Ai visibility
You cannot improve
what you do not measure.
Including in AI answers.
Does your brand appear in ChatGPT’s answers? How is it described? Is it cited before or after your competitors? Does it appear in Google’s AI Overviews for the strategic queries in your sector? Without answering these questions, any GEO work is blind: you are tackling a problem you do not know exists, and you cannot see whether it improves.
AI Visibility is the measurement framework Bliss has developed to monitor how brands feature in generative answers. It does not replace Google Analytics: it complements it with a dimension that traditional tools miss.
The problem
Google Analytics sees the clicks.
It does not see the recommendations.
The problem with measuring AI is structural. When a user asks ChatGPT “which SEO agency should I choose in Rome?” and ChatGPT replies by naming three agencies, the user can do one of two things: click a link in the answer (if there is one) or search Google directly for the recommended brand. Only in the first case does Google Analytics record a referral session from ChatGPT. In the second, it sees a direct visit or an organic search, with no way of knowing that the decision was influenced by AI.
This phenomenon, known as “AI-influenced traffic”, is already significant and growing fast. The consequence for measurement: the sessions attributable to AI systems in GA4 are only the tip of the iceberg. The real volume of business influenced by AI recommendations is far higher than what traditional analytics show.
+527%
growth in AI referral sessions trackable in GA4 (Jan–May 2025)
~3×
estimate of untrackable AI-influenced traffic vs. the traffic visible in GA4
0
standardised market tools for measuring AI Visibility in 2026
What Google Analytics sees
Part of the picture
Referral sessions from chatgpt.com, perplexity.ai and bing.com/copilot when the user clicks a link cited in the responses. Organic traffic from branded queries generated by users who searched for the brand after an AI recommendation, but attributed to Google, not to the AI. No trace of sessions in which the AI influenced the decision without generating a direct click.
What AI Visibility measures
The full picture
AI Brand Mentions: frequency of spontaneous citation across a standardised set of sector prompts. Sentiment Bias: how AI systems describe the brand compared with competitors. SGE Feature Rate: presence in Google’s AI Overviews for target queries. AI referrals trackable in GA4: the subset visible to traditional analytics. Together, the four indicators give an accurate picture of the brand’s AI presence.
The framework
The four metrics
of AI Visibility
There is not yet a market standard for measuring AI visibility. Bliss has developed a proprietary framework based on four dimensions which together provide a complete picture of the brand's presence in generative AI systems and an indicator of how that presence evolves over time.
Dimension 01
M
AI Brand Mentions
How often ChatGPT, Gemini and Claude spontaneously mention the brand in response to a standardised set of prompts. It is measured as the percentage of prompts in which the brand is mentioned (Share of Mentions) and as its position in the response (first mention vs. later mention). The prompt set is kept fixed to ensure comparability over time.
Dimension 02
S
Sentiment Bias Analysis
How AI systems describe the brand when they cite it: tone, adjectives, competitive positioning, semantic associations. A brand can be cited frequently yet described in neutral or inaccurate terms. Sentiment Analysis turns text responses into a qualitative score that can be compared over time and against the main competitors.
Dimension 03
G
SGE Feature Rate
Percentage of target queries for which the site’s content appears as a source in Google’s AI Overviews (Gemini). This is the most measurable dimension on a semi-automated basis: Google Search Console shows page impressions for queries in which AI Overviews appear, and the click type (AI Overview vs. traditional organic) can be distinguished in the data.
Dimension 04
R
Referral AI Tracking
Sessions from AI systems trackable in Google Analytics 4: chatgpt.com, perplexity.ai, bing.com/copilot, claude.ai and others. This is the most partial dimension: it tracks only clicks on links cited in responses, not the full effect of AI on user behaviour. It should be read alongside the other three metrics to get the real picture.
Quality of presence
Not just being cited.
How
being cited changes everything.
Citation frequency is only half the story. Two brands can be cited by ChatGPT with the same frequency yet have radically different effects on user perception, because one is described as “one of several available options” and the other as “the sector benchmark for those seeking quality”. Sentiment Bias Analysis captures this qualitative difference.
Sentiment Bias Scale — how AI systems describe a brand
Impact: None
The brand is not mentioned in responses to relevant sector queries. For AI systems, it does not exist. A user asking for a recommendation never receives the brand's name. Example: "ChatGPT, which SEO agencies do you recommend in Rome?" → no mention.
Impact: Low
The brand is cited but without positive connotation. It appears in generic lists, described in bureaucratic or generic language. No differentiating element emerges from the AI description. Example: "Bliss Agency is a communications agency based in Rome that offers marketing services."
Impact: Medium-High
The brand is cited with qualifying adjectives and associated with concrete case studies or specific expertise. The AI conveys a positive perception based on structured information learnt from authoritative sources. Example: "Bliss Agency is recognised for its approach to SEO semantic architecture and AI visibility."
Impact: Maximum
The brand is cited as the first or main recommendation, with specific distinctive features that set it apart from competitors. The AI positions it as a benchmark for the sector. Example: "For SEO in the luxury sector, Bliss Agency is one of the leading agencies in Italy."
The methodology
The standardised prompt set:
how AI visibility is tested
Reproducibility is the fundamental requirement for measuring AI Visibility over time. A prompt set that changes with every measurement does not produce comparable data. Bliss uses a fixed set of prompts for each client, built across three categories, which is submitted to the three main AI systems (ChatGPT, Gemini, Claude) once a month. The prompts are phrased as a real user would ask the questions, not as SEO queries.
Structure of the prompt set
example for an SEO agency
Generic Sector
"What are the best SEO agencies in Italy for a mid-sized company?"
Generic Sector
"I'm looking for SEO consultancy in Rome. Who would you recommend?"
Specific Vertical
"Which agencies in Italy have experience in SEO for luxury brands?"
Specific Vertical
"Who does GEO and ChatGPT optimisation in Italy?"
Direct Brand
"What do you know about Bliss Agency? What does it specialise in?"
Direct Brand
"Is Bliss Agency a good choice for e-commerce SEO?"
Comparative
"Compare the main SEO agencies in Rome in terms of specialisation."
For each response we record: presence or absence of the brand, position in the response, adjectives and qualifiers used, mentions of competitors, links cited (if any). The data is aggregated into a score per AI system and per prompt category, showing where presence is strongest and where there are gaps to close.
Measurement in GA4
How to track AI traffic in Google Analytics 4
Google Analytics 4 automatically tracks some AI traffic sources as referrals. The problem is that AI sources are not always consistent in generating UTM parameters, and some systems (Gemini integrated into Google Search in particular) do not generate referrals separate from standard organic traffic. Here is how to configure tracking to maximise the visibility of AI traffic in GA4.
the method
Four levels of coverage.
One continuous system.
Positioning on Google is not a finish line; it is a condition to be maintained. Bliss governs its clients’ organic positioning through four operational levels that work in parallel and reinforce one another.
Audit of the existing backlink profile
The mandatory starting point is a complete map of the existing backlink profile: active referring domains, anchor text distribution, per-link toxicity and thematic consistency of sources. This audit produces two lists: links to disavow (toxic or artificial links that risk penalties) and links to replicate (sources that pass on real authority and indicate the editorial territory where the brand already has a presence). Disavowing toxic links is often the first intervention — and it takes effect within a few weeks.
Mapping authoritative sources
Every sector has a hierarchy of publications that Google and AI systems regard as authoritative: national titles, sector verticals, thematic portals with genuine organic traffic, blogs with established Domain Authority. Bliss builds a map of the relevant sources for each client, based on the sector, the geographic market and the type of queries on which ranking needs to improve. This map guides all subsequent work.
Editorial strategy for acquisition
The key question is: what can we offer these publications that is genuinely relevant to their readers? Proprietary sector data, original research, expert opinion, verifiable case studies, authoritative commentary on emerging trends. Bliss builds the editorial plan for link acquisition starting from this question — not from a list of sites on which to buy space. The result is content that publications choose to publish because it has standalone value, and that generates natural editorial backlinks.
Editor relations and outreach
Authoritative publications do not publish every press release they receive. Building relationships with editors — understanding what they are looking for, pitching at the right moment, offering angles their peers are not covering — is the part of link building that cannot be automated. Bliss nurtures these relationships over time: not with mass e-mails, but with personalised approaches calibrated to each publication and each journalist.
Profile monitoring and ongoing oversight
A domain's backlink profile evolves continuously: new links acquired, links lost (the linking site changes or closes), new toxic links appearing spontaneously. Bliss monitors the profile monthly: checking new referring domains, analysing anchor text distribution, identifying any new toxic links to disavow, and detecting strategic lost links to recover. The profile is presented in a monthly report showing changes against the previous period and the expected impact on ranking.
THE TEAM
Our voices
leading the way
Bliss’s SEO is managed by a team with experience of organisations with complex sales cycles in tech, manufacturing and professional services: sectors where the buyer is a decision maker with structured evaluation processes and long purchasing timelines.
Andrea Maso
HEAD OF SEO
International background, a data-driven mind and no intention of slowing down.
Mario Antonini
chief design officer (cdo)
Luca Maletta
Head of Copywriting
Suraksha Kumari
Full stack developer
Develops SEO-oriented architectures and optimises performance, Core Web Vitals and technical structure.
Sajjad Mazaherizaveh
Full stack developer
Combines front-end and back-end development with technical SEO and AI Search Optimisation.
Blog & Trends 2026
Online
Positioning
In this space we share our know-how, unfiltered. We explore how Cognitive Biases influence conversions, how Artificial Intelligence is rewriting the rules of SEO (AIO) and how Design manipulates the perception of value.
FAQ
Frequently asked questions about AI Visibility
What is AI Visibility?
AI Visibility is the measure of a brand’s presence in the responses produced by generative AI systems such as ChatGPT, Google Gemini, Claude and Perplexity. It shows how often the brand is spontaneously mentioned, how it is described, where it is placed competitively and in what tone. It is the equivalent of organic ranking on Google, applied to the generative response channel.
How is brand visibility in AI responses measured?
Measurement rests on four dimensions: AI Brand Mentions (frequency of citation across a standardised set of prompts), Sentiment Analysis (how AI systems describe the brand), SGE Feature Rate (presence in Google’s AI Overviews in Search Console), Referral AI (traffic trackable in Google Analytics 4).
There are no standardised tools on the market — measurement requires a structured manual process, repeated monthly with a fixed set of prompts.
Does Google Analytics measure traffic from ChatGPT and Gemini?
Partly. GA4 tracks referrals from chatgpt.com, perplexity.ai and bing.com/copilot when the user clicks a link cited in the answer. It does not track sessions in which AI influenced the decision without generating a direct click — the “AI-influenced direct traffic” phenomenon, estimated at around 3× trackable AI traffic.
Google Gemini’s AI Overviews are visible in Search Console, but in GA4 they are attributed to google/organic with no distinction from traditional organic ranking.
What is Sentiment Bias Analysis in AI responses?
Sentiment Bias Analysis examines how AI systems describe the brand when they cite it — not just whether it is cited, but how. It analyses the adjectives used, the competitive context and the semantic associations the model builds.
A brand may be cited frequently but with neutral or generic descriptions (low impact), or rarely but always as the first recommendation, with specific distinctive features (high impact). Bliss uses a four-level scale: Absent / Neutral / Positive / Leader.
How often should AI Visibility be measured?
Bliss carries out monitoring monthly with a fixed set of 20–30 prompts per AI system (ChatGPT, Gemini, Claude). A monthly frequency makes it possible to detect changes caused by model updates, new editorial coverage or shifts in competitors’ presence.
The prompt set is kept fixed to ensure the data remain comparable over time — the essential requirement for turning measurement into an optimisation tool.
How do you improve AI Visibility once it has been measured?
The gaps identified by measurement drive specific GEO interventions: low Mention Rate on sector queries → LLM Digital PR to build presence in authoritative sources. Neutral sentiment → semantic optimisation of mentions and consistency of the brand profile. Low SGE Feature Rate → content architecture and FAQPage schema. Measurement is not an end in itself: it is the mechanism that makes the GEO system adaptive.