AI visibility is the extent to which a brand, product or website is named, cited or recommended in answers produced by generative artificial intelligence (AI) systems such as ChatGPT, Gemini, Claude, Perplexity, Microsoft Copilot and Google's AI Overviews and AI Mode. It exists as a term because those systems answer in prose. A results page shows where a site ranks; a synthesised answer shows only who made it in, so marketers needed a word for being in it.

The label belongs to vendors and agencies rather than to any standards body. More than 20 vendors sell tools under the name, by the IAB's count, and each measures something slightly different. The closest thing to a shared definition is guidance the Interactive Advertising Bureau (IAB) published on August 3, 2026, which set out vocabulary and disclosure rules but no certification. It is the scoreboard for the practice known as GEO, or generative engine optimization, which is covered in its own explainer.

How it is measured

Almost every commercial tool works the same way. The vendor, or the client, assembles a prompt library: questions a buyer might ask, such as "best running shoes for flat feet" or "alternatives to Salesforce". Software sends each prompt to several platforms, often repeatedly, and parses the answers for brand names and cited links. The counts are then rolled up into rates and scores.

The IAB document, Measuring Visibility in the AI Era, groups the resulting metrics into four levels it calls the "4 P's":

  • Presence: mention rate, the share of responses that name the brand; citation rate, the share that cite one of its pages as a source; and share of voice, its mentions as a share of all brand mentions in a defined category.
  • Prominence: position, meaning where the brand appears in the answer and whether it comes first.
  • Portrayal: sentiment, framing, and separate hallucination and factual inaccuracy rates.
  • Persuasion: recommendation strength and the click-through rate after a citation.

The gap between mentions and citations matters. Semrush's AI Visibility Index 2026, built on 126 million US prompts collected between January and April 2026, found that engines often name brands without citing their websites, and vice versa. Wikipedia's mentions reportedly fell 50% over those four months while its citations held up. Semrush rolls its inputs into a single AI Visibility score from 0 to 100, without publishing the formula.

Amplitude's Visibility Score, launched on October 30, 2025, counted how often a brand appeared across hundreds of prompts on ChatGPT, Claude and Google AI Overviews. Evertune's AI Brand Score, distributed through impact.com from December 2025, estimates how likely a model is to recommend a brand when the prompt does not name it, drawing on more than 1 million AI responses a month per brand. Ahrefs' Brand Radar distinguishes pages a model found during retrieval from pages it actually cited, and added Grok on April 24, 2026.

First-party data from the platforms

A second category of data comes from the engine operators themselves. Rather than simulating users, it reports what happened to real queries.

Bing Webmaster Tools opened an AI Performance report in public preview on February 10, 2026, covering Copilot and Bing's AI summaries. It shows total citations, cited pages and grounding queries, the searches the AI system ran to fetch sources, which often differ from what the user typed. Microsoft Clarity, its free analytics product, made a Citations dashboard generally available on May 13, 2026, adding "share of authority": a domain's citations divided by all citations, calculated daily.

Google followed in June 2026 with a generative AI performance report in Search Console, covering AI Overviews, AI Mode and Discover. It reached every site worldwide by August 31, 2026, with impressions, pages, countries and devices but no clicks and no queries. Merchant Center gained AI performance insights, including share of voice, on May 27, 2026.

None of these reports covers rival platforms. Bing says nothing about ChatGPT, and Search Console says nothing about Copilot.

Origin and evolution

The idea predates the label. A paper posted to arXiv on November 16, 2023 by researchers at Princeton University and the Indian Institute of Technology (IIT) Delhi, which coined GEO, measured a source's "visibility" inside generated answers by its share of the text, weighted by position.

Commercial use followed Google's launch of AI Overviews in May 2024 and ChatGPT search that October. By late 2025, products were arriving monthly. Adobe made LLM Optimizer generally available on October 14, 2025, claiming a 1,100% year-on-year rise in AI traffic to US retail sites in September 2025.

Microsoft's Clarity team shipped three citation releases in 25 days across July and August 2026, including Topic Insights on July 9 and a branded versus non-branded split. Cloudflare announced an AEO Visibility Dashboard on August 6, 2026, in early access, probing only Claude and GPT models from a benchmark panel built once per category.

Why it matters for marketers

When an assistant shortlists three brands, the rest are absent, and the absence leaves no trace in web analytics. Semrush found that only 36 of more than 1,200 brands it tracked stayed in the top 100 on all four platforms in every month, and that ChatGPT cited 15.4 sources per response against 3.3 for Gemini. The same company can be prominent on one engine and missing on another.

Wording matters as much as platform. Recrawled, a UK answer engine optimisation (AEO) agency, put 1,608 questions to four platforms on September 4, 2026 about 40 UK software firms. Nine were never named in any of their 24 shortlist answers, and the share of firms named by at least one platform fell from 63% for UK-specific questions to 0% when a question pitted the market leader against its nearest rival. Recrawled sells AEO services.

Publishers sit on the other side: when no click follows, a citation count is the only sign their pages were used.

Limitations and disputes

The central objection is statistical. AI answers are non-deterministic: the same prompt can return different brands on each run. Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai had 600 volunteers run 12 prompts 2,961 times on ChatGPT, Claude and Google's AI in November and December 2025. According to SparkToro's write-up of January 27, 2026, the chance of two identical brand lists was under 1 in 100, and of the same order roughly 1 in 1,000. Fishkin wrote that "any tool that gives a 'ranking position in AI' is full of baloney", while accepting that "visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric". He disclosed that the work was not peer-reviewed and was done with a partner whose business is visibility tracking.

The IAB accepts the premise. It treats a single response as a sample of one, asks for results as ranges such as 22% plus or minus four points, classes programmes with fewer than 50 queries as exploratory, and warns that retraining alone can move a brand's share of voice five points overnight. It has not set acceptable variability thresholds. It also notes that vendors produce different results for the same brand.

Personalisation compounds the problem. In November 2025 the SEO consultant Lily Ray showed that ChatGPT rewrites prompts using location, memory and custom instructions, so a tracker's clean account rarely sees what a customer sees. Andrea Volpini of WordLift said there was "no reliable way to track LLM users".

Prompt libraries are the weakest point, since vendors choose the questions; the IAB requires providers to state library size, refresh cadence and whether queries are synthetic.

Then there is the question of what to do with the number. Benjamin Houy shut down his tracking tool, Lorelight, on October 31, 2025, after customers churned because the data did not change what they did. "There was no secret formula," he wrote. Visibility can also be gamed: an Ahrefs experiment with a fictitious brand found most platforms repeated detailed fabrications over the official FAQ, though Roger Montti of Search Engine Journal argued that 49 of its 56 prompts were leading questions.

AI visibility and adjacent terms

Search ranking is a position in an ordered list, stable enough to report daily. AI visibility is a frequency across sampled answers. Both the IAB and SparkToro reject a single rank as meaningful for AI.

GEO and AEO are names for the work of improving AI visibility, not for the measurement. The labels are contested, as the GEO explainer sets out; AI visibility is the outcome they claim to move.

Referral traffic counts visits arriving from an AI platform. A brand can be named in thousands of answers and receive no visits, which is why the absence of clicks from Search Console's report matters to publishers.

Share of voice is one input into AI visibility, not a synonym. In display advertising, viewability is unrelated: it measures whether an ad impression had a chance to be seen on screen.

Recent developments

Scrunch reported on September 24, 2026 that a cited self-promoting listicle lifted its author's recommendation rate from about 4% to 7%, a small, sampled shift of the kind the IAB says should be read as a range. Microsoft says its own visibility tools are for "directional monitoring and insight rather than as a source of guaranteed accuracy", in the July 2026 Clarity Topic Insights release. As of October 2026, no audited standard exists. The next fixed date is December 3, 2026, when the main conduct requirements set by the UK Competition and Markets Authority (CMA) on Google's AI features for publishers take effect.

Timeline

  • November 16, 2023: Princeton and IIT Delhi researchers post the GEO paper, measuring source visibility inside generated answers
  • May 2024: Google launches AI Overviews in the US
  • October 31, 2024: OpenAI launches ChatGPT search
  • October 14, 2025: Adobe makes LLM Optimizer generally available
  • October 30, 2025: Amplitude launches AI Visibility
  • October 31, 2025: Benjamin Houy announces the shutdown of Lorelight
  • November 3, 2025: Lily Ray documents ChatGPT prompt personalisation and its effect on tracking tools
  • December 16, 2025: impact.com and Evertune announce AI brand monitoring reports
  • January 27, 2026: SparkToro publishes its study of AI recommendation consistency
  • February 10, 2026: Bing Webmaster Tools opens its AI Performance report in public preview
  • April 24, 2026: Ahrefs adds Grok to Brand Radar
  • May 13, 2026: Microsoft Clarity Citations becomes generally available
  • May 27, 2026: Google adds AI performance insights, including share of voice, to Merchant Center
  • June 3, 2026: Google announces the generative AI performance report in Search Console
  • June 29, 2026: Semrush publishes its AI Visibility Index 2026
  • July 9, 2026: Microsoft Clarity releases Topic Insights
  • August 3, 2026: The IAB publishes Measuring Visibility in the AI Era
  • August 6, 2026: Cloudflare announces its AEO Visibility Dashboard in early access
  • August 31, 2026: The Search Console AI report reaches all sites worldwide
  • September 24, 2026: Scrunch publishes its study on self-ranking listicles
  • September 29, 2026: Recrawled publishes The AI Shortlist 2026
  • December 3, 2026: The CMA's main conduct requirements on Google's AI features for publishers take effect

Summary

Who. Brands, agencies and publishers track AI visibility, using vendors such as Semrush, Ahrefs, Adobe, Amplitude, Evertune and Cloudflare, plus first-party reports from Microsoft and Google. The IAB published the first industry guidance in August 2026; researchers including Rand Fishkin have tested its reliability.

What. A measure of how often and how prominently a brand or website is named, cited or recommended in AI-generated answers, usually expressed as mention rate, citation rate, share of voice or a proprietary score. It has no standard definition and no audited methodology.

When. The concept appeared in academic GEO research in November 2023. Commercial tools multiplied from late 2025, platform reports arrived from February 2026, and the IAB guidance followed on August 3, 2026.

Where. In ChatGPT, Gemini, Claude, Perplexity, Copilot, Grok and Google's AI Overviews and AI Mode, measured through vendor dashboards, Bing Webmaster Tools, Microsoft Clarity, Google Search Console and Merchant Center.

Why. AI answers shortlist a few brands and settle many queries without a click, so presence in the answer has become a commercial concern. Because answers vary from run to run and by user, the figures are estimates from samples, and how far they can be trusted remains disputed.