Share of Model Benchmark Report

Share of Model: Compare AI Visibility With Clear Definitions

Define mentions, source links and recommendations before comparing brands. Build a useful scorecard with consistent questions, denominators and collection conditions.

v1.1 · October 2026 · Tenten GEO Research

CH.01

Measure a Defined Part of Visibility

AI answers add another place to observe how a brand appears. Sampled visibility can inform decisions alongside traffic, customer research and business outcomes.

A flowing cloud of violet particles crosses a glowing cream horizontal bar and thin parallel lines on a dark background — Conceptual artwork for measurement; the bar is not observed brand performance.

Buyers may encounter your brand in search, reviews, conversations and AI answers. An observation from one product or prompt set does not reveal every buyer’s shortlist. Keep the question clear: what appeared in the responses we actually collected?

Share of Model is used here as a reporting label, not a standardized market-share measure. Define each metric separately and compare brands within the same observations. Your own history can also be useful; a competitor column does not automatically make a sample representative.

View	What it can describe	Boundary to keep
Media and community presence	Coverage, discussion and responses in selected channels	Coverage and exposure require their own definitions
Sampled AI answers	Names, source links and recommendations in collected responses	A prompt sample is not all real-user demand
Website behavior	Observed visits and actions, subject to collection limits	A link appearing does not prove a click
Commercial outcomes	Qualified inquiries, opportunities and recorded revenue	Attribution and causality need additional evidence

Complementary views of performance; no single visibility metric replaces the others

Start with the decision

Choose a question the report should help answer, such as whether a product description is accurate or which reader need deserves attention. Then collect the evidence that fits it.

CH.02

Read Benchmarks With Their Definitions

A published average is useful only when its population and measurement match the comparison. It is not a passing grade for every brand.

The AthenaHQ State of AI Search 2026 PDF downloaded on 5 October shows 16.3% average brand mentions on page 5. Page 18 defines brand mentions using responses where at least one brand appears, while share of voice uses all brand mentions as its denominator. The earlier 17.2% value still appears in indexed summaries. Record the source version and definition rather than treating either number as your expected result.

→Check which products, question types, languages and dates the sample covers before comparing it with yours.
→A conditional rate that excludes answers without brands differs from a rate across all eligible answers.
→Do not assume a low rate proves poor content, or that one language has more opportunity, without relevant evidence.

Use a comparable baseline

Keep your own counts, conditions and prior period visible. If an external benchmark lacks enough detail for a fair comparison, describe it as context and leave the comparison unscored.

CH.03

Build the Scorecard From the Same Observations

Brands can be compared only when the collection and counting rules align. Define the rules before interpreting changes.

Extract each brand’s presence from the same response set where possible. Separate discovery prompts from prompts that name a brand, since naming it in the question changes the context. A larger prompt set does not automatically represent the market.

Step 1

Define the questions and scope

Choose relevant reader tasks, products, modes, languages and locations. Preserve a versioned question set; no fixed 100–300 prompt count is required.

Step 2

Define the brands and matching rules

Choose relevant brands and record names, aliases and ambiguity rules. Keep branded and unbranded questions separate and preserve the comparison roster.

Step 3

Collect and document responses

Record time, mode, interface or API, search setting and failures. Use a planned repeat policy; do not keep only favorable answers.

Step 4

Calculate with explicit denominators

Report counts and rates together. A missing response is not a zero mention; no eligible observations means unavailable, not 0%.

Metric	Definition for this framework	Comparison rule
Brand mention rate	Eligible responses naming the brand at least once / all eligible collected responses	Count each brand once per response; report exclusions and counts
Difference between brands	The percentage-point difference between comparable mention rates	If also reporting a ratio, a zero comparison rate makes it undefined; show counts instead
Explicit recommendation rate	Eligible recommendation-task responses explicitly recommending the brand / eligible responses for that task	Define what counts; first position or mere mention is not automatically a recommendation
Own-domain source-link rate	Eligible responses showing a source link to the approved brand domain / eligible collected responses	Record domain rules and source role; appearing as a link is not a visit

Explicit definitions for this guide; vendor metrics with the same name may use different denominators

Keep the comparison honest

Multiple brands can appear in one answer, so their mention rates may sum to more than 100%. A normalized share of mentions is a different calculation. Label a changed question set, model or brand roster as a changed series.

CH.04

Separate Names, Sources and Clicks

A brand name, a source citation and a clickable URL describe different observations. None establishes a purchase on its own.

A mention can be favorable, neutral or unfavorable. A displayed source link does not prove the linked page supports every statement in the answer. Record the context and check support for important claims rather than assigning fixed business weights to each signal.

Link behavior depends on the product, mode, question and collection method. A rate of answers containing any external link is not the chance that your brand will be cited. Keep each product’s observations separate before creating any explained aggregate.

Signal	What is observable	What needs separate evidence
Mention	The brand is named in the recorded response	Accuracy, sentiment, recommendation and real-user exposure
Source citation	The response attributes information to a source	Whether the source supports the associated statement
Clickable link	A URL is presented in the observed interface	Actual clicks, qualified visits and downstream outcomes

Related signals with different meanings; no universal value ranking

Read the response, not only the score

Keep representative examples and unresolved classifications with the report. Use traffic and business records when available, while keeping visibility, referral and attribution limits clear.

CH.05

Choose Tools by the Measurement You Need

Confirm the collection method, coverage and usable output before comparing total cost. A plan name or price does not establish accuracy.

Check current official plans and distinguish database research from custom prompt tracking. On 5 October 2026, Otterly lists Lite at US$29 monthly, while Profound’s current site describes trial access and custom Enterprise pricing. This is a dated example, not a complete price comparison or a recommendation for one vendor.

Need	Questions to ask	Evidence to request	Cost to include
A focused baseline	Does the method cover the products and questions relevant to the decision?	A sample export, collection conditions and metric definitions	Plan limits, setup and analyst time
Repeatable comparisons	Can question sets, brand rules and historical method changes be tracked?	Response history, failure handling and versioned configuration	Collection frequency, add-ons and ongoing review
Multiple teams or markets	Are access, export and local-language requirements supported?	Relevant documentation and a permitted representative demonstration	Total scope, staff, support and contract terms

A procurement framework, not a vendor ranking or a claim that expensive plans cover every product

→Identify whether observations come from consumer interfaces, APIs or a vendor’s existing index.
→Inspect raw responses and counting rules where permitted, including missing data and ambiguous brand matches.
→Choose only features that support the decision and fit the available budget and staff capacity.
→Agree on evaluation scope and cancellation terms; a longer contract is a commercial choice, not proof of tool quality.

Use the results you can explain

Select a tool whose method and output fit your work. If the team cannot interpret the metric or act on the result, clarify the need before increasing spend.

CH.06

Report the Decision on One Page

A concise report should show the scope, the observation, the uncertainty and the next action. It can include a reason to maintain or stop work.

Use counts and rates together, and keep methods consistent across periods. A percentage increase from a small base or a changed prompt set can look dramatic while saying little about wider exposure.

Block 1

Scope and headline observation

Name the products, modes, question set, period and eligible response count. Show the chosen metric with its denominator and comparison basis.

Block 2

Comparable change

Show the prior period and current counts under the same method. Mark changes in configuration, missing data and uncertainty.

Block 3

Examples and reader needs

Include useful response examples and specific factual or information gaps. Missing from a list alone does not mandate a rewrite.

Block 4

Decision and resources

State what to continue, change or stop, why, and what it would cost. Separate proposed business benefits from outcomes already measured.

Report section	Include	Purpose
Scope and observation	Definition, dates, products, counts and denominator	Understand exactly what was measured
Comparison	Comparable history or brands, with method changes marked	Assess change without overstating precision
Examples and limits	Relevant responses, factual gaps and unknowns	Explain what the number does and does not support
Next decision	Action or no change, owner, cost and review trigger	Use the report to make a practical choice

Four parts of an actionable report; cadence depends on the decision, not a reporting quota

Keep the meaning visible

A useful scorecard explains what was observed and which decision it supports. It does not turn a sampled mention rate into market share, buyer preference or guaranteed revenue.

Sources & methodology

Revised on 5 October 2026. AthenaHQ’s current State of AI Search 2026 PDF, pages 5, 17 and 18, reports 16.3% average brand mentions, describes millions of responses across seven models and distinguishes conditional brand mentions from share of voice. Its source-monitoring section names December 2025–March 2026; that statement is not assumed to establish the exact collection window or sample size for every benchmark. Source: https://athenahq.ai/reports/Athena-State-of-AI-Search-Report-2026.pdf. Indexed summaries still show the earlier 17.2%; the two versions are not evidence of a market decline. Current product examples: https://otterly.ai/pricing; https://www.tryprofound.com/pricing. The earlier 100% expectation comparison, unverified Spotlight 77%/31% chart and unmatched price multiples are removed. The scorecard definitions and review framework are Tenten’s recommendations, not standardized market shares or audited conversion results.

Discuss your AI search priorities

Request a 30-minute benchmark interpretation

Send a booking request. We’ll email you to confirm a time within 24 hours.