What Is Share of Model and How Is It Different from Share of Voice?
Last updated: July 6, 2026
← Guide 8: Visibility Measurement
Share of Model replaces Share of Voice as the primary visibility metric in AI-mediated discovery. A company can have 40% Share of Voice and 0% Share of Model. SoM predicts whether you appear in AI-generated buying conversations. Share of Voice does not.
Signal: Spend, impressions, mentions
Predicts: Human brand awareness
Signal: Entity authority, semantic coherence, citations
Predicts: AI-generated answer inclusion
Recommendation share = revenue signal
Most teams track one and call it the other
SOV tracks human attention. SoM tracks machine recommendation. They are not correlated. Optimizing for one does not move the other.
Table of Contents
Why does Share of Voice fail to measure AI visibility?
Share of Voice answers a question buyers are no longer asking. It measures how often humans encounter your brand in paid and earned media. AI systems don’t check impression counts. They check whether your entity has sufficient authority signals to surface in a generated answer.
You can double your Share of Voice and see zero change in Share of Model. Different inputs. Different outputs. Different mechanisms entirely.
What Share of Voice actually measures
Share of Voice is your brand’s percentage of total category exposure, measured in media spend, impressions, mentions, or organic search presence, relative to competitors. Nielsen defines it as a brand’s percentage of total media expenditure in a category. The formula is consistent: your metrics divided by total market metrics, expressed as a percentage.
It answers a real question: how much of the human-facing conversation does your brand own?
That was the right question for twenty years. Buyers are no longer asking it. Their first research move is opening OpenAI’s ChatGPT and typing “best [category] tools for [use case].” Share of Voice tells you nothing about what happens next.
What AI systems actually check
AI systems don’t retrieve content based on how many impressions your campaign generated last quarter. They surface companies based on a probabilistic assessment of relevance and authority. That assessment draws from training data, entity relationships, and live web citations in the case of retrieval-augmented systems like Perplexity AI.
The signals that move AI responses are structural:
- How clearly your entity is defined across the web.
- How consistently your category language is used.
- Whether third-party sources connect your company to the right problems.
- Whether your content is formatted in ways AI can extract and reproduce with confidence.
A 2025 analysis of 680 million citations found three signals that correlate strongly with AI citation rates:
- Brand search volume is the strongest predictor of LLM citations, with a correlation of 0.334. That outweighs traditional backlinks.
- Adding statistics to content increases AI visibility by 22%.
- Using original quotations from named experts increases visibility by 37%.
None of those inputs show up in a Share of Voice report.
The gap. Why high SOV companies still disappear in AI answers
In 13 years running TreDigital, the version of this problem I saw most often was a company that had done everything right by traditional standards. Strong SEO. Consistent PR. High impression share. Recognizable brand in their category.
Then generative AI entered their buyers’ research process. The company’s AI mentions dropped to near zero. Their competitors were being recommended by name. Smaller. Less well-funded. Lower traditional SOV. The gap wasn’t in awareness. It was in whether AI could classify them at all.
AI systems couldn’t classify them reliably. Their category language was inconsistent across surfaces. Their entity signals were fragmented. Their content was written for human readers, not for machine extraction. This is what we call Identity Fragmentation at the measurement layer. You look visible to humans. AI can’t figure out what you do.
High SOV. Zero SoM. That’s the gap.
Search era: visibility = ranking. AI era: visibility = recommendation. Google rankings do not predict AI citations. Neither does Share of Voice. Understanding where Share of Model fits relative to SEO, AEO, and GEO is the starting point.
What is Share of Model?
Share of Model is the percentage of relevant AI-generated responses that include your company. Measured across a standardized prompt set, across multiple AI platforms, tracked over time.
Share of Model is not a standalone metric. It is the output of the entire Algorithmic Authority Stack. Every layer (identity, expertise, semantics, content, distribution, trust, measurement) either builds or destroys your SoM score. You can’t optimize the number directly. You fix the layers. The number follows.
Run 50 buyer-intent prompts across OpenAI’s ChatGPT. Your company appears in 9 of them. Your SoM on that prompt set, on that platform, on that date is 18%. That number means nothing in isolation. It means everything compared to your competitors on the same prompts, or compared to your own baseline three months ago.
Guide · Layer 5 of 7
Why Does AI Cite My Competitors Instead of Me?
Your content exists. AI cites your competitors anyway. The off-page signal sequence that builds Algorithmic Touchpoint Presence. G2, Reddit for B2B AI visibility, trade press, and directory presence. Platform by platform.
Read the full guide →The three variables that define your SoM
Three inputs drive your score. Each one must be measured separately:
- Prompt coverage. Do you appear across category, problem, and evaluation queries?
- Platform distribution. Do you appear on ChatGPT, Perplexity, Gemini, and Claude? Or just one?
- Response position and sentiment. Are you recommended, listed, or mentioned in passing?
Prompt coverage. How many of the queries your buyers type into AI systems mention your company in the response? A complete prompt set includes category-level queries (“best [category] platforms for mid-market manufacturing”), problem-level queries (“how do I solve [specific problem]”), and evaluation queries (“compare [your company] vs [competitor]”). Missing any of these creates blind spots.
Platform distribution. Your SoM on OpenAI’s ChatGPT and your SoM on Perplexity AI are not the same number. They often diverge significantly because each platform uses different training data, retrieval architecture, and citation logic. A 2025 study analyzing 5.5 million responses found that Perplexity AI cites sources in almost every response. OpenAI’s GPT-4o cites far less frequently. A separate analysis of 680 million citations found that only 11% of domains are cited by both ChatGPT and Perplexity. A brand that appears consistently in one may be nearly absent in the other. One platform number is not your SoM. It’s one data point. This is the failure mode we call Surface Dependency. Measuring one platform and assuming the others will match.
Response position and sentiment. Named first is not the same as named fifth. Named with a recommendation (“Company X is particularly well-suited for mid-market manufacturing workflows”) is not the same as named in passing. Named incorrectly is worse than not being named at all.
Mention share vs recommendation share
This is the distinction most AI visibility content completely misses. It needs to be precise.
Mention share is how often your company’s name appears in AI-generated responses. Regardless of context or characterization.
Recommendation share is how often your company is actively recommended as the preferred option for the buyer’s specific use case.
Most companies measure mention share and call it SoM. They’re measuring the easier number, not the one that matters.
SOV tracks human attention. SoM tracks machine recommendation. Within SoM: mention share is visibility. Recommendation share is the revenue signal.
A company can appear in an AI response as a historical footnote, a cautionary comparison, or a generic list item. None of those mentions predicts buyer consideration. Recommendation share is being actively framed as the right answer for the buyer’s specific problem. That’s what moves pipeline. That’s what you need to track separately.
Citation Share measures whether AI cites your domain. Share of Voice measures how often your brand gets mentioned. Share of Model measures whether you appear across ChatGPT, Perplexity, Gemini, and Claude, or concentrate in one. Tools track the first two. The third is where structural risk lives.
Why your SoM differs across platforms
| Platform | Citation behavior | What it prioritizes | Implication for SoM |
|---|---|---|---|
| OpenAI’s ChatGPT | Cites infrequently in standard responses; more in web-search mode | Brand search volume, entity clarity, training data coverage | Hardest to move fast. Reflects longer-term authority signals. |
| Perplexity AI | Cites in almost every response; highest citation frequency of any major platform | Recency, structured content, third-party coverage | Most responsive to content changes. |
| Google Gemini | Variable citation; applies Google’s E-E-A-T framework | Expertise signals, authoritative sources, entity validation | Responds to same signals as Google’s organic results. |
| Anthropic’s Claude | Prioritizes data-backed, precisely worded sources; deprioritizes SEO-optimized thin content | Concrete facts, named sources, dated claims | Penalizes generic content more severely than other platforms. |
This divergence is structural. It’s why a single-platform SoM number is not a strategy. Your buyers use multiple platforms for different stages of their research. Your SoM measurement needs to reflect that.
What happens when SoM is high but the model is wrong
High SoM with inaccurate characterization is a liability, not an asset. A company can appear consistently in AI responses, but be associated with a product line deprecated in 2023, or a pricing model that no longer exists, or a target segment they moved away from two years ago. That’s not visibility. That’s misinformation at scale.
If OpenAI’s ChatGPT mentions your company consistently but positions you for a segment you don’t serve, that mention is damaging. This is the hallucination problem applied to brand measurement. The fix isn’t to reduce your SoM. It’s to correct the underlying entity signals so the model’s characterization becomes accurate. You can’t fix what you don’t measure. Any SoM tracking system needs to log not just whether the brand appeared, but what the model said about it.
New data puts a number on the downstream effect. Similarweb found AI-recommended brands saw 2.5x more site visits, and 55.9% of that traffic arrived through branded search, not an AI referral link (Search Engine Journal, June 26 2026). Share of Model rises before your referral report moves. Track branded search volume as the leading proxy. Aleyda Solis named the blind spot: measuring AI impact through AI referral traffic alone misses demand that lands in Search and Direct.
Monthly Intelligence Report
What changed in AI retrieval this month.
One brief. The patterns your competitors aren’t tracking yet. Covers ChatGPT, Perplexity, and Google AI Overviews. Published monthly.
Monthly. No spam. Unsubscribe anytime.
How do you measure Share of Model?
Four steps. No proprietary tool required to start:
- Step 1. Design your prompt set. 50 to 100 queries mapped to your buyer’s real questions.
- Step 2. Choose your platforms. Match to where your buyers actually research.
- Step 3. Score each response. Record presence, position, characterization, and competitors.
- Step 4. Run on a fixed cadence. Monthly minimum, more often after structural changes.
I’ve run baseline SoM audits manually with a spreadsheet and 20 prompts. The insight from a clean manual baseline outweighs anything you’ll get from an automated tool running the wrong prompts.
Step 1. Design your prompt set
Your prompt set needs to reflect how buyers actually research, not how you want to be found. Three prompt types:
Category-level prompts. Generic research queries buyers run at the beginning. “What are the best [category] tools for [use case]?” “Who are the leading [category] vendors for [company size]?” These test your baseline category presence.
Problem-level prompts. Mapped to specific pain points your product solves. “How do I [solve problem your product addresses]?” “What’s the best way to handle [operational challenge]?” These test your association with the problems you own.
Evaluation-level prompts. Comparison queries buyers run when they’re close to a decision. “Compare [your company] vs [competitor A] vs [competitor B].” “Is [your company] a good fit for [specific industry or use case]?” These test recommendation share, not just mention share.
Start with 15 to 20 prompts across all three types. Run them manually. Use that baseline to build toward a larger automated set.
Step 2. Choose your platforms
Match platform selection to your buyer’s actual behavior. A 2026 analysis of 2 million LLM sessions found that B2B and enterprise buyers favor Microsoft Copilot (embedded in Excel, Word, and Teams), while finance professionals use Perplexity AI at 24% market share in that vertical. Technical evaluators prefer Anthropic’s Claude for deep analysis.
The platforms you measure are the platforms where your buyers actually research. Not the platforms with the largest general user base.
- OpenAI’s ChatGPT, Perplexity AI, and Google Gemini cover the majority of research behavior for most B2B companies.
- Microsoft Copilot is worth adding for enterprise Microsoft environments.
- Anthropic’s Claude matters more for technical and developer audiences.
Step 3. Score each response
For every prompt response, record:
| Dimension | What to record |
|---|---|
| Brand present? | Yes / No |
| Position | First named / middle of list / last / not named |
| Mention type | Recommended / listed / mentioned in passing / compared negatively |
| Characterization | Accurate / partially accurate / inaccurate / hallucinated |
| Competitor mentions | Which competitors appeared, in what position |
| Citation source | Was your content cited? Which URL? |
Step 4. Run on a fixed cadence
Monthly is the minimum. Parametric models update through retraining cycles. Retrieval-augmented models like Perplexity AI update in near real-time. A single snapshot tells you nothing about trajectory. Monthly tracking shows whether your fixes are working and whether model updates are affecting your position.
Correct entity signals today, and parametric models like OpenAI’s standard ChatGPT may not reflect those corrections for weeks or months. Retrieval-augmented models will reflect changes faster. Your tracking cadence needs to account for this lag. Don’t declare a fix successful after one measurement cycle.
One caveat the metric hides: the scoreboard has a bias. On July 1, 2026, 5W Public Relations tested 32,200 prompts and found most assistants over-recommend their parent company’s products. OpenAI’s ChatGPT self-recommended 2.0 times more often. Anthropic’s Claude, 1.2 times, the lowest. When you read your Share of Model, treat each platform’s self-preference as a variable, not a constant. Source: BigGo Finance, July 2026.
| Platform type | Typical update lag | Implication |
|---|---|---|
| Parametric (no live retrieval) | Weeks to months (retraining cycles) | Fixes take time to appear. Don’t expect immediate results. |
| Retrieval-augmented (live web search) | Hours to days | Content and citation changes can appear quickly. |
A live example of the latency problem cutting the other way: when ChatGPT moved to GPT-5.5 on May 23, 2026, SISTRIX measured a 47% citation shift across 3.8 million responses within 48 hours. Parametric updates do not always take months to surface. When the model itself changes, your Share of Model can move overnight. Monthly is the floor. Run an extra cycle after any known model release. (Source: SISTRIX, June 2026.)
How do you measure SoM across the buyer journey?
Your SoM at the discovery stage will look different from your SoM at the evaluation stage. Both need separate tracking. A company can have strong category SoM and weak evaluation SoM. Buyers hear their name early. They’re not being recommended at the decision point.
This is one of the most common structural gaps in our audits: a company that appears in generic category prompts but disappears in comparison and evaluation prompts. They have mention share at the top of the funnel. They have near-zero recommendation share at the bottom. That’s a pipeline problem disguised as a visibility win.
Discovery SoM
Prompts: “What are the best [category] tools?” “Who are the leading [category] vendors for [company size or industry]?”
What this tests: Whether AI systems include you in the initial category framing. A buyer at this stage doesn’t know your name. They’re asking AI for a starting list. If you’re not in the answer, you’re not in their consideration set. You never get a second chance at discovery.
Evaluation SoM
Prompts: “Compare [your company] vs [Competitor A].” “Is [your company] good for [specific use case]?” “What are the pros and cons of [your company]?”
What this tests: Whether AI systems characterize you accurately and favorably when a buyer is close to a decision. This is where recommendation share separates from mention share. A company can appear in a comparison response and still lose. If the model characterizes them as the lower-value option, appearing hurts more than it helps.
Troubleshooting SoM
Prompts: “How do I fix [problem your product solves]?” “What’s the best way to handle [operational challenge]?”
What this tests: Whether AI systems associate you with expertise in the problem, not just with the product. This is the most durable form of SoM. Companies that appear in troubleshooting responses are embedded in the knowledge layer, not just the vendor layer. That’s harder to displace.
| Journey stage | Prompt type | What it measures | Primary risk if low |
|---|---|---|---|
| Discovery | Category / vendor lists | Initial inclusion | Never entered consideration |
| Evaluation | Comparisons / fit assessment | Recommendation quality | Named but not chosen |
| Troubleshooting | Problem / solution queries | Expertise association | Visible as vendor, not authority |
What does a competitive SoM benchmark look like?
No published industry benchmark for Share of Model exists. Research confirmed this gap in March 2026: available frameworks describe methodology, but no third party has established a median score or target range for B2B companies.
That gap is itself a signal. The market hasn’t standardized SoM measurement yet. Companies that establish their baseline now will have 12 to 18 months of trajectory data before their competitors start tracking.
What follows is the first structured B2B SoM benchmark framework, built from audit data across 50+ companies. Not vendor claims. Not theoretical ranges.
How to set your baseline and track competitors
Run your initial prompt set. 20 to 30 prompts minimum. Across OpenAI’s ChatGPT, Perplexity AI, and Google Gemini. Record every company named in every response. Calculate your mention rate: your appearances divided by total prompts. Do the same for your top three competitors.
That is your baseline. Run the same prompt set in 30 days. Trajectory matters more than the absolute number.
Tier progression maps directly to the Stack:
- Moving from Tier 1 to Tier 2 requires fixing structural failures in Layers 1 through 3.
- Tier 2 to Tier 3 requires Layer 4 (training-ready content) and Layer 5 (Algorithmic Touchpoint Presence).
- Breaking into Tier 4 requires all seven layers working together.
That’s what the Algorithmic Authority Audit is designed to diagnose and fix.
FAQ
Is Share of Model the same as Share of Voice in AI?
No. Share of Voice measures brand presence in human-facing channels: paid media, social, traditional search. Share of Model measures presence inside AI-generated responses. A different surface that runs on different signals. High Share of Voice does not predict high Share of Model. They must be measured and optimized separately.
Can I use my existing SOV tools to measure Share of Model?
No. Existing SOV tools are built for traditional media monitoring and search visibility. They don’t read AI-generated responses. Some emerging AI visibility platforms now offer prompt-based monitoring across ChatGPT, Perplexity AI, and Gemini. For most B2B teams starting out, a manual prompt set is sufficient to establish a baseline before investing in automated tooling.
How often should I run my SoM prompt set?
Monthly at minimum. Parametric models update on retraining cycles measured in weeks to months. Retrieval-augmented models like Perplexity AI reflect content changes in days. Running monthly captures trajectory without generating noise from normal model volatility. After a significant structural fix (new entity signals, major content changes, a PR push), run an additional cycle two to four weeks after the change.
What’s a realistic SoM target for a B2B company in a competitive category?
In a competitive B2B category, 0 to 5% is typical before structural visibility work. 15 to 25% is attainable within 6 to 12 months with fixes across Layers 1 to 4 of the Algorithmic Authority Stack. In narrower, less competitive categories, 30%+ is achievable within 12 to 18 months. No third-party benchmark exists yet. The framework in this piece is the first structured approach built from audit data.
Does my SoM on ChatGPT predict my SoM on Perplexity?
No. A 2025 analysis of 680 million citations found that only 11% of domains are cited by both OpenAI’s ChatGPT and Perplexity AI. The platforms use fundamentally different retrieval architectures. A company can have strong SoM on one and near-zero on the other. Measure both.
What’s the difference between mention share and recommendation share?
Mention share is how often your company’s name appears in an AI response. Recommendation share is how often the AI actively recommends your company as the preferred option for the buyer’s specific use case. A company can have high mention share and low recommendation share: appearing in lists without being endorsed. Recommendation share correlates to buyer consideration. It’s harder to move and requires deeper structural authority work.
Which layer of the Algorithmic Authority Stack drives Share of Model?
Share of Model is the output of all seven layers, measured at Layer 7 (Visibility Measurement). Layer 1 (Market Identity Clarity) determines whether AI can classify you. Layer 3 (Semantic Density) determines whether your language is legible. Layer 4 (Training-Ready Content) determines whether your content can be extracted. Layer 5 (Algorithmic Touchpoint Presence) determines whether you appear across multiple AI platforms, not just one. You can’t optimize for SoM directly. Fix the layers. The metric follows.
Related Reading
- Guide 8: Visibility Measurement. The complete measurement system SoM fits inside.
- Google Rankings ≠ AI Citations. Why the metric that used to predict visibility no longer does.
- Layer 5: Algorithmic Touchpoint Presence. Why single-platform SoM isn’t AI visibility.
- How to Get Cited by Perplexity AI. The platform most responsive to content changes.
- What Is the Difference Between SEO, AEO, and GEO?. Where SoM fits in the full visibility stack.
Is your company invisible to AI?
Six questions, about 90 seconds. Find out which of the seven layers is breaking first, and whether you are failing to be retrieved or failing to be cited.
Take the AI Visibility Test