AI Visibility Measurement

Measuring AI visibility means counting how often your brand or pages appear in answers from AI search tools, which is reach a site can gain or lose while its traffic looks unchanged. Search Console now reports impressions from Google’s own AI features,¹ and since November 2026 OpenAI’s Business Hub counts the ChatGPT answers that reference a claimed business’s website, though not those that only name the brand.² Nothing first-party shows whether a brand is mentioned in a Perplexity answer that replaces a click entirely, or synthesised into a digest by a personal agent such as Gemini Spark, now task mode in the Gemini app, with no observable surface at all. Across Conductor’s May to September 2025 measurement window, AI referral traffic represented 1.08% of total website traffic across 10 industries analysed.³ That figure counts clicks arriving from AI surfaces, which is the opposite end of the problem: an answer that satisfies the question outright produces the reach and none of the referral. Measuring AI visibility requires a separate set of metrics and tools.

Why do traditional SEO metrics miss AI visibility?

Traditional search metrics track clicks and rank positions. AI search introduces a different outcome: a user reads an AI-generated answer that synthesises several sources, sees a brand name or a cited link, and may or may not click through to the site. The metric that matters is whether the brand appears in the answer, not whether it ranked in the list below it.

Ahrefs research found that 38% of pages cited in Google AI Overviews also ranked in the traditional top ten in early 2026, against 76% in a July 2025 study.⁴ Ahrefs improved its citation parsing between the two, so the pair is not like-for-like, and the explanations it offers for the gap are query fan-out and Google’s adoption of Gemini 3. The current figure carries the point on its own: a page’s rank for the query does not reliably predict whether it is cited, so measuring rankings alone misses citations. It does not show that citations ignore rankings, because AI answers draw on fan-out queries that rank trackers never check (see why AI citations don’t match Google’s rankings).

The consequences of stale content also differ between the two systems. In traditional search, outdated information reduces ranking quality but the user can judge what they read. In grounding, Bing’s search team writes, “a stale fact produces a misleading response”, and that response can cite your site.⁵

The core metrics

Citation rate measures the proportion of relevant queries in which your site or brand is cited in an AI-generated answer. A brand tracking 100 industry-relevant prompts and appearing in 40 of the resulting answers has a citation rate of 40% for that prompt set. The traffic quality case for tracking citation rate: in Knotch’s client-panel tracking, quoted in Conductor’s benchmarks report rather than measured by it, LLM-referred visitors converted at twice the rate of other traffic sources and in one-third the number of sessions.³

Share of voice measures your citations as a proportion of all citations across your competitive set. It is the AI counterpart of the share of voice metric used in organic search, and the two do not translate into each other. If five brands collectively appear 200 times across a prompt set and your brand accounts for 40 of those appearances, your AI share of voice is 20%.

Prompt coverage measures how many of the queries relevant to your category produce answers that cite your brand at all. A brand with high citation rate on a narrow prompt set but zero coverage across a broader set has a fragile position.

Sentiment records whether your brand is mentioned positively, neutrally, or negatively within AI-generated answers. Negative framing in a cited answer is a brand risk that citation rate alone does not surface.

These four metrics together give a more complete picture than any single number. Citation rate without share of voice misses competitive context. Share of voice without sentiment misses quality.

How do AI visibility tools work?

AI visibility platforms submit a defined set of prompts to AI search engines, through the engines’ APIs or by automating their live interfaces, record the full responses, and analyse how often and how prominently a brand or domain appears. Unlike traditional rank trackers, which check a URL’s position in a results list, AI visibility tools parse generated text for mentions, citations, and sentiment.

The accuracy of the output depends on the prompt set. Prompts should reflect the actual queries your audience uses when researching your category. A prompt set built from keyword research and customer interview data produces more actionable results than one built from broad industry terms.

Sampling frequency matters too. One run of a prompt set is a single snapshot, so weekly or monthly tracking is more useful for spotting trends.

One consequence of that method is now on the record, because Google has stated it: “Google does not evaluate third-party SEO tools or vendors directly, and they have no access to our internal metrics.”⁶ Every AI visibility figure these platforms report is their own observation of model output, not a number Google supplies or validates. Two tools can return different citation rates for the same brand in the same week and both be reporting honestly, because each ran its own prompts against its own sample. Treat the trend within one tool as the signal and the absolute number as an estimate.

Tools available in 2026

A category of dedicated AI visibility platforms has developed alongside the growth of AI search. The main options differ in which AI engines they cover, how many prompts they track, and how they surface citation and sentiment data.

Profound tracks up to nine answer engines, among them ChatGPT, Google AI Overviews and AI Mode, Perplexity, Gemini, Copilot and Claude, with prompt-level tracking.

Semrush AI Visibility Toolkit is an add-on to Semrush’s existing platform, covering brand mentions and citations across major AI surfaces, useful for teams already working within Semrush.

Otterly offers prompt-level citation tracking across ChatGPT, Perplexity, and other major AI surfaces with a tiered plan structure suited to smaller prompt sets at lower price points.

Peec AI covers ChatGPT, Perplexity, Google AI Overviews and AI Mode among other engines, with unlimited users on every plan, making it suited to agency use cases.

AthenaHQ tracks citations and brand mentions across ChatGPT, Perplexity, Google AI Overviews and AI Mode, Gemini, Claude, Copilot, Grok and others, with credit-based pricing that scales with prompt volume.

Pricing and engine coverage across this category change frequently as platforms mature; the descriptions above were checked against each provider’s site in October 2026. Check each provider’s current pricing directly before committing.

Two free options from Microsoft work differently enough to sit outside the comparison: both report citations Microsoft observed for a verified domain rather than running a prompt set. Bing Webmaster Tools’ AI Performance report, in public preview since February 2026, counts citations of your pages across “Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations” and lists the grounding queries that retrieved them;⁷ since June 2026 it also shows Citation Share, your proportion of all citations for a grounding query, an observed counterpart to AI share of voice.⁸ Since October 2026 a beta Topics Portfolio tab also lists queries in the topics you are cited for where other sites are cited and yours is not.⁹ Microsoft Clarity’s Citation dashboard reports the same kind of observed citation and grounding activity. Both surface the queries that actually retrieved your content, including ones you would not have thought to track. The trade-offs are shared: neither can measure a topic where you earn no citations, and coverage is narrow, since both name Microsoft Copilot and unnamed partners as their source, so neither can be read as coverage of ChatGPT, Gemini, Perplexity or Google AI Overviews. They are a useful observed baseline to sanity-check simulated citation rates against, not a replacement for a competitive prompt panel.

A third free source covers ChatGPT. OpenAI’s Business Hub, in beta since 10 November 2026, reports 28 days of mentions, website clicks and discovery topics to a business that has claimed its ChatGPT profile and passed an identity check (OpenAI launches Businesses in ChatGPT).² Like the Microsoft tools, it counts what happened in real answers instead of running a prompt set. For an online business a mention counts only when an answer references its website, so the figure measures linked visibility and will run below a tracker’s brand-mention count. Discovery topics group the themes of those answers without showing any prompt, and a metric stays blank until activity passes a reporting threshold, so a small site may see nothing at first.

Building a measurement framework

A functional AI visibility framework requires three things: a representative prompt set, a baseline measurement period, and a consistent reporting cadence.

Prompt set design: Start with 50 to 100 prompts that reflect how your target audience searches for your category. Include informational queries (“what is the best X for Y”), comparison queries (“X vs Y”), and transactional queries (“how to choose X”). Refresh the set quarterly as query patterns shift.

Baseline period: Run the same prompt set for four to six weeks before making content changes. Without a baseline, it is impossible to attribute shifts in citation rate to specific actions.

Reporting cadence: Monthly tracking suits most brands. Weekly tracking is worthwhile during periods of active content change or when monitoring the effect of a specific optimisation.

Integrating AI visibility with existing reporting

AI visibility measurement sits alongside existing SEO reporting, not in place of it. Google Search Console remains the source for traditional rankings, impressions and clicks, and GA4 captures traffic from AI referrers where the source is passed through.

GA4. Since May 2026 GA4 has an ‘AI Assistant’ default channel group, with Google’s documentation naming ChatGPT, Gemini, Claude, Deepseek, Copilot and Grok,¹⁰ and Perplexity confirmed as covered in a Google Analytics LinkedIn post in June 2026.¹¹ If a platform you care about still lands in Referral, add a custom channel definition for it.¹² Some AI traffic arrives as direct or unattributed, particularly from the ChatGPT and Perplexity mobile apps, and UTM parameters on links you control are the only clean fix.

Search Console. Clicks from within an AI Overview and clicks on the organic result below it both register as organic in the Performance report, so a falling CTR against stable impressions is a directional signal that an AI feature is absorbing clicks, not attribution. The generative AI performance reports, one for Search covering AI Overviews and AI Mode and a separate one for Discover, show impressions by page, country and date, plus device for Search, and have been available to all websites since 31 August 2026.¹ They carry no clicks and no queries, so they show where Google’s AI features surface your pages and not which searches trigger them. Two logging faults sit inside these series and need annotating: impressions were over-reported from 13 May 2025 to 27 April 2026,¹³ and Generative AI impressions were suppressed from 13 to 17 August 2026. The report can also list URLs that do not exist: links Google’s AI features invented, counted with impressions in the same table as real pages, which a check against your own sitemap separates out (Search Console reports hallucinated URLs as real pages).

The reporting pack. The simplest start is two rows added to the existing monthly template, AI citation rate and AI share of voice, each labelled with the tool that produced it and that tool held constant. The absolute number is one platform’s estimate from its own prompt sample, so only the trend within one tool is comparable month to month, and a figure that arrives without its source attached will be read as a measurement rather than an estimate.

Footnotes

1. Introducing Search Generative AI performance reports in Search Console — Google Search Central. Carries Google’s note: “As of August 31, 2026, we’ve rolled out these insights to all websites worldwide.” ↩ ↩²

2. Setting up your business on ChatGPT — OpenAI Help Center ↩ ↩²

3. 2026 AEO/GEO Benchmarks Report — Conductor, a live page (marked “last updated 6 July 2026” when read on 21 September 2026). Conductor’s own platform data: US averages across 1,215 of its enterprise customer domains in 10 industries, May to September 2025. ↩ ↩²

4. AI Overview Citations From Top-10 Pages Dropped From 76% to 38% — Ahrefs ↩

5. Evolving role of the index: From ranking pages to supporting answers — Bing ↩

6. Good SEO is good GEO — Brendon Kraham, Think with Google ↩

7. Introducing AI Performance in Bing Webmaster Tools Public Preview — Bing Webmaster Blog, 10 February 2026. ↩

8. New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare — Bing Search Blog ↩

9. Introducing Topics Portfolio (Beta) in AI Performance — Bing Webmaster Blog ↩

10. What’s new in Google Analytics: AI Assistant channel — Google Analytics ↩

11. Google Analytics AI Assistant traffic measurement roll-out — Google Analytics LinkedIn ↩

12. Custom channel groups — Google Analytics ↩

13. Data anomalies in Search Console — Google Search Central ↩