Share of Model is useful only when the sample behind it is visible. For a defined set of prompts, engines, markets, and dates, it answers one narrow question: in what percentage of eligible answers did the brand appear? It does not estimate every answer on the internet, prove preference, or guarantee demand.

What Share of Model can—and cannot—tell you

A transparent calculation is: eligible sampled answers that mention the brand, divided by all eligible sampled answers, multiplied by 100. Define an eligible answer before collection—for example, a completed answer for an approved category question in the selected country and language. Keep the prompt set and inclusion rule fixed inside each reporting period.

Model outputs vary by product, version, location, personalization, and time. That makes the result a directional sample, not a population estimate or cross-vendor benchmark. A change can signal that the sampled answer set changed; it does not, by itself, explain why or prove an SEO, revenue, or reputation outcome.

Keep three measurement layers separate

1. Controlled AI-answer sample

This is where Share of Model lives. Use a predeclared question set, record the test conditions, code each eligible answer with the same mention rule, and calculate the share. Report the sample size alongside the percentage. Use the result to compare like-for-like samples over time, not to claim universal market share.

2. Google generative AI visibility

Google's Search Generative AI performance reports provide a different first-party layer. Google says these insights have been available to websites worldwide since 31 August 2026 and show impressions, pages, countries, devices for Search, and dates for visibility in generative AI features. The data also remains included in overall performance reporting. These platform impressions are not the same denominator as a controlled prompt sample, so do not merge the two percentages.

3. Referral and on-site outcomes

Referral analytics answer whether a cited link produced a visit and what happened on the site. OpenAI states that ChatGPT search referral URLs include utm_source=chatgpt.com. Measure this layer with aggregate, privacy-safe counts by approved source and page path. Do not persist raw prompts, generated answers, calculator results, full URLs, query strings, referrers, or user identifiers in the content workflow. A lead or CRM outcome remains unavailable until the receiving system confirms it.

What improves eligibility for AI search

For Google's AI features, the current guidance is refreshingly ordinary: apply foundational SEO. A page must be indexed and eligible to appear with a snippet. Crawling must be allowed, important content should be available as text, internal links should make the page discoverable, and structured data should match what visitors can see. Meeting those conditions does not guarantee indexing, citation, or visibility.

Google also says there is no special schema or AI text file required for its generative search features. It advises against rewriting content for machines, manufacturing mentions, or producing a separate thin page for every query variation. The durable advantage is non-commodity evidence: original experience, clear methods, useful comparisons, cited claims, and a technically accessible page.

A repeatable Share of Model protocol

Define the decision and question set

Start with a business decision, not a vanity score. A brand team might need to learn whether it appears in category education, consideration, comparison, or risk questions. Freeze a balanced set of questions for that job, document exclusions, and avoid changing the list midway through a comparison window.

Fix the environment

Record the model product, country, language, collection date, run count, and whether the session was fresh. If any condition changes, annotate the break rather than presenting the new number as a clean continuation. The goal is reproducibility, even though model outputs remain probabilistic.

Code mentions consistently

Choose the mention rule before reviewing results. Decide whether aliases count, whether a linked citation without a written name counts, and how to treat unavailable or refused answers. Apply the same rule to the brand and comparator set. Retain only the approved aggregate evidence required for review.

Calculate, annotate, and repeat

Calculate the sampled share, publish the sample size and conditions, and list any missing runs or methodology changes. Repeat on a cadence that matches the decision—monthly is often more interpretable than daily noise. Use the Share of Model calculator to structure a controlled sample and compute the percentage; treat its output as a diagnostic, not a market benchmark.

Where brand-owned AI influencers fit

A brand-owned synthetic influencer can create a consistent body of category education, product explanation, disclosure, and campaign evidence. That makes the asset useful to people and easier to connect across the brand's site. It does not automatically earn inclusion in an answer. The content still needs named ownership, truthful claims, visible evidence, strong internal routing, and transparent synthetic-media governance.

Turn the diagnostic into action

Interpret the layers together without collapsing them. If controlled samples show no mentions and the relevant pages are not indexed, fix crawlability and page quality first. If the brand appears but the supporting evidence is inaccurate, improve the source page and claims. If referrals arrive but qualified action does not, review the landing page and CTA. Keep outcome claims blocked until durable receipt and reconciliation exist.

Primary sources

Google Search Central: AI features and your website explains eligibility, technical foundations, and overall Search Console reporting.

Google Search Central: Optimizing for generative AI features covers current SEO guidance and unsupported shortcuts.

Google Search Central: Search Generative AI performance reports documents the dedicated report's rollout and available dimensions.

OpenAI Help Center: Publishers and Developers FAQ explains OAI-SearchBot access and ChatGPT referral attribution.