Table Of Contents
AI share of voice is the percentage of AI-generated responses, across a defined set of buyer-intent prompts, in which your brand is mentioned relative to the total number of responses sampled. Measuring it takes a defined prompt set, 20 to 30 runs per prompt per platform, and a competitor log built from the brands AI engines actually name. Keep mention rate, citation rate and share of voice separate, because each one tells you something different. No free tool shows whether ChatGPT or Gemini named your brand in an answer, so prompt-level tracking is the only complete method. Use the calculator below to run your own numbers, then use MAPS to close the gap.
If your brand isn't showing up when buyers ask AI engines for recommendations, you're losing pipeline to competitors who are. The question is: how do you measure that gap with enough precision to act on it?
That's what AI share of voice is designed to answer. It's a metric, a measurement process, and a diagnostic tool rolled into one. This guide walks through the full methodology: how to define the metric correctly, build a prompt set that reflects real buyer intent, run consistent tests across ChatGPT, Perplexity, Gemini, and Google AI Overviews, calculate a defensible number, and use that number to close the gap.
- A precise definition of AI share of voice and how it differs from mention rate and citation rate
- The formula, with a worked example
- A step-by-step process for building and running your prompt set
- Free tools that give you partial visibility right now
- The MAPS framework for improving your results
- Benchmarks, caveats, and a FAQ section
What Is AI Share of Voice?
AI share of voice is the percentage of AI-generated responses, across a defined set of buyer-intent prompts, in which your brand is mentioned relative to the total number of responses sampled.
It answers one question: out of all the times AI engines respond to the questions your buyers are actually asking, how often does your brand appear in the answer?
This is a fundamentally different measurement from traditional share of voice, which counts ad impressions or organic search rankings. AI engines don't rank pages in a list. They synthesize answers. Your brand either appears in that answer or it doesn't. The metric captures presence, not position.
Three metrics you need to keep separate
Most teams confuse three related but distinct measurements. Conflating them produces misleading numbers.
| Metric | What it counts | What it tells you |
|---|---|---|
| Mention rate | Responses where your brand name appears in the answer text, with or without a link | How often AI engines include you in the conversation |
| Citation rate | Responses where your domain is linked as a source | How often AI engines treat your content as evidence |
| AI share of voice | Your mention rate relative to all competitors mentioned across the same prompt set | Your competitive position in AI-generated answers |
A mention without a citation means the AI named your brand but didn't link to your content. A citation without a mention in the prose is rare but possible (a source link in the footnotes of an AI Overview, for example). Both matter, but they measure different things.
The practical implication: a brand can have a 20% mention rate and still have low AI share of voice if five competitors are each being mentioned in 40-60% of responses. The share calculation requires competitor data, not just your own numbers.
Why this metric matters now
AI engines are now a primary research channel for B2B buyers. When a plant manager or procurement lead asks ChatGPT "best industrial automation suppliers for mid-size manufacturers," the response shapes consideration before any website is visited. The same is true when an operations leader asks "top B2B payroll software for 500-employee companies" or "what CRM do manufacturers use." If your brand isn't in that answer, you don't exist in that buyer's evaluation set.
The gap is measurable. In an illustrative 30-day test across six priority buyer prompts, a B2B industrial equipment manufacturer appeared in responses to only one of six prompts, with a mention rate of 10% (12 of 120 responses) on that single prompt and zero mentions on the remaining five. Meanwhile, responses named dozens of alternative suppliers. That's not a ranking problem. That's an AI visibility problem, and it requires a different diagnostic.
The AI Share of Voice Formula
The formula has two layers: one for your own mention rate, and one for relative share across competitors.
Layer 1: Mention rate (your baseline)
Mention Rate = (Responses mentioning your brand ÷ Total responses sampled) × 100
Run each prompt a fixed number of times across each platform. A sample of 20-30 responses per prompt per platform gives you a workable baseline. More responses reduce variance; fewer introduce noise.
Layer 2: AI share of voice (the competitive number)
AI Share of Voice = Your brand mentions ÷ Total brand mentions across all tracked brands × 100
"Total brand mentions" is the sum of every mention of every brand (including yours) across the same prompt set. This requires tracking competitor appearances, not just your own.
Important caveat: this is a relative metric. If you track only three competitors but the AI is mentioning fifteen, your denominator is wrong and your share figure is inflated. Build your competitor list from what the AI actually names, not from who you think your competitors are.
Always report both. Share of voice without mention rate hides prompt-level gaps. Mention rate without share of voice hides competitive context.
Worked example
Setup: A B2B industrial equipment manufacturer tracks 3 buyer prompts, 40 responses each, across 4 platforms (ChatGPT, Perplexity, Gemini, Google AI Overviews). Total responses: 480 per prompt, 1,440 total. The figures below are an illustrative log, not a claim about any named company.
Results logged:
| Prompt | Your brand mentions | Competitor A mentions | Competitor B mentions | Competitor C mentions |
|---|---|---|---|---|
| "best industrial automation suppliers for mid-size manufacturers" | 0 | 22 | 18 | 31 |
| "who supplies industrial automation with remote monitoring for mid-size manufacturers" | 48 | 19 | 12 | 9 |
| "what automation systems do mid-size manufacturers use" | 0 | 34 | 27 | 16 |
| Total | 48 | 75 | 57 | 56 |
Total brand mentions across all brands: 48 + 75 + 57 + 56 = 236
Your AI share of voice: 48 ÷ 236 × 100 = 20.3%
Your mention rate across all prompts: 48 ÷ 1,440 × 100 = 3.3%
These two numbers tell different stories. A 20.3% share of voice sounds reasonable until you see that it's driven entirely by one prompt and that your mention rate is 3.3% overall. The share number flatters; the mention rate reveals the concentration risk.
What this means for how you read the data: always report both. Share of voice without mention rate hides prompt-level gaps. Mention rate without share of voice hides competitive context.
Every figure here is the illustrative mention count from the table above. Your brand wins one prompt and scores zero on the other two.
| Prompt | Your brand | Competitor A | Competitor B | Competitor C |
|---|---|---|---|---|
| "best industrial automation suppliers for mid-size manufacturers" | 0 | 22 | 18 | 31 |
| "who supplies industrial automation with remote monitoring for mid-size manufacturers" | 48 | 19 | 12 | 9 |
| "what automation systems do mid-size manufacturers use" | 0 | 34 | 27 | 16 |
This is the concentration pattern the article describes: appearing on one prompt rather than broadly.
Totals are 48, 75, 57 and 56, summing to 236 brand mentions across the prompt set. Share of voice divides by that 236. Mention rate divides by all 1,440 responses sampled.
| Measure | Value |
|---|---|
| Your brand, total mentions | 48 |
| Competitor A, total mentions | 75 |
| Competitor B, total mentions | 57 |
| Competitor C, total mentions | 56 |
| AI share of voice | 20.3% |
| Mention rate | 3.3% |
Same brand, same data, two very different-looking results. That difference is why the article insists on reporting both.
Separating mentions from citations in your log
For each response, record two binary flags:
- Mentioned (Y/N): your brand name appeared anywhere in the response text
- Cited (Y/N): your domain appeared as a linked source
The article defines the two diagonal cases. The other two simply restate the flags, because the article gives definitions here rather than citation figures, and nothing below is a measured rate.
No citation counts or rates appear in this figure. The article supplies definitions for these cases, not a quantitative citations dataset.
Calculate separate rates for each. A high citation rate with a low mention rate means the AI trusts your content but doesn't recommend your brand by name. A high mention rate with a low citation rate means the AI knows your brand but isn't grounding responses in your content. Both patterns require different fixes.
How to Build a Representative Prompt Set
The quality of your AI share of voice measurement is entirely determined by the quality of your prompt set. A prompt set built around vanity queries ("who is [your brand name]") tells you nothing useful. A prompt set built around real buyer questions tells you whether you're in the consideration set when it matters.
Start with buyer intent, not brand awareness
Your prompts should reflect the questions buyers ask during evaluation, not the questions you wish they were asking. Pull from three sources:
- Sales call notes and CRM data: what questions do prospects ask before they convert? What alternatives do they mention? These are the prompts that matter most.
- Search Console queries: your top informational and commercial queries reveal what buyers research before engaging. Filter for question-format queries.
- Competitor gap analysis: run a handful of category prompts in ChatGPT and note every brand that appears. Those are the prompts where you need to compete.
Prompt categories for B2B brands
A well-structured prompt set covers four intent categories:
| Category | Example prompt structure | Why it matters |
|---|---|---|
| Category/vendor discovery | "Best [category] for [industry or company size]," e.g. "best industrial automation suppliers for mid-size manufacturers" or "top B2B payroll software for 500-employee companies" | Captures top-of-funnel AI recommendations |
| Problem-solution | "How do I [solve specific problem] for a [company type]" | Captures mid-funnel advisory responses |
| Comparison | "Compare [your category] vs [adjacent category]" | Captures decision-stage evaluations |
| Product- or solution-specific | "What [product type] do [industry] use," e.g. "what CRM do manufacturers use" | Captures high-intent, narrow queries |
Prompt set size and sampling guidelines
- Minimum viable set: 10-15 prompts covering at least two intent categories
- Recommended set: 25-40 prompts across all four categories
- Sample size per prompt per platform: 20-30 responses (run the same prompt multiple times; AI responses vary)
- Platforms to cover: ChatGPT, Perplexity, Gemini, and Google AI Overviews at minimum. Add Microsoft Copilot if your buyers are in enterprise environments.
Why sample size matters: AI engines don't return identical answers every time. The same prompt can produce different brand mentions across runs, especially on ChatGPT and Gemini. A single test per prompt is not a data point; it's an anecdote. Twenty or more runs per prompt per platform gives you a stable rate.
Prompt phrasing discipline
Use the same prompt wording across every platform and every measurement period. Even small changes in phrasing (adding "in 2026," changing "best" to "top," switching from first-person to third-person) can shift which brands appear. If you change a prompt, treat it as a new prompt and start a fresh baseline.
Write prompts the way buyers actually phrase questions to AI, not the way you'd write a search query. Conversational phrasing ("what's the best way to...") tends to trigger more brand-specific recommendations than keyword-style phrasing ("industrial automation suppliers Chicago" or "B2B payroll software 500 employees").
How to Track Results Consistently Across Platforms
Each AI platform has different behavior, different citation patterns, and different response structures. A tracking process that treats them identically will produce misleading comparisons. Here's how to handle each one.
Platform-by-platform tracking notes
ChatGPT responses vary significantly across runs. The same prompt can return different brand lists depending on the model version, time of day, and browsing mode (with or without web access). Track with web browsing enabled for vendor discovery prompts; disable it for advisory prompts to test the model's trained knowledge. Log which mode you used.
Perplexity is citation-heavy by design. Every response includes source links, making citation tracking straightforward. Mention rate and citation rate tend to be more correlated here than on other platforms. Note the "Focus" setting used (All, Web, Academic, etc.); different settings surface different sources.
Gemini responses often pull from Google's index and can reflect recent content changes faster than other platforms. Track with the default model; avoid experimental modes for consistency. Gemini is more likely to cite Google-indexed sources, so citation rate here is a partial proxy for your organic search authority.
AI Overviews appear inline in Google Search and don't trigger on every query. Track two things separately: (1) whether an AI Overview appeared at all for your prompt (trigger rate), and (2) whether your brand was mentioned or cited when one did appear. A low trigger rate on your prompts is itself a signal worth monitoring.
Note on Google AI Overviews and AI Mode: as of August 31, 2026, Google Search Console provides a Generative AI performance report showing impressions of your pages in AI Overviews and AI Mode, combined. This is first-party data on link impressions only, not brand mentions in the answer text, and not clicks or queries. See the free tools section below for details on what Search Console does and doesn't show.
Building your tracking log
Use a consistent spreadsheet or database with these columns per response:
| Column | What to record |
|---|---|
| Prompt ID | Reference to your prompt library |
| Platform | ChatGPT / Perplexity / Gemini / Google AI Overviews |
| Run date | Date and time of the test |
| Response # | Run number for this prompt on this platform |
| Brand mentioned (Y/N) | Did your brand name appear in the response text? |
| Brand cited (Y/N) | Did your domain appear as a linked source? |
| Competitor brands named | List every other brand mentioned |
| Competitor domains cited | List every other domain linked |
| Notes | Unusual response, format change, error |
Measurement cadence
- Weekly: run your full prompt set on all platforms. This is the minimum cadence for detecting meaningful changes.
- Monthly: calculate your mention rate, citation rate, and AI share of voice across the full month's data. Monthly aggregates smooth out run-to-run variance.
- Quarterly: review and update your prompt set. Remove prompts that never trigger AI Overviews. Add new prompts that reflect emerging buyer questions.
Do not compare across platforms directly. A 15% mention rate on Perplexity and a 15% mention rate on ChatGPT are not equivalent. Platform audiences differ, citation behaviors differ, and response structures differ. Report platform-level rates separately, then aggregate for an overall figure with a clear note that it's a cross-platform average.
Free Tools to See Whether AI Mentions Your Brand
You don't need a paid platform to start building visibility. Several free tools provide partial, indirect signals that are worth setting up before you invest in anything else. The key word is "partial": no free tool shows whether ChatGPT or Gemini named your brand in an answer. What these tools do tell you is whether AI engines are sending traffic to your site, whether your links appeared in Google's AI features, and, in Microsoft's case, whether your pages were cited as sources in Bing-grounded experiences such as Copilot.
The signal distinction that matters: referral traffic and bot activity are downstream indicators. They tell you that AI engagement happened. Microsoft Clarity's Citations report and Bing Webmaster Tools' AI Performance report (still in preview) come closest to a share-of-voice view, but only for Bing-grounded experiences, and they measure citations as a source, not uncited brand-name mentions. Prompt-level mention tracking is the only method that tells you whether your brand appeared in the answer text itself across ChatGPT, Gemini, and the rest of the set. Use the free tools as supporting data, not as a substitute for prompt tracking.
Google Search Console: Generative AI performance report
As of August 31, 2026, Google Search Console includes a dedicated Generative AI performance report for all websites worldwide. Google's help documentation describes the Search report as impressions only: it shows how often links to your pages appeared in Google AI Overviews and AI Mode, combined, with dimensions for pages, countries, dates, and devices.
What it shows:
- Impressions of your URLs in AI Overviews and AI Mode (combined, not separated)
- Which of your pages appeared as sources
- Trends over time, by country and device
What it does not show:
- Whether your brand was named in the answer text
- Which queries triggered the AI feature (queries are not in this report)
- Clicks, CTR, or average position
- Competitor pages cited in the same responses
- Mentions or citations in ChatGPT, Gemini's standalone app, or other third-party assistants
The report only appears for properties with sufficient impression volume, so smaller sites may not see it immediately. Search Labs experiment data is excluded. If the report is available for your property, check it monthly as a trend indicator. A rising impression count tells you your content is being surfaced in Google's AI features; it doesn't tell you whether buyers are being recommended your brand.
Also in Search Console: the standard Performance report includes AI Overviews and AI Mode data under the "Web" search type, mixed in with regular organic clicks and impressions. There is currently no official way to isolate AI-driven clicks from standard organic clicks in that report. The dedicated Generative AI report is the official way to see AI-feature impressions on their own.
GA4: AI Assistant channel
As of May 13, 2026, Google Analytics 4 includes a dedicated AI Assistant default channel that classifies users who arrive from sources such as ChatGPT, Gemini, Copilot, Claude, Deepseek, and Grok. The official rule sets the medium to ai-assistant and the campaign to (ai-assistant) when the referrer matches Google's list of AI assistants. Google announced the change in What's new in Google Analytics. No manual setup is required for sources Google already recognizes.
Important split: clicks from Google's own AI Overviews and AI Mode stay in Organic Search. They are not included in the AI Assistant channel, and GA4 has no official dimension that isolates them.
Practical setup: in Acquisition reports, filter Session default channel group to AI Assistant. You can also review Session source / Session source/medium for hostnames such as chatgpt.com, gemini.google.com, copilot.microsoft.com, and claude.ai. Track session volume, engagement rate, and goal completions from this channel month over month. An increase in AI Assistant traffic is a positive signal that your content is being cited and clicked.
Note on Perplexity: Google's published AI Assistant channel description names ChatGPT, Gemini, Deepseek, Copilot, and Grok (and Claude in the what's-new note). It does not name Perplexity. Google later added a Source Group dimension that includes Perplexity, but that is a source-grouping feature, not confirmation that Perplexity sits inside the AI Assistant default channel. Verify in your own property whether Perplexity traffic lands in AI Assistant or Referral.
Limitation: GA4 only captures sessions where a user clicked a link from a recognized AI assistant (or arrived with the ai-assistant medium). It does not capture brand mentions where no link was clicked, which is the majority of AI mentions. Hidden referrers can appear as Direct.
Microsoft Clarity: AI channel groups, citations, and bot activity
Microsoft Clarity (free) includes three features directly relevant to AI visibility tracking.
AIPlatform channel group: Clarity classifies sessions in which a visitor clicked a link inside a generative AI chat into a dedicated "AIPlatform" channel, covering ChatGPT, Copilot, Gemini, Perplexity, and Claude. A separate PaidAIPlatform channel covers paid ads inside those chats. Find both under the Referrer card in your Clarity dashboard, switching to the Channels tab. Microsoft's documentation covers the full setup. This lets you measure AI-referred session volume, compare engagement against other channels, and view session recordings from AI-referred visitors. Clarity counts only direct visits from standalone AI websites (for example, chatgpt.com or copilot.microsoft.com); AI features embedded in search engines or productivity tools are not included. If the referrer is hidden, traffic may appear as Direct.
AI Visibility Citations report: This is the closest free tool to a share-of-voice view, with an important limit: it covers Bing-grounded experiences only, not ChatGPT. Generally available since May 13, 2026, the Citations report shows how pages on a verified domain were referenced as sources in Microsoft Copilot, Bing AI summaries, and select Bing-grounded partner experiences. Official metrics include page citations, share of authority (your domain's share of citations versus other cited domains on overlapping queries), grounding queries, cited pages, and AI referral traffic. In August 2026, Clarity added branded versus non-branded query segmentation; in September 2026, it added customizable Brand Terms so you can group abbreviations, spellings, and localized names. Find it under Dashboards, then AI Visibility, then Citations. A project admin must verify domain ownership first.
What Citations does not show: uncited brand-name mentions, or whether ChatGPT, Gemini, or other non-Bing-grounded assistants named or cited you. Treat it as Microsoft's first-party citation report, not universal AI share of voice.
AI Visibility Bot Activity: After you connect a supported CDN or use the latest official WordPress plugin, Clarity's Bot Activity dashboard uses server-side logs to show how verified AI crawlers request your content: bot operator, AI request share, activity by purpose, and the pages most requested. Supported CDNs are Fastly, Amazon CloudFront, Cloudflare, Azure Front Door, and Akamai. Microsoft is explicit that bot activity represents requests only and does not indicate that content was retrieved, grounded, cited, or surfaced in an AI-generated response.
What Clarity does not show overall: whether ChatGPT or Gemini named your brand in an answer. Citations come closest for Copilot and other Bing-grounded experiences. Bot activity confirms your content was requested; it does not confirm it was cited or that your brand was recommended.
Bing Webmaster Tools: AI Performance (public preview)
Bing Webmaster Tools' AI Performance report has been in public preview since February 10, 2026. It measures how your content was cited as a visible source in Microsoft Copilot, AI-generated summaries in Bing, and select partner AI integrations, the same Microsoft grounding ecosystem that powers Clarity Citations.
Official views include total citations, cited pages, grounding queries, and visibility trends. Follow-on capabilities announced as preview on June 16, 2026 include Intents, Topics, Citation Share (your site's percentage of all citations for a given grounding query; observational, and it does not expose competitor domains), and Compare. Topics Portfolio, added October 13, 2026, is still in beta: a personalized topic list with directional figures for topic size, sites cited, and opportunity queries where other sites appear and yours does not.
Mark these features accordingly: the core report family remains in preview, and Topics Portfolio is beta. Data is aggregated and sampled, not a complete log of every citation. This report does not claim coverage of ChatGPT or Google AI Overviews.
Cloudflare AI Crawl Control (free tier)
If your domain is on Cloudflare and proxied, AI Crawl Control is available on the free plan with no extra product configuration. The free tier detects well-known AI crawlers by user-agent, lets you allow or block individual crawlers, and shows operator-level metrics for a maximum of 24 hours. It is a crawler-access tool, not a mention or citation report. Referral views (traffic from domains such as chatgpt.com) are documented as paid-plan only.
What the free stack tells you (and what it doesn't)
| Tool | Brand mentions in the answer | Citations as a source | Clicks and traffic | Crawler activity |
|---|---|---|---|---|
| Search Console Generative AI report | No. The report does not show whether your brand was named in the answer text. | Limited. Impressions of your URLs as sources in AI Overviews and AI Mode only, combined. | No. No clicks, CTR or average position in this report. | No. Not a crawler report. |
| GA4 AI Assistant channel | No. Does not capture brand mentions where no link was clicked. | No. Not a citation report. | Yes. Sessions arriving from recognized AI assistants. | No. Not a crawler report. |
| Microsoft Clarity AIPlatform channel | No. Does not show whether the brand was recommended. | No. Not a citation report. | Yes. AI-referred session volume and engagement from standalone AI sites. | No. Not a crawler report. |
| Clarity AI Visibility Citations | No. Does not show uncited brand-name mentions. | Limited. Bing-grounded experiences only, not ChatGPT. | Yes. AI referral traffic is an official metric of this report. | No. Bot activity is a separate dashboard. |
| Bing Webmaster Tools AI Performance | No. Does not claim coverage of brand mentions in answer text. | Limited. Public preview, Bing-grounded only, aggregated and sampled. | No. Citation reporting, not a traffic report. | No. Not a crawler report. |
| Clarity Bot Activity and Cloudflare AI Crawl Control | No. Crawling is not a mention. | No. Requests only. Does not indicate content was cited. | No. Cloudflare referral views are paid-plan only. | Yes. Which verified AI crawlers request your content. |
| Signal | Tool | What it confirms | What it misses |
|---|---|---|---|
| Pages cited in Google AI features | Search Console Generative AI report | Your URLs appeared as sources in AI Overviews and AI Mode (impressions only, combined) | Brand mentions, queries, clicks, competitor data, third-party assistants |
| AI referral clicks | GA4 AI Assistant channel | Users clicked through from recognized AI assistants (ChatGPT, Gemini, Copilot, Claude, and others) | Mentions without clicks; AI Overviews/AI Mode traffic (stays in Organic Search); confirm Perplexity in-property |
| AI-referred session behavior | Microsoft Clarity AIPlatform channel | Engagement quality of AI traffic from standalone AI sites | Whether brand was recommended; embedded AI features |
| Citations in Copilot / Bing AI | Clarity AI Visibility Citations | Your pages were cited as sources, plus share of authority, grounding queries, and branded vs non-branded splits | Uncited brand mentions; ChatGPT, Gemini, and other non-Bing-grounded assistants |
| Citations in Copilot / Bing AI (preview) | Bing Webmaster Tools AI Performance | Citation volume, Citation Share, and (beta) Topics Portfolio in Bing-grounded experiences | Same coverage limit as Clarity Citations; report family is still preview |
| AI crawler activity | Clarity Bot Activity; Cloudflare AI Crawl Control (free: last 24 hours) | Which AI crawlers request your content | Citation or mention outcome (crawl ≠ citation) |
The free stack is a starting point. It tells you that AI engagement is or isn't happening, and Microsoft's citation reports come closest for Copilot and Bing AI. No free tool shows whether ChatGPT or Gemini named your brand. To know whether your brand is winning or losing in AI-generated answers across the platforms your buyers use, you need prompt-level tracking against a defined competitor set.
Benchmarks and Caveats
There are no universal benchmarks for AI share of voice. Anyone claiming otherwise is either working from a narrow sample or selling you something. What exists are directional reference points that are useful for calibration, not for absolute judgment.
What the data suggests (directionally)
Based on tracked prompt data across B2B product and service categories, a few patterns emerge:
- Most B2B brands start at or near zero. The majority of B2B companies have never run a structured prompt test and have no baseline. A 0% mention rate on your first measurement is normal, not alarming.
- Mention rates above 30% on any single prompt are relatively strong for a non-dominant brand in a competitive category. Rates above 50% suggest meaningful AI authority for that specific topic.
- Cross-prompt concentration is common. Brands that appear frequently tend to appear on 2-3 specific prompts rather than broadly. Broad coverage across 10+ prompts is rare and typically belongs to category-defining brands.
- Citation rates lag mention rates. It's common to be mentioned more often than you're cited. The gap between the two tells you whether AI engines are recommending your brand from trained knowledge or from your indexed content.
These are directional observations, not benchmarks. Your category, competitive density, prompt set composition, and platform mix all affect what "good" looks like. A 15% mention rate in a category with 50 named competitors is a very different result from a 15% mention rate in a category with 5.
Caveats you must disclose when reporting
1. Compare like with like. Never compare a mention rate calculated on 10 responses to one calculated on 120. Never compare a Perplexity-only rate to a cross-platform average. Specify your sample size, prompt set, and platform mix in every report.
2. AI responses are non-deterministic. The same prompt on the same platform on the same day can return different results. This is not a measurement error; it's a property of large language models. It's why you need 20+ runs per prompt, not one.
3. Platform weights are arbitrary. If you aggregate across ChatGPT, Perplexity, Gemini, and Google AI Overviews, you're implicitly weighting each platform equally unless you adjust. In most B2B contexts, the platforms your buyers actually use should receive higher weight. You don't necessarily know that without user research.
4. Your prompt set is not the universe. A strong mention rate on your 25 tracked prompts tells you nothing about the thousands of prompts you didn't track. Think of your prompt set as a sample of buyer intent, not a complete picture.
5. Results shift. AI models are updated continuously. A mention rate measured in Q1 may look different in Q3 with no changes to your content, because the model changed. Trend data over multiple measurement periods is more meaningful than a single snapshot.
How to Improve Your AI Share of Voice: The MAPS Framework
Measuring AI share of voice tells you where you stand. The MAPS framework tells you what to do about it. MAPS is a four-pillar structure for building the content and credibility signals that AI engines use when deciding which brands to include in their answers.
Each pillar addresses a different reason why AI engines might be ignoring your brand.
M
Pillar 1: Model Buyer Intent
AI engines recommend brands that are associated with specific, well-defined problems. If your content is written around your products or services rather than your buyers' questions, you're invisible to the prompts that matter.
What this means in practice:
- Map your content to the exact questions buyers ask at each stage of evaluation. Not "what we do" but "how do you solve [specific problem] for a [specific role] in [specific industry]."
- Use the same language buyers use. If your buyers ask "what's the best industrial automation setup for a mid-size plant" or "what CRM do manufacturers use," your content should use that phrasing, not a sanitized version of it.
- Build content that answers the category-level questions your prompt set covers. If you have zero content addressing "how mid-size manufacturers should evaluate industrial automation suppliers" or "what CRM do manufacturers use," don't expect to appear when those prompts are run.
The diagnostic: run your prompt set and identify every prompt where your mention rate is zero. For each one, ask: do we have any content that directly answers this question? If not, that's your first content priority.
A
Pillar 2: Answer Clearly
AI engines extract answers from content. They favor content that states a clear, direct answer in the first paragraph of a section, not buried in the middle of a 1,500-word piece.
What this means in practice:
- Structure content with direct, self-contained answers at the top of each section. The first 40-60 words of any section should be extractable as a standalone answer.
- Use question-format headings for informational content. "What is AI share of voice?" performs better as an extraction target than "Understanding AI Visibility Metrics."
- Avoid hedging language in definitions and explanations. "AI share of voice is approximately the percentage of..." is weaker than "AI share of voice is the percentage of..."
- Include explicit definitions, formulas, and step-by-step processes. These are the formats AI engines cite most reliably.
The diagnostic: take your best-performing content page and read only the first sentence of each section. If those sentences don't form a coherent, useful set of answers, the content isn't structured for AI extraction.
P
Pillar 3: Prove & Place
AI engines cite sources they trust. Trust is built through third-party validation: mentions on authoritative sites, inclusion in ranked lists and comparisons, and consistent association with the category you want to own.
What this means in practice:
- Pursue coverage on the types of sources AI engines actually cite for your category. In B2B markets, this typically includes review platforms (G2, Capterra), industry association directories, ranked lists in trade publications, and comparison guides on high-authority domains.
- Get your brand into the formats that AI engines favor. For B2B categories, ranked lists and comparison articles are consistently cited more than general blog posts. If you're not in those formats, you're not in the citation pool.
- Build your entity footprint: consistent brand name, consistent description, consistent association with your core category across your own site, LinkedIn, press mentions, and third-party profiles.
The diagnostic: search your brand name in Perplexity. What sources does it cite? If the top citations are your own website and nothing else, you have a third-party authority gap. Competitors with strong AI share of voice are typically cited from multiple external sources.
S
Pillar 4: Structure & Stay Fresh
AI engines favor content that is well-structured and recently updated. Stale content and poorly structured pages are less likely to be extracted as sources, even if the underlying information is correct.
What this means in practice:
- Use semantic HTML structure: H1, H2, H3 in logical hierarchy. AI engines parse document structure when extracting content.
- Add schema markup for your key pages: FAQ schema, HowTo schema, and Article schema increase the likelihood of structured extraction.
- Update high-priority content regularly. A guide last updated in 2023 competes poorly against a guide updated in 2026 for the same query.
- Remove outdated statistics, dead links, and deprecated information. Content quality signals matter for AI citation decisions.
The diagnostic: check the last-modified date on your top 10 content pages. If the majority haven't been updated in 12+ months, you have a freshness problem that's compounding your AI visibility gap.
Applying MAPS to a real gap
Consider a B2B industrial equipment manufacturer tracking the prompt "best industrial automation suppliers for mid-size manufacturers" with a 0% mention rate across 120 responses. The figures are illustrative. The MAPS diagnosis:
- Model Buyer Intent: do they have content addressing how mid-size manufacturers should evaluate industrial automation suppliers? If not, start there.
- Answer Clearly: is that content structured with a direct definition and clear value proposition in the first paragraph? If it buries the lede, rewrite the opening.
- Prove & Place: are they listed on G2, in industry association directories, or in trade publication comparisons? Are they included in any "best industrial automation suppliers" roundup articles? If not, that's the placement gap.
- Structure & Stay Fresh: when was the relevant content last updated? Is FAQ schema implemented? If the page is two years old with no schema, that's a structural disadvantage.
A 0% mention rate on a single prompt rarely has a single cause. MAPS gives you a structured way to diagnose which of the four levers is most broken.
Frequently Asked Questions
How is AI share of voice different from traditional share of voice?
Traditional share of voice measures your brand's presence in paid media, organic search rankings, or social conversations relative to competitors. AI share of voice measures your brand's presence in AI-generated answers. The key difference is that AI engines synthesize recommendations rather than rank pages, so the metric is binary per response (mentioned or not) rather than positional.
How many prompts do I need to get a meaningful measurement?
A minimum of 10-15 prompts covering at least two intent categories gives you a baseline. For a reliable competitive picture, 25-40 prompts across all four intent categories (discovery, problem-solution, comparison, product- or solution-specific) is the recommended range. More prompts reduce the risk that your overall number is driven by performance on one or two outlier queries.
Do I need to test every AI platform?
Not necessarily. Start with the platforms your buyers actually use. For most B2B audiences, ChatGPT and Perplexity are the highest-priority starting points, with Gemini and Google AI Overviews added as a second layer. The important rule is to track platforms separately and never aggregate without disclosing your platform mix.
Can I improve AI share of voice without creating new content?
Yes, partially. Structural improvements (schema markup, heading hierarchy, clearer section openings) and third-party placement (G2 or industry directory listings, trade publication inclusions, analyst mentions) can improve citation rates without new content. But if your brand is absent from AI answers because you have no content addressing the relevant buyer questions, structural fixes alone won't close the gap. Both content and structure need to be addressed.
How long does it take to see results after making changes?
It varies by platform and by the type of change. Perplexity and Gemini tend to reflect content changes faster than ChatGPT, which relies more heavily on training data. Third-party placements (getting listed in a ranked article) can show up in Perplexity within days of the source being indexed. Changes to your own content may take weeks to months to influence AI responses. Measure monthly and look for trend direction rather than immediate jumps.
What's the difference between a brand mention and a brand citation in AI responses?
A mention is when your brand name appears in the response text. A citation is when your domain appears as a linked source. You can be mentioned without being cited (the AI names your brand but doesn't link to your content) and cited without being prominently mentioned in the prose (a source link in an AI Overview without your brand name in the answer text). Track both separately; they indicate different things about your AI authority.
Is there a free way to see if AI is citing my content?
Yes, with limitations. Google Search Console's Generative AI performance report (available worldwide since August 31, 2026) shows impressions of your pages in AI Overviews and AI Mode, combined; it is impressions only, with no clicks or queries in that report. Microsoft Clarity's AI Visibility Citations report (generally available since May 13, 2026) and Bing Webmaster Tools' AI Performance report (public preview since February 10, 2026) show when your pages were cited as sources in Copilot and other Bing-grounded experiences. Clarity's AIPlatform channel and GA4's AI Assistant channel (launched May 13, 2026) show sessions arriving from AI platforms; AI Overviews and AI Mode clicks in GA4 stay in Organic Search. No free tool shows whether ChatGPT or Gemini named your brand in the answer text, or how you compare to competitors across those assistants. That requires prompt-level tracking.
How do I know which competitors to include in my share of voice denominator?
Build your competitor list from what AI engines actually name in response to your prompts, not from your internal competitive set. Run your prompt set once and log every brand that appears. Those are your AI competitors for this measurement. The list often differs from your perceived competitive set and is more accurate for share of voice purposes.
Start with Your Baseline
AI share of voice is a measurable number. It requires a defined prompt set, consistent sampling, and a competitor log, but it's not complex. The brands that will build AI visibility over the next 12 months are the ones that start measuring now, when most competitors haven't.
The first step is establishing your baseline: run your prompt set, log the results, and find out where you actually stand. That number, however uncomfortable, is more useful than any estimate.
If you want a structured starting point, the OneIMS GrowthScore assesses your current AI visibility across the dimensions that matter most, including prompt-level presence, content structure, and third-party authority signals. It gives you a scored baseline and a prioritized action plan built around the MAPS framework.
The measurement is the easy part. What you do with the data is where the work begins.
Find out where you actually stand.
Get a scored baseline across prompt-level presence, content structure, and third-party authority signals, plus a prioritized action plan built around MAPS.
Get My GrowthScore