AI Search Measurement

How to Measure AI Search Visibility in 2026

Measure AI search visibility by testing a fixed set of customer prompts across ChatGPT, Perplexity, Gemini and Google, then recording brand mentions, linked citations, recommendation position, sentiment and competitors. Connect that visibility to AI referral traffic, branded search, enquiries and revenue. Rankings alone cannot show whether an answer engine names or recommends your business.

AI answers vary by platform, location, personalisation and time. The solution is not a single “visibility score”; it is a repeatable measurement system that separates exposure from commercial impact.

AEO · GEO · AI SEO · Updated September 2026

Which metrics matter?

Track whether the brand appears, whether the domain is cited, how prominently it is presented, whether the description is accurate, and whether that exposure influences traffic or conversions. Semrush similarly recommends separating AI mentions, citations and cited pages from conventional organic traffic reporting.1

Mention

The answer names the brand, product or expert. Record presence, position, sentiment and factual accuracy even when no link appears.

Citation

The answer links to the website or displays it as a source. Record the cited URL, platform, prompt and surrounding claim.

Lead

The user visits, searches the brand, calls, submits a form or buys. This is the business layer that visibility tools cannot prove on their own.

Creative Orbit framework — Mention → Citation → Lead: measure recognition first, attributable source selection second, and commercial action third. Reporting only the final click hides zero-click influence; reporting only mentions hides whether visibility produces business.

Build a prompt baseline

Create a stable library of real buying questions rather than testing only your business name. AirOps recommends beginning with 20–30 industry-relevant prompts, including branded, unbranded and competitor-comparison questions.2

Cover five intent groups

  • Discovery: “Who provides [service] in Wollongong?”
  • Problem: “How do I fix [commercial problem]?”
  • Comparison: “Compare [approach A] and [approach B] for an Illawarra business.”
  • Trust: “Which [provider type] has experience with [specific need]?”
  • Decision: “What should I ask before hiring a [provider]?”

Record the exact prompt, date, platform, location, account state and result. Keep the core set unchanged for trend reporting. Add experimental prompts separately so an expanding test library does not create a false improvement.

Create a practical scorecard

A useful scorecard preserves the answer evidence and calculates simple rates that can be checked manually.

Record these fields

  • Brand mentioned: yes or no.
  • Domain cited: yes or no, plus the exact destination URL.
  • Recommendation prominence: first, later, peripheral or absent.
  • Sentiment and accuracy: positive, neutral, negative or factually wrong.
  • Competitors mentioned and cited.
  • Source type: own site, directory, review platform, media, forum or video.

Calculate the mention rate as answers mentioning the brand divided by answers tested. Calculate the citation rate the same way using linked citations. Competitive citation share compares your citation count with the total citations earned by the tracked competitor set. These metrics show direction, not absolute market share, because generated answers can change between runs.

Track traffic and leads

Configure analytics to separate AI referrals, then connect sessions with enquiry, booking and revenue events. OpenAI states that ChatGPT uses the UTM parameter utm_source=chatgpt.com on referral URLs, enabling publishers to identify inbound traffic.3

Add three supporting signals

  • AI referral conversions: calls, forms, bookings or sales from identified AI sources.
  • Branded search movement: changes in Search Console impressions and clicks for the business name.
  • Assisted evidence: a short enquiry-form question asking how the prospect found the business.

AI influence is often under-attributed. A person may see the brand in an answer, search the name later and arrive through Google. Treat referral traffic as confirmed impact, branded demand as supporting evidence, and self-reported discovery as qualitative context.

Find the actual visibility gap

The most valuable result is not a score; it is the reason a competitor appears. Review which source types and pages the platform repeatedly selects.

  • If competitors are cited from their own guides, improve topical depth and direct answers.
  • If directories dominate, correct and strengthen listings on the recurring sources.
  • If local media or associations appear, develop genuine editorial and community authority.
  • If the brand is mentioned inaccurately, clarify services, locations and entity relationships.
  • If pages are never retrieved, audit crawler access, indexability, internal links and server responses.

Any public website can appear in ChatGPT search, but OpenAI says OAI-SearchBot must not be blocked if page content is to be discovered, surfaced and clearly cited.4 Google separately says pages must be indexed and eligible for snippets to appear as supporting links in AI Overviews or AI Mode.5

A structured AI search optimisation campaign should therefore report eligibility, mentions, citations and outcomes—not present a fluctuating generated answer as a guaranteed ranking.