exposes failed runs and repeated-run disagreement.<\/p>\n<p><strong>MET-05 \u2014 Historical comparability<\/strong><\/p>\n<p>Explain what happens to historical trends when prompts, competitors, segments, engines, or formulas change.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Versioned methodology and a restatement policy.<\/p>\n<h3>Segmentation, exports, and workflow<\/h3>\n<p><strong>OPS-01 \u2014 Multi-label taxonomy<\/strong><\/p>\n<p>Allow one prompt to belong to multiple buyer-defined segments, such as product, persona, market, intent, funnel stage, campaign, and priority.<\/p>\n<p><strong>Acceptance evidence:<\/strong> A demonstration using the buyer\u2019s sample taxonomy.<\/p>\n<p><strong>OPS-02 \u2014 Answer-level export<\/strong><\/p>\n<p>Provide one record per answer observation or a documented relational schema that produces the same result.<\/p>\n<p><strong>Acceptance evidence:<\/strong> CSV, JSON, API, or warehouse sample containing stable prompt, run, answer, engine, brand, citation, and segment identifiers.<\/p>\n<p><strong>OPS-03 \u2014 Data reconciliation<\/strong><\/p>\n<p>Ensure dashboard totals can be reconciled with exported records.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Independent calculation matching the interface after documented rounding.<\/p>\n<p><strong>OPS-04 \u2014 Integrations and limits<\/strong><\/p>\n<p>List supported APIs, warehouse destinations, BI integrations, rate limits, latency, backfill options, and additional fees.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Technical documentation and pricing schedule.<\/p>\n<p><strong>OPS-05 \u2014 Reporting portability<\/strong><\/p>\n<p>Confirm that customers can create independent reports without manually copying dashboard values.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Export used to reproduce one vendor chart. A practical reporting structure is available in maxaeo\u2019s <a href=\"https:\/\/maxaeo.ai\/blog\/ai-visibility-report\">AI visibility report template<\/a>.<\/p>\n<h3>Governance, security, and commercial terms<\/h3>\n<p><strong>GOV-01 \u2014 Access control<\/strong><\/p>\n<p>Describe roles, least-privilege permissions, SSO, user provisioning, workspace separation, and export controls.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Access-control matrix and live demonstration.<\/p>\n<p><strong>GOV-02 \u2014 Auditability<\/strong><\/p>\n<p>Log changes to prompts, taxonomies, users, monitoring settings, classifications, and deletions.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Searchable audit-log sample.<\/p>\n<p><strong>GOV-03 \u2014 Customer-data use<\/strong><\/p>\n<p>State whether customer prompts, answers, exports, annotations, or feedback are used to train or improve vendor or third-party models.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Contract language, privacy terms, and opt-out controls.<\/p>\n<p><strong>GOV-04 \u2014 Retention and deletion<\/strong><\/p>\n<p>Document active retention, backups, deletion timing, account closure, legal holds, and data export before termination.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Retention schedule and deletion procedure.<\/p>\n<p><strong>GOV-05 \u2014 Security documentation<\/strong><\/p>\n<p>Provide the applicable DPA, subprocessor list, incident-notification terms, penetration-test summary, certification scope, and data-location options.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Current documents covering the proposed service.<\/p>\n<p><strong>COM-01 \u2014 Pricing units<\/strong><\/p>\n<p>Disclose limits and prices for prompts, runs, engines, locations, languages, projects, competitors, seats, exports, API calls, retention, and workspaces.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Completed pricing worksheet and order form.<\/p>\n<p><strong>COM-02 \u2014 Implementation responsibility<\/strong><\/p>\n<p>Identify who configures prompts, taxonomies, competitors, access, integrations, validation, and training.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Named responsibilities, timeline, dependencies, and acceptance criteria.<\/p>\n<p><strong>COM-03 \u2014 Support and service levels<\/strong><\/p>\n<p>Provide support channels, response targets, escalation paths, availability commitments, maintenance rules, and service credits.<\/p>\n<p><strong>Acceptance evidence:<\/strong> SLA and support policy.<\/p>\n<p><strong>COM-04 \u2014 Ownership and portability<\/strong><\/p>\n<p>Confirm ownership and export rights for buyer-created prompts, tags, annotations, mappings, reports, and configuration.<\/p>\n<p><strong>Acceptance evidence:<\/strong> Contract language and a documented exit process.<\/p>\n<h2>How should engine coverage be evaluated?<\/h2>\n<p><strong>Engine coverage means verified access to each required answer surface under documented conditions\u2014not the number of logos on a pricing page.<\/strong><\/p>\n<p>Request separate responses for ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, and Google AI Overviews when those surfaces are relevant to the buyer.<\/p>\n<p>For each surface, verify:<\/p>\n<ul>\n<li>How the answer is collected.<\/li>\n<li>Whether the record identifies the surface and collection method.<\/li>\n<li>Whether account state and personalization are controlled.<\/li>\n<li>Which markets and languages are supported.<\/li>\n<li>Whether location is requested, inferred, or unavailable.<\/li>\n<li>Whether citations and search links are captured separately.<\/li>\n<li>Whether model or mode information is stored when exposed.<\/li>\n<li>How failed runs are detected and reported.<\/li>\n<li>How interface or collection changes are documented.<\/li>\n<\/ul>\n<p>Do not award full coverage credit when a vendor groups materially different surfaces under one parent brand. A conversational assistant, search-grounded response, and generated search summary are not automatically comparable datasets.<\/p>\n<h2>Which calculations must vendors disclose?<\/h2>\n<p>Every metric should have a written formula, a defined denominator, and an answer-level drill-down. Reject proprietary composite scores that cannot be reconstructed from exported observations.<\/p>\n<table>\n<thead>\n<tr>\n<th>Metric<\/th>\n<th>Transparent calculation<\/th>\n<th>Required clarification<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Collection success rate<\/td>\n<td>Completed, parseable answers \u00f7 scheduled runs \u00d7 100<\/td>\n<td>Are retries counted as new attempts or replacements?<\/td>\n<\/tr>\n<tr>\n<td>Mention rate<\/td>\n<td>Eligible answers mentioning the brand \u00f7 eligible answers \u00d7 100<\/td>\n<td>Are refusals, errors, and incomplete answers eligible?<\/td>\n<\/tr>\n<tr>\n<td>Recommendation rate<\/td>\n<td>Eligible answers explicitly recommending the brand \u00f7 eligible answers \u00d7 100<\/td>\n<td>What language qualifies as an explicit recommendation?<\/td>\n<\/tr>\n<tr>\n<td>Top-three rate<\/td>\n<td>Answers placing the brand in positions 1\u20133 \u00f7 eligible ranked answers \u00d7 100<\/td>\n<td>How are prose and unnumbered lists ordered?<\/td>\n<\/tr>\n<tr>\n<td>AI share of voice<\/td>\n<td>Brand mentions \u00f7 mentions of all tracked brands \u00d7 100<\/td>\n<td>Is each brand counted once per answer or once per occurrence?<\/td>\n<\/tr>\n<tr>\n<td>Citation rate<\/td>\n<td>Eligible answers containing a qualifying citation \u00f7 eligible answers \u00d7 100<\/td>\n<td>Are citations, search links, and plain-text domains separated?<\/td>\n<\/tr>\n<tr>\n<td>Owned-source share<\/td>\n<td>Citations to buyer-controlled domains \u00f7 all captured citations \u00d7 100<\/td>\n<td>How are subdomains, redirects, and syndicated pages classified?<\/td>\n<\/tr>\n<tr>\n<td>Description consistency<\/td>\n<td>Answers matching approved claims \u00f7 answers describing the brand \u00d7 100<\/td>\n<td>Is matching automated, manually reviewed, or both?<\/td>\n<\/tr>\n<tr>\n<td>Prompt-weighted visibility<\/td>\n<td>Sum of prompt result \u00d7 approved prompt weight \u00f7 sum of eligible weights<\/td>\n<td>Who controls weights, and are historical weights versioned?<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Metrics should be available by prompt, engine, product, market, persona, and other buyer-defined segments. Account-wide averages can conceal zero visibility for a priority product or region.<\/p>\n<p>Require the vendor to calculate three metrics from ten buyer-supplied answer records. Recalculate them independently. Any difference should be explainable through documented exclusions, weighting, deduplication, or rounding.<\/p>\n<h2>How can buyers test reproducibility?<\/h2>\n<p>Reproducibility does not require identical prose from a generative engine. It requires a consistent sampling method that exposes ordinary answer variation instead of hiding it inside a smoothed trend.<\/p>\n<p>Use this controlled proof-of-concept protocol for every finalist:<\/p>\n<ol>\n<li>Create 20 prompts: five category recommendations, five comparisons, five problem-to-solution questions, and five entity or reputation questions.<\/li>\n<li>Run every prompt three times per engine on three separate days.<\/li>\n<li>Keep wording, language, location, device, account state, and answer surface constant.<\/li>\n<li>Preserve every attempt, including errors, refusals, and retries.<\/li>\n<li>Export all answers, citations, classifications, and metadata.<\/li>\n<li>Reconstruct the agreed metrics outside the platform.<\/li>\n<li>Compare brand, ranking, citation, and description stability.<\/li>\n<\/ol>\n<p>This produces <strong>180 scheduled observations per engine<\/strong>:<\/p>\n<p><strong>20 prompts \u00d7 3 runs \u00d7 3 days = 180 observations<\/strong><\/p>\n<p>Testing eight engines produces <strong>1,440 scheduled observations<\/strong>:<\/p>\n<p><strong>180 observations \u00d7 8 engines = 1,440 observations<\/strong><\/p>\n<p>This is large enough to expose missing metadata, overwritten answers, duplicate records, inconsistent classification, and export defects. It is <strong>not<\/strong> a statistically representative estimate of every buyer query or long-term market visibility.<\/p>\n<h3>Measure agreement without expecting identical prose<\/h3>\n<p>Use at least three transparent diagnostics:<\/p>\n<p><strong>Brand-set agreement<\/strong><\/p>\n<p><strong>Jaccard agreement = brands found in both runs \u00f7 brands found in either run<\/strong><\/p>\n<p>A score of 1 means the same brand set appeared in both answers. A score of 0 means the two answers shared no brands.<\/p>\n<p><strong>Citation-URL agreement<\/strong><\/p>\n<p><strong>Citation agreement = normalized URLs found in both runs \u00f7 normalized URLs found in either run<\/strong><\/p>\n<p>Require the vendor to disclose URL-normalization rules, including query parameters, redirects, fragments, and trailing slashes.<\/p>\n<p><strong>Rank-band stability<\/strong><\/p>\n<p><strong>Rank-band stability = runs in which the brand remains in the same rank band \u00f7 eligible runs<\/strong><\/p>\n<p>Define rank bands before testing, such as positions 1\u20133, 4\u201310, mentioned but unranked, and absent.<\/p>\n<p>Low agreement does not automatically mean the platform is defective; generative answers vary. The disqualifying problem is <strong>hidden variation<\/strong>\u2014for example, overwritten runs, silent retries, or a trend that excludes inconvenient failures.<\/p>\n<h2>How should the proof of concept be evaluated?<\/h2>\n<p>The POC should test the buyer\u2019s prompts, classifications, reporting workflow, and security requirements. A scripted vendor demonstration cannot substitute for controlled evidence.<\/p>\n<h3>1. Build a balanced prompt cohort<\/h3>\n<p>Avoid using only branded prompts. They mostly test whether the engine recognizes an existing entity.<\/p>\n<p>Include:<\/p>\n<ul>\n<li><strong>Category discovery:<\/strong> \u201cWhich platforms are best for [job]?\u201d<\/li>\n<li><strong>Comparison:<\/strong> \u201cCompare [brand] and [competitor] for [use case].\u201d<\/li>\n<li><strong>Problem-led:<\/strong> \u201cHow should a [role] solve [problem]?\u201d<\/li>\n<li><strong>Requirement-led:<\/strong> \u201cWhich [category] vendors support [requirement]?\u201d<\/li>\n<li><strong>Entity:<\/strong> \u201cWhat does [brand] do, and who is it for?\u201d<\/li>\n<\/ul>\n<p>Label each prompt by market, persona, intent, funnel stage, product, and business priority.<\/p>\n<h3>2. Inspect evidence before reviewing charts<\/h3>\n<p>Randomly select ten observations and confirm:<\/p>\n<ul>\n<li>The raw answer matches the submitted prompt.<\/li>\n<li>The recorded surface and run conditions are correct.<\/li>\n<li>Detected brands appear in the answer.<\/li>\n<li>Recommendations are classified separately from mentions.<\/li>\n<li>Citations resolve to the captured URLs.<\/li>\n<li>Timestamps and run identifiers are coherent.<\/li>\n<li>Errors and retries remain visible.<\/li>\n<\/ul>\n<p>Then test five entity edge cases: an acronym, a former company name, a similarly named competitor, an ambiguous product name, and an answer that criticizes the brand.<\/p>\n<h3>3. Reconstruct one dashboard result<\/h3>\n<p>Select one engine, segment, and period. Export its records and independently calculate collection success, mention rate, recommendation rate, and AI share of voice.<\/p>\n<p>The vendor passes when the independent results match the interface after documented rounding. An unexplained discrepancy should reduce the metric-transparency score even if the chart appears plausible.<\/p>\n<h3>4. Test export integrity<\/h3>\n<p>Verify that:<\/p>\n<ul>\n<li>All successful and failed observations have stable run identifiers.<\/li>\n<li>Brand and citation records join to the correct answer.<\/li>\n<li>Null values are documented rather than replaced with zeros.<\/li>\n<li>Timestamps use a documented timezone.<\/li>\n<li>Segments and prompt versions can be reconstructed.<\/li>\n<li>The export contains no unexplained duplicates.<\/li>\n<li>Dashboard filters produce the same population as export filters.<\/li>\n<\/ul>\n<h3>5. Record POC acceptance results<\/h3>\n<p>Use a simple acceptance table:<\/p>\n<table>\n<thead>\n<tr>\n<th>Test<\/th>\n<th>Pass condition<\/th>\n<th>Result<\/th>\n<th>Evidence<\/th>\n<th>Contract condition<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Raw-answer linkage<\/td>\n<td>Every sampled dashboard value links to an answer<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Metadata completeness<\/td>\n<td>Required fields exist for every sampled record<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Failure visibility<\/td>\n<td>Failed attempts remain visible and exportable<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Metric reconstruction<\/td>\n<td>Independent calculations match documented results<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Citation integrity<\/td>\n<td>Captured URLs match answer citations<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Export joins<\/td>\n<td>No orphaned prompt, answer, brand, or citation records<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Entity edge cases<\/td>\n<td>Each case is correctly handled or correctable<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>Access controls<\/td>\n<td>Required roles and workspace boundaries function<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/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\/1784634900907-15-922-2.jpg\" alt=\"Analyst validating an AI visibility RFP template against raw answers, citations, and exported calculations\"><\/figure>\n<h2>What segmentation and exports should the RFP require?<\/h2>\n<p><strong>Segmentation turns a blended visibility score into a diagnosis. Exports make that diagnosis independently verifiable.<\/strong><\/p>\n<p>Prompts should support multiple labels, including:<\/p>\n<ul>\n<li>Product or service.<\/li>\n<li>Category and use case.<\/li>\n<li>Audience or buyer role.<\/li>\n<li>Search intent.<\/li>\n<li>Funnel stage.<\/li>\n<li>Market and language.<\/li>\n<li>Business unit.<\/li>\n<li>Campaign or strategic priority.<\/li>\n<li>Agency client.<\/li>\n<\/ul>\n<p>Segment definitions must be versioned. Adding easy branded prompts to an existing cohort can raise a trend even when no engine answer has improved.<\/p>\n<p>The minimum useful export should preserve:<\/p>\n<ul>\n<li>Stable prompt and prompt-version identifiers.<\/li>\n<li>Run and answer identifiers.<\/li>\n<li>Engine and answer surface.<\/li>\n<li>Exact prompt and full response.<\/li>\n<li>Scheduled and completed timestamps.<\/li>\n<li>Market, language, device, and account state.<\/li>\n<li>Run status, error type, and retry relationship.<\/li>\n<li>Brand entity, classification, position, and confidence.<\/li>\n<li>Exact citation URL, label, order, and resolved URL.<\/li>\n<li>Segment identifiers and historical taxonomy.<\/li>\n<li>Metric eligibility and exclusion reason.<\/li>\n<\/ul>\n<p>PDF reports are useful for executives, but they are not a substitute for answer-level data. Analysts should be able to reproduce an independent report without copying numbers from charts.<\/p>\n<h2>How should governance and security be scored?<\/h2>\n<p>Governance should reflect the sensitivity of prompt libraries, competitive intelligence, unreleased products, user access, and exported response data.<\/p>\n<p>The official <a href=\"https:\/\/www.nist.gov\/itl\/ai-risk-management-framework\" target=\"_blank\" rel=\"noopener\">NIST AI Risk Management Framework<\/a> organizes AI risk work around Govern, Map, Measure, and Manage. Procurement teams can translate those functions into concrete platform controls:<\/p>\n<ul>\n<li>Role-based access and least-privilege permissions.<\/li>\n<li>SSO and user provisioning where required.<\/li>\n<li>Workspace separation for business units or agency clients.<\/li>\n<li>Audit logs for prompts, taxonomies, classifications, exports, and settings.<\/li>\n<li>Encryption, data-location, and key-management commitments.<\/li>\n<li>Documented retention, backup, deletion, and account-closure procedures.<\/li>\n<li>A current subprocessor list and incident-notification terms.<\/li>\n<li>Explicit rules for using customer data to train or improve models.<\/li>\n<li>Human correction and approval workflows for sensitive classifications.<\/li>\n<li>Export and portability rights at contract termination.<\/li>\n<\/ul>\n<p>Certifications support a review but do not replace it. Inspect the certification scope, exceptions, covered services, report period, and contract language.<\/p>\n<h2>Which commercial terms deserve close review?<\/h2>\n<p>Commercial comparisons should normalize what the buyer is actually purchasing. \u201cUnlimited projects\u201d may have little value when prompts, engines, runs, markets, exports, seats, or historical retention remain restricted.<\/p>\n<p>Request prices and limits for:<\/p>\n<ul>\n<li>Prompts and prompt versions.<\/li>\n<li>Scheduled and completed runs.<\/li>\n<li>Engines and answer surfaces.<\/li>\n<li>Markets, languages, and locations.<\/li>\n<li>Competitors and tracked entities.<\/li>\n<li>Seats, roles, and workspaces.<\/li>\n<li>Raw-answer retention.<\/li>\n<li>CSV exports, API calls, and warehouse delivery.<\/li>\n<li>SSO, provisioning, and audit logs.<\/li>\n<li>Onboarding and taxonomy design.<\/li>\n<li>Migration and historical backfill.<\/li>\n<li>Premium support and professional services.<\/li>\n<\/ul>\n<p>Calculate total cost using the same scope for every vendor:<\/p>\n<p><strong>Year-one cost = subscription + implementation + integrations + required add-ons + expected overages + internal implementation labor<\/strong><\/p>\n<p>A secondary operational comparison can be useful:<\/p>\n<p><strong>Cost per successful observation = year-one platform cost \u00f7 completed, usable answer records<\/strong><\/p>\n<p>Do not use cost per observation as the sole buying metric. A cheap record without raw evidence, reliable metadata, or exportability has limited decision value.<\/p>\n<p>Also review:<\/p>\n<ul>\n<li>Overage calculation and notification rules.<\/li>\n<li>Renewal uplift and cancellation notice.<\/li>\n<li>Minimum commitments.<\/li>\n<li>Payment milestones tied to implementation acceptance.<\/li>\n<li>Ownership of prompts, tags, mappings, and annotations.<\/li>\n<li>Data-access rights after termination.<\/li>\n<li>Deletion and backup-expiration timing.<\/li>\n<li>Support response targets and service credits.<\/li>\n<li>Remedies when contracted coverage becomes unavailable.<\/li>\n<\/ul>\n<h2>What does a worked vendor score look like?<\/h2>\n<p>Consider a hypothetical vendor that passes every mandatory gate but provides only partial documentation for metric weighting.<\/p>\n<table>\n<thead>\n<tr>\n<th>Category<\/th>\n<th align=\"right\">Rating<\/th>\n<th align=\"right\">Weight<\/th>\n<th align=\"right\">Weighted result<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Engine and surface coverage<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">14<\/td>\n<td align=\"right\">10.50<\/td>\n<\/tr>\n<tr>\n<td>Raw-answer evidence<\/td>\n<td align=\"right\">4<\/td>\n<td align=\"right\">12<\/td>\n<td align=\"right\">12.00<\/td>\n<\/tr>\n<tr>\n<td>Reproducibility and data quality<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">14<\/td>\n<td align=\"right\">10.50<\/td>\n<\/tr>\n<tr>\n<td>Citation and source capture<\/td>\n<td align=\"right\">4<\/td>\n<td align=\"right\">12<\/td>\n<td align=\"right\">12.00<\/td>\n<\/tr>\n<tr>\n<td>Taxonomy and segmentation<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">9<\/td>\n<td align=\"right\">6.75<\/td>\n<\/tr>\n<tr>\n<td>Metric transparency<\/td>\n<td align=\"right\">2<\/td>\n<td align=\"right\">12<\/td>\n<td align=\"right\">6.00<\/td>\n<\/tr>\n<tr>\n<td>History and change management<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">5<\/td>\n<td align=\"right\">3.75<\/td>\n<\/tr>\n<tr>\n<td>Exports and integrations<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">8<\/td>\n<td align=\"right\">6.00<\/td>\n<\/tr>\n<tr>\n<td>Governance, security, and privacy<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">9<\/td>\n<td align=\"right\">6.75<\/td>\n<\/tr>\n<tr>\n<td>Service and commercial fit<\/td>\n<td align=\"right\">3<\/td>\n<td align=\"right\">5<\/td>\n<td align=\"right\">3.75<\/td>\n<\/tr>\n<tr>\n<td><strong>Total<\/strong><\/td>\n<td align=\"right\"><\/td>\n<td align=\"right\"><strong>100<\/strong><\/td>\n<td align=\"right\"><strong>78.00<\/strong><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>A score of 78 supports shortlisting, not automatic selection. The buyer should require a complete metric dictionary and successful recalculation before signing.<\/p>\n<p>If the same vendor withheld raw answers, its score would no longer matter because it would fail a mandatory gate. This is why the template combines qualification gates with weighted scoring.<\/p>\n<p>Teams comparing citation-focused products can also use the <a href=\"https:\/\/maxaeo.ai\/blog\/ai-visibility-tools-citation-tracking\">AI visibility tools buyer\u2019s guide and scorecard<\/a> to examine differences in source evidence and workflows.<\/p>\n<h2>Which vendor responses are red flags?<\/h2>\n<p>Treat a response as high risk when it prevents independent verification or obscures the population being measured.<\/p>\n<p>Watch for:<\/p>\n<ul>\n<li>\u201cAll major AI engines\u201d without a surface-level coverage matrix.<\/li>\n<li>API data presented as consumer-interface data without validation.<\/li>\n<li>Screenshots that omit the prompt, timestamp, surface, or market.<\/li>\n<li>Raw answers visible in the interface but unavailable in exports.<\/li>\n<li>Sentiment or recommendation scores without classification rules.<\/li>\n<li>Citation tracking that stores domains but not exact URLs.<\/li>\n<li>Failed runs that disappear from success-rate denominators.<\/li>\n<li>Silent retries that overwrite original attempts.<\/li>\n<li>Share-of-voice calculations with undisclosed competitors.<\/li>\n<li>Historical trends based on changing prompt cohorts.<\/li>\n<li>One proprietary score that cannot be reconstructed.<\/li>\n<li>Identical-looking results across engines without a collection explanation.<\/li>\n<li>Security claims unsupported by scoped documents.<\/li>\n<li>Standard capabilities that actually require paid services.<\/li>\n<li>Contract terms that restrict ownership or export of configured data.<\/li>\n<li>Refusal to run buyer-supplied prompts during evaluation.<\/li>\n<li>Pricing that cannot be normalized to the requested scope.<\/li>\n<\/ul>\n<p>Require the vendor to demonstrate the underlying record whenever a response relies on a dashboard screenshot or marketing claim.<\/p>\n<h2>How should the final buying decision be made?<\/h2>\n<p>The final decision should combine evidence integrity, weighted fit, proof-of-concept performance, security review, implementation effort, and normalized total cost.<\/p>\n<p>Use this sequence:<\/p>\n<ol>\n<li><strong>Exclude vendors that fail a mandatory gate.<\/strong><\/li>\n<li><strong>Score remaining vendors against the same 100-point model.<\/strong><\/li>\n<li><strong>Run the controlled POC with identical prompts and conditions.<\/strong><\/li>\n<li><strong>Recalculate selected metrics from exported records.<\/strong><\/li>\n<li><strong>Resolve discrepancies, limitations, and delivery commitments in writing.<\/strong><\/li>\n<li><strong>Complete security, privacy, legal, and integration reviews.<\/strong><\/li>\n<li><strong>Normalize year-one and renewal costs to the same usage assumptions.<\/strong><\/li>\n<li><strong>Document the final decision and any contractual conditions.<\/strong><\/li>\n<\/ol>\n<p>Retain these artifacts in the procurement record:<\/p>\n<ul>\n<li>Final vendor response.<\/li>\n<li>Completed scorecard.<\/li>\n<li>Raw POC export.<\/li>\n<li>Metric-reconstruction workbook.<\/li>\n<li>Security and privacy review.<\/li>\n<li>Pricing-normalization worksheet.<\/li>\n<li>List of exceptions and contract conditions.<\/li>\n<li>Implementation acceptance criteria.<\/li>\n<\/ul>\n<p>Re-run the mandatory gates before renewal and after material changes to collection methods, metric definitions, data exports, or engine coverage.<\/p>\n<h2>Common questions about AI visibility RFPs<\/h2>\n<h3>How many AI engines should an RFP require?<\/h3>\n<p>Require the engines and answer surfaces your buyers actually use. Many B2B evaluations consider ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, and Google AI Overviews, but depth matters more than logo count. Verify the interface, market, language, metadata, citations, and raw-answer access for each surface.<\/p>\n<h3>How long should an AI visibility proof of concept run?<\/h3>\n<p>For vendor data-quality testing, run the same 20-prompt cohort three times per engine on three separate days. This reveals collection and export defects without claiming to measure long-term visibility. A longer period is necessary when the objective is to estimate weekly or monthly market trends.<\/p>\n<h3>Can an API response represent what users see in an AI product?<\/h3>\n<p>Not automatically. APIs and consumer interfaces may differ in model configuration, retrieval, system instructions, personalization, location, and citations. A vendor should report the datasets separately unless it can provide a dated validation showing that the API reliably represents the specified consumer surface.<\/p>\n<h3>Can an AI visibility platform guarantee brand recommendations?<\/h3>\n<p>No. A platform can observe recommendations, preserve their evidence, diagnose recurring content or source gaps, and measure changes after optimization. It cannot control an independent AI engine or guarantee placement. Treat guarantees as a commercial red flag.<\/p>\n<h3>How much historical data should buyers require?<\/h3>\n<p>The retention period should cover the organization\u2019s planning, reporting, and renewal cycles. Twelve months is a practical starting requirement for year-over-year review, but buyers may need more. Confirm that retention includes raw answers, citations, metadata, prompt versions, and failures\u2014not only aggregated charts.<\/p>\n<h3>Who should own the RFP process?<\/h3>\n<p>SEO or marketing should define prompts, competitors, markets, and success metrics. Analytics should validate formulas and exports. IT, security, legal, and procurement should review integrations, data handling, service terms, and cost. Assign one accountable program owner to resolve cross-functional tradeoffs.<\/p>\n<h3>How should agencies adapt this AI visibility RFP template?<\/h3>\n<p>Agencies should add client-level workspace isolation, reusable taxonomies, granular permissions, white-label reporting, bulk exports, API limits, portfolio views, and client-specific retention controls. Pricing should be normalized per client, prompt, engine, and run rather than per seat alone.<\/p>\n<p><script type=\"application\/ld+json\">\n{\n  \"@context\": \"https:\/\/schema.org\",\n  \"@graph\": [\n    {\n      \"@type\": \"Article\",\n      \"headline\": \"AI Visibility RFP Template: 100-Point Vendor Scorecard\",\n      \"description\": \"Copy this AI visibility RFP template with pass\/fail gates, a 100-point scorecard, POC tests, pricing questions, and required vendor evidence.\",\n      \"author\": {\n        \"@type\": \"Organization\",\n        \"name\": \"maxaeo\"\n      },\n      \"publisher\": {\n        \"@type\": \"Organization\",\n        \"name\": \"maxaeo\"\n      },\n      \"image\": \"image-placeholder\"\n    },\n    {\n      \"@type\": \"FAQPage\",\n      \"mainEntity\": [\n        {\n          \"@type\": \"Question\",\n          \"name\": \"How many AI engines should an RFP require?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Require the engines and answer surfaces your buyers actually use. Coverage depth matters more than logo count: verify the interface, market, language, metadata, citations, and raw-answer access for every contracted surface.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"How long should an AI visibility proof of concept run?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"For vendor data-quality testing, run the same 20-prompt cohort three times per engine on three separate days. Use a longer period when the objective is to estimate weekly or monthly market trends.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Can an API response represent what users see in an AI product?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Not automatically. APIs and consumer interfaces may differ in model configuration, retrieval, system instructions, personalization, location, and citations. Require a dated equivalence study or report the datasets separately.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Can an AI visibility platform guarantee brand recommendations?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"No. A platform can monitor recommendations, diagnose source and content gaps, and measure changes, but it cannot control an independent AI engine or guarantee placement.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"How much historical data should buyers require?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Twelve months is a practical starting requirement for year-over-year review. Confirm that retention includes raw answers, citations, metadata, prompt versions, and failed runs rather than only aggregated charts.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"Who should own the AI visibility RFP process?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"SEO or marketing should define the measurement scope, analytics should validate calculations, and IT, security, legal, and procurement should review integrations, data handling, service terms, and cost. One accountable program owner should resolve tradeoffs.\"\n          }\n        },\n        {\n          \"@type\": \"Question\",\n          \"name\": \"How should agencies adapt an AI visibility RFP template?\",\n          \"acceptedAnswer\": {\n            \"@type\": \"Answer\",\n            \"text\": \"Agencies should add client-level workspace isolation, granular permissions, reusable taxonomies, white-label reporting, bulk exports, API limits, portfolio views, and pricing normalized per client, prompt, engine, and run.\"\n          }\n        }\n      ]\n    }\n  ]\n}\n<\/script><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Copy this AI visibility RFP template with pass\/fail gates, a 100-point scorecard, POC tests, pricing questions, and required vendor evidence.<\/p>\n","protected":false},"author":1,"featured_media":1551,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1553","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\/1553","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=1553"}],"version-history":[{"count":0,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts\/1553\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/media\/1551"}],"wp:attachment":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/media?parent=1553"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/categories?post=1553"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/tags?post=1553"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}