AI visibility monitoring: how to measure whether AI assistants recommend your business

HUB LLC · 12 September 2026 · Search & discoverability

Rank tracking tells you where a page appears for a keyword. Analytics tells you where visits came from. Neither answers a question more companies are now asking: when a potential customer asks an AI assistant to research a market, is our business part of the answer, and how do we compare with competitors?

That can be measured, within limits. This article explains how to design an AI visibility measurement that is repeatable, transparent and compliant, and how to turn the results into work that improves the underlying information.

What “AI visibility” should mean

“Visibility” is too vague to act on. Break it into observations that can be recorded for every answer:

• Mention: the brand, store or product is named.
• Recommendation: it is included in a shortlist or suggested as a fit for the request.
• Citation: where the system shows sources, the company’s own domain is among them.
• Association: the system correctly understands what the company sells and where it operates.
• Accuracy: the facts it states about services, locations, prices or products are correct.

A business can be mentioned often and still be described wrongly, or cited as a source without being recommended. Each observation leads to a different kind of fix.

A measurement is a sample, not a ranking

AI answers vary with the model version, product mode, location, language, personalization, conversation history and time. A visibility check is therefore a sampled measurement under recorded conditions, not a universal position. Every result should store:

• the date, time and exact prompt;
• the market and language;
• the provider, the interface tested and any available model or mode information;
• the full answer and any visible sources or citations;
• the detected mentions, and whether a person confirmed them.

Run each prompt repeatedly over time and report ranges and trends rather than single answers. A change of one mention in ten runs is noise until it persists.

The prompt set decides the quality of the result

Testing only the company name measures awareness. Commercial questions measure acquisition opportunity. Group prompts by intent so results can be read by stage of the buying process:

Build the list from real evidence: questions from sales calls, onsite search terms, Search Console queries, product categories and the positioning of named competitors. Review it quarterly, but keep a stable core so trends remain comparable.

Scores must show their working

A single proprietary number is easy to sell and hard to trust. If a report includes a score, each component should be visible together with the evidence, test date and prompt behind it. Useful components are:

• Prompt coverage: the share of monitored prompts where the company appears.
• Recommendation presence: the share where it is included in a shortlist or comparison.
• Citation presence: the share of source-showing answers that cite the company’s domain.
• Competitive share of voice: the company’s mentions relative to tracked competitors.
• Information accuracy: the share of mentions with no factual error.
• Content support: whether the website has a strong page answering each monitored intent.
• Technical discoverability: crawling, indexing and structured data barriers.

Competitor share of voice

For each prompt, record whether the company and each tracked competitor appear, then aggregate across the prompt set and repeated runs. An illustrative example for one topic:

The aggregate is less useful than the topic breakdown. If a competitor appears repeatedly for Magento migration prompts and you do not, look for reasons: a dedicated landing page, clearer case studies, more third-party coverage, better structured information or more specific service descriptions. That turns a dashboard into competitive research rather than a vanity metric.

Follow the sources

Where an AI or search experience shows sources, record which domains support each answer. Answers about a market may lean on review sites, trade publications, marketplaces, documentation or forums rather than on any company’s own website.

Useful views include the most frequently cited domains, the sources behind competitor recommendations, the company’s own pages that are cited, relevant third-party sources where the company is absent, and outdated or incorrect sources. Legitimate responses are to improve first-party pages, publish better technical documentation or original research, correct business listings and earn relevant industry coverage. Fabricated reviews and manufactured citations are not a strategy; they are a liability.

Check accuracy, not only presence

Flag answers that describe the company with an outdated service, the wrong markets, old pricing, a previous brand name or a missing core capability, or that confuse it with a similarly named business. For each confirmed error, identify which first-party page, structured data or business profile should state the correct information more clearly. For companies with complex offerings, this is information-quality monitoring as much as marketing measurement.

Measure within the rules

How results are collected matters as much as what is collected. Use official APIs, approved integrations and each provider’s documented terms. Do not build monitoring around scraping consumer interfaces or bypassing access controls. Google, for example, classes automated queries and scraping results for rank-checking without express permission as violations of its spam policies and Terms of Service. Source: Google Search spam policies.

Two practical consequences follow. First, an answer obtained through a developer API may differ from what a signed-in consumer sees in an app, so a report should state which interface was tested. Second, some signals come first-hand: Google includes traffic from AI Overviews and AI Mode in the Search Console Performance report, within the overall Web search type rather than as a separate figure. Source: Google Search Central. Referral data from AI assistants in your analytics adds another observed signal.

Avoid false positives

Simple text matching fails for short or ambiguous names. For each brand, maintain a canonical name, alternative spellings, domains, product brand names, meanings to exclude and country context. Low-confidence matches should go to a person for review instead of being counted automatically.

Keep observations and hypotheses apart:

Observed: Competitor A appeared in 8 of 10 runs of the migration prompts; your company appeared in 3.
Hypothesis: Competitor A’s dedicated migration guide may contribute to a stronger association with the topic.

A monitoring report should never present correlation as proven cause.

Turn results into work

A dashboard that only reports that visibility went down will soon be ignored. Every finding should point to an action in one of five areas:

• Technical: indexing, rendering, structured data, internal linking, sitemaps, canonicals or performance.
• Content: missing service pages, weak category information, no comparison content, outdated FAQs or pages that do not answer the monitored intent.
• Product data: missing attributes, poor identifiers, inconsistent price or availability and incomplete feeds.
• Authority: case studies, original research, technical documentation and credible third-party references.
• Business information: unclear company identity, markets served, contact details or policies.

Audit, implement, monitor, improve

Visibility monitoring is the measurement step in a loop. A readiness audit asks whether a store is technically and structurally prepared. AI Commerce Ready implementation fixes what the audit finds. Monitoring then shows whether the brand and its products are surfaced more often and more accurately for the questions that matter, and it informs the next round of improvements.

AI visibility does not replace SEO. Rankings, organic traffic and clicks still matter; AI visibility adds another view of how customers discover suppliers. A page can rank well yet rarely inform AI answers, or be cited often without sending much traffic, and both signals are worth having. Our zero-click SEO guide covers the search side.

If you want to know how visible your company or store is for the questions your customers ask, tell us your market, your competitors and the questions that matter. We will propose a prompt set and a measurement plan before any monitoring begins.