Competitor Share of Model Evaluation Template: A Practical SoM Framework

By maxaeo.ai ｜ Published 2026-10-13 ｜ Updated 2026-10-13

A competitor share of model evaluation template turns AI search answers into a repeatable comparison of brands, prompts, models, citations, and recommendation position. Instead of asking whether a company “shows up,” it measures how much competitive answer space each brand captures and where that visibility comes from.

Most AI visibility frameworks cover share of voice, prompt gaps, model-level reporting, and cited sources. The harder measurement problems are usually left open: keeping the denominator stable, weighting recommendation position, separating buyer intent, and distinguishing a passing mention from a commercially useful recommendation. (aisearchtoolrank.com)

This framework fills those gaps with a copy-ready evaluation structure for SaaS and other considered purchases.

What is share of model in competitive AI search?

Share of model (SoM) is the percentage of weighted AI answer visibility captured by a brand across a fixed set of prompts, models, and competitors. It is similar to traditional share of voice, but it accounts for how AI systems mention, compare, cite, and recommend brands.

A basic formula is:

Brand SoM = Brand weighted visibility
            ÷ Total weighted visibility for all tracked brands
            × 100

A simple mention count is useful for directional tracking, but it treats every appearance equally. A brand listed first in a recommendation answer should not receive the same score as a brand mentioned once in a long comparison.

The evaluation should therefore record four separate signals:

• Presence: Was the brand mentioned?
• Position: Where did it appear in the answer?
• Recommendation strength: Was it described as a fit, alternative, leader, or poor choice?
• Evidence: Did the answer cite the brand’s website or third-party sources?

This distinction matters because a competitor can have a high mention rate but weak recommendation framing, while another brand may appear less often but win the highest-intent prompts.

The core competitor share of model evaluation template

Use one row for every prompt × model × run combination. Do not combine all platforms into one number before storing the underlying observations.

Keep the raw answer as a required field. A percentage without the underlying answer cannot explain what changed or whether the model made a factual mistake.

How to calculate a defensible SoM score

A practical scoring model should reward useful visibility without pretending that AI answers behave like conventional rankings.

Use the following illustrative weights:

Then apply a position multiplier:

For each observation:

Weighted visibility =
Base points × Position multiplier × Intent weight

Recommended intent weights:

• Category discovery: 0.8
• Problem-aware research: 1.0
• Vendor comparison: 1.2
• Alternative or switching prompt: 1.3
• Purchase-ready recommendation: 1.5

These numbers are not universal industry standards. They are a transparent starting point. The value comes from applying the same rules to every competitor and preserving the raw evidence for review.

Example calculation

Suppose a SaaS brand is recommended second on a vendor-comparison prompt and cited in the answer:

Base points = 2.0
Position multiplier = 0.80
Intent weight = 1.20

Weighted visibility = 2.0 × 0.80 × 1.20 = 1.92

If the four tracked competitors collectively earn 12.0 weighted points on that prompt, the brand’s prompt-level SoM is:

1.92 ÷ 12.0 × 100 = 16%

Report this alongside raw mention rate. A brand with 16% SoM and a 40% mention rate has a different competitive position from a brand with 16% SoM and a 10% mention rate.

The model-level comparison matrix

A single blended score can hide major differences between AI engines. Build a model matrix before calculating the portfolio-level number.

Calculate each model separately using the same prompt set and scoring rules. Only then create a portfolio score, using a documented weighting scheme such as equal model weighting or audience-based weighting.

This prevents a strong result on one platform from masking poor visibility on another. AI share-of-voice tools commonly recommend model-level breakdowns, while more advanced competitive studies add fixed prompt sets, discovered competitors, and confidence ranges. (digitalotters.com)

For a deeper methodology, see the AI engine position tracking framework for probabilistic answers.

Prompt design: the part most templates miss

The prompt set determines what your SoM means. A list of branded queries will measure reputation, not category competitiveness.

Build the evaluation set across five intent groups:

1. Category prompts: “What are the best tools for…?”
2. Use-case prompts: “Which platform helps a SaaS team…?”
3. Comparison prompts: “A versus B for…”
4. Alternative prompts: “What are alternatives to…?”
5. Risk prompts: “What should buyers watch out for when choosing…?”

For each group, create variations by company size, use case, geography, and buying constraint. Keep the core set locked for trend reporting, but maintain a separate discovery set for newly appearing competitors.

This two-set approach solves a common denominator problem: discovered brands can be monitored without silently changing the historical benchmark. The same principle is useful in traditional competitor analysis, where adding or removing domains can distort share calculations. (aisearchtoolrank.com)

Add the evidence layer: citations, sources, and gaps

SoM tells you who appears. The evidence layer helps explain why.

Add these fields to your reporting view:

A useful original diagnostic is the Evidence-Adjusted SoM Gap:

Evidence-Adjusted SoM Gap =
Competitor SoM − Your SoM
+ Citation gap
+ High-intent prompt gap

Use the result as a prioritization signal, not as a financial forecast. A competitor with a modest SoM lead but a large exclusive-source advantage may require a source-building response. A competitor with a large SoM lead but weak citation support may be vulnerable to accurate, well-structured content.

MaxAEO’s LLM competitive evidence extraction workflow is useful when the next action depends on identifying the specific pages and domains behind an AI answer.

How to use the template in a monthly review

A repeatable review can follow five steps:

1. Freeze the benchmark: Keep the core prompts, competitor set, models, locale, and scoring rules unchanged.
2. Run repeated observations: Record the same prompts on the same engines at a consistent cadence.
3. Calculate three views: Mention rate, weighted SoM, and citation rate.
4. Inspect material changes: Review raw answers whenever SoM moves materially or a competitor enters the top recommendation position.
5. Assign an action: Improve a missing comparison page, correct an inaccurate claim, strengthen a cited source, or monitor without intervention.

For operational teams, a daily monitoring workflow can store answers and trend lines while keeping the monthly benchmark stable. MaxAEO tracks brand mentions, competitive position, recommendation placement, sentiment, and cited sources across eight AI engines, with daily updates and bilingual market coverage. A daily AI search tracking workflow provides a practical operating model.

Common questions

Is SoM the same as AI share of voice?

Not exactly. AI share of voice often counts mentions or citations, while SoM can add recommendation strength, position, intent, and evidence quality. Teams should define the formula before comparing periods.

How many prompts should a benchmark include?

There is no universal number. Start with a balanced core set covering category, use-case, comparison, alternative, and purchase-ready prompts. Expand only when new buyer intents or competitors are discovered.

Should every AI model have the same weight?

Not necessarily. Equal weighting is easiest to audit. Audience-based weighting may be more useful when your buyers clearly favor particular AI platforms, but the rationale must remain documented.

Can a high mention rate still indicate weak visibility?

Yes. A brand may be mentioned frequently but appear late, receive neutral treatment, or lack citation support. Pair mention rate with weighted position, sentiment, recommendation type, and citation rate.

What should happen after finding a prompt gap?

Inspect the winning competitor’s cited sources and answer framing first. Then decide whether the gap requires a comparison page, clearer product documentation, third-party coverage, or a factual correction. MaxAEO’s generative search competitor comparison matrix can support that prioritization.

A useful SoM report is not a leaderboard alone. It is a controlled measurement system that connects prompts, models, recommendations, citations, and concrete content decisions.