ai optimisation
Three different AIs,
One single goal:
becoming the answer
ChatGPT, Gemini and Claude do not work in the same way. They have different datasets, different citation mechanisms and their own update logic. Confusing them is often the first mistake.
Bliss manages optimisation for generative AI systems as an integrated whole: the same actions that strengthen organic SEO on Google drive citability in ChatGPT and Claude.
Each platform has its own priorities. Each presence requires specific action. Bliss governs presence in generative AI systems as an integrated system.
The context
The way people search has changed
Search behaviour is changing structurally. Users, especially those with high spending power and business decision-makers, now query generative AI systems as they would an advisor. They do not search “best SEO agencies Rome” and then open ten tabs: they ask ChatGPT or Gemini directly for a recommendation, and they trust the answer.
And so, a brand that does not appear in AI answers effectively ceases to exist for that portion of the market, however well positioned it is organically on Google. The Search Generative Experience is compressing traffic to the traditional top organic positions while opening a new channel of visibility: citations in generative answers.
Optimisation for these systems is the natural evolution of SEO. The same variables that determine a domain’s authority on Google, editorial quality, authoritative backlinks and the brand’s semantic consistency, now determine the likelihood of being cited by ChatGPT, Gemini and Claude. The difference lies in the level of specificity required for each platform.
+527%
+40%
20–30%
Citability factors
What determines whether an AI
cites your brand or your competitor's
Academic research on GEO (KDD ’24 paper, Politecnico di Milano) has identified the factors that statistically increase the likelihood of a generative AI system citing a piece of content in its answer. These factors are not arbitrary: they reflect how language models assess the authority and citability of information.
The cross-cutting factors
They work across all three systems
Some optimisation elements increase citability on ChatGPT, Gemini and Claude simultaneously. They are the mandatory starting point of any GEO strategy, because they deliver results across all platforms without requiring customisation for each one.
The three platforms
How each AI works
and how to optimise for each one
Every AI model builds its own view of the world. ChatGPT, Gemini and Claude do not share the same knowledge base, do not update on the same timescales and do not give weight to the same sources. Bliss covers each of these models with an action plan that separates cross-platform actions from platform-specific ones.
ChatGpt
How ChatGPT ®
decides
what to cite and why
OpenAI · 180M+ active users
ChatGPT operates in two modes. In no-browsing mode, answers come exclusively from training data, that is, from the information acquired while the model was trained, up to a cut-off date. As a result, a brand without a significant editorial presence in the training corpus does not exist for that version of the model.
In web browsing mode (ChatGPT Plus and the API with tools), the system uses Bing as its search engine and synthesises answers from what it finds online. ChatGPT’s citation behaviour is straightforward: it cites the sources it considers most authoritative and relevant to the specific query. A brand cited frequently in recognised publications is more likely to be included as a source.
Specific actions for ChatGPT
Editorial presence in outlets that ChatGPT recognises
Content with a Q&A structure and a direct answer up front
Proprietary data and original, citable statistics
Up-to-date, consistent Wikipedia and Wikidata pages
Backlinks from .gov, .edu and academic sources where possible
llms.txt to communicate the site's structure to the model
Google Gemini®
and AI Overviews:
SEO becomes GEO
Google · AI Overviews in millions of searches
Gemini is the AI system most directly connected to traditional SEO ranking.
Google’s AI Overviews (the generative boxes that appear above the organic results) are powered by Gemini and use the same signals as organic ranking: E-E-A-T, domain authority, backlink quality, Core Web Vitals, structured data.
Here the difference from traditional SEO lies in the type of content favoured: pages optimised for traditional rankings are not automatically optimised for citability in AI Overviews.
Gemini’s advantage over the other systems is measurability: Google Search Console makes it possible to monitor impressions for queries where AI Overviews appear, and AI Overview traffic can be tracked more precisely than on ChatGPT or Claude.
Google Gemini-specific actions
FAQPage schema on every page with real questions
Organization schema with sameAs on the Knowledge Graph
Content that answers "who / what / how" questions
Factual statement at the opening of each H2 section
Core Web Vitals within Google's quality thresholds
Monitoring AI Overviews in Google Search Console
Claude® the AI system most used in B2B and enterprise settings
Anthropic · used by businesses and developers
Claude (Anthropic) is trained on a corpus that favours high-quality editorial content, journalism, academia and technical documentation. Its user profile differs from ChatGPT’s: it is more widespread among developers, professionals and companies that integrate it into their workflows via API. The likelihood of being cited in Claude depends not on how many people use the consumer product (claude.ai), but on how many companies have integrated Claude into their internal processes.
Claude has access to web browsing in some configurations, but its baseline knowledge of the brand depends mainly on training data, with news outlets, academic papers and technical documentation perceived as authoritative.
For B2B brands, Claude is the system to secure most urgently, because the managers who use it for comparative research on suppliers and partners do not open Google. They trust the answer they receive implicitly.
Claude-specific actions
Presence in international publications (English and Italian)
llms.txt to communicate structure and strategic pages
Presence on Crunchbase, LinkedIn and industry directories
Citable technical documentation and white papers
Content with high information density and a clear structure
Case studies with verifiable data and declared methodology
the method
Five levels.
One system.
Before applying this work to its partners, Bliss applied it to itself.
The starting point was around 10 active referring domains, no coverage in national publications, no Wikipedia entry, and AI engines that did not cite Bliss in any context. The signals that AI systems use to build a brand’s profile were absent or inconsistent.
We worked on each of these elements: building an editorial presence in publications that AI models could recognise as authoritative, making the signals distributed across the web consistent, and optimising the site’s semantic structure so that it is readable, not just indexable.
One year on, Bliss appears in ChatGPT, Gemini and Perplexity answers to strategic industry queries: “migliori agenzie branding Italia”, “agenzia comunicazione lusso Roma”, “consulenza GEO Italia”. This is the work of AI optimisation, built on structured signals, authoritative sources and a consistent semantic identity.
How do ChatGPT, Gemini and Claude describe you right now?
The first step is a snapshot of the current state: establishing how the three AI systems analyse the brand, cite it and position it against competitors. This audit identifies presence gaps and opportunities for differentiation.
The brand as an entity that AI models recognise.
Generative AI systems read entities. A brand without a clear definition in the Knowledge Graph is read ambiguously or ignored by the models. In this phase we implement the complete Organization schema with all relevant fields to optimise the consistency of information and build the map of semantic relationships between the brand and the entities in its sector.
Content that models choose as sources.
It is the structure of content that determines whether an AI system will use it as a source in its answer. By working on the architecture of strategic content, in this phase we act to maximise citability, with the aim of producing content that a language model "sees" as the most complete and verifiable answer to a specific query.
The editorial presence that feeds training data.
Language models learn by reading the web. The sources they regard as authoritative (national and international news outlets, academic papers, recognised industry portals) become the sources they draw on in their answers. To build our partners' editorial presence, we define an LLM Digital PR plan. The result is visibility in generative systems that accumulates over time and does not stop when advertising spend stops.
Monitor share of voice in AI answers.
Once these steps are complete, ongoing oversight is applied through periodic measurement of visibility in AI systems. From here, our proprietary framework produces a monthly report that combines the most relevant data with traditional SEO metrics in a single dashboard, able to support positioning over time.
Case study
How Bliss built its own presence in AI answers
We applied the LLM Digital PR method to ourselves before offering it to our partners. The starting point was a brand with solid offline foundations but almost no digital editorial presence: around 10 active referring domains, no Wikipedia entry, no coverage in national outlets. ChatGPT, Gemini and Claude did not cite Bliss in any context.
A year on, Bliss now appears in ChatGPT, Gemini and Perplexity answers to strategic industry queries (“best branding agencies Italy”, “luxury communications agency Rome”, “GEO consultancy Italy”). All this because we built a network of editorial citations that AI models can read as a signal of authority.
FAQ
Frequently asked questions on optimisation for ChatGPT, Gemini and Claude
How do you optimise a website for ChatGPT?
ChatGPT with web browsing uses Bing as its engine and favours structured content, verifiable factual data, citations of authoritative sources and direct answers to specific questions. To increase the likelihood of being cited: content with a clear semantic structure, presence in sources ChatGPT considers authoritative, schema markup that defines the brand entity, and an editorial backlink profile from recognised publications.
For the no-browsing version, training data is the decisive factor: editorial presence in publications that form part of the training corpus of OpenAI’s models is the only way to enter the model’s knowledge base.
How do you appear in Google Gemini's AI Overviews?
Google Gemini powers AI Overviews using the same signals as organic ranking (E-E-A-T, domain authority, content quality), with greater weight given to content that answers specific questions directly and concisely.
To optimise: FAQPage and HowTo schema on pages that answer questions, factual statements at the opening of each section, semantic domain authority built through editorial backlinks, and integration into Google’s Knowledge Graph through Organization and sameAs schema.
What is GEO (Generative Engine Optimization)?
GEO is the discipline that optimises a brand’s presence in generative AI systems (ChatGPT, Google Gemini, Perplexity, Claude), which produce synthesised answers instead of lists of links. GEO works on the brand’s semantic authority as an entity, on the citability of content and on presence in the sources that language models use as references.
It complements traditional SEO: the same signals that improve organic ranking also increase the likelihood of being cited in generative answers. Bliss manages SEO and GEO as a single system.
How does optimisation for Anthropic's Claude work?
Claude is trained on a corpus that favours high-quality editorial content. Its knowledge of a brand depends mainly on its training data and, when connected to the web, on the sources it finds. Claude’s user base is predominantly B2B: companies and professionals who integrate it into their workflows via API.
To increase citability in Claude: editorial presence in recognised publications, content with high information density, schema markup that defines the brand entity unambiguously, an llms.txt file, presence on Crunchbase and LinkedIn.
Are traditional SEO and GEO mutually exclusive?
No, they reinforce each other. The same signals Google uses for organic ranking (domain authority, E-E-A-T, structured data, editorial backlinks) are also the signals AI systems use to determine which sources to cite. A site with excellent organic SEO is already well positioned for Gemini’s AI Overviews.
GEO adds a further layer: it specifically optimises semantic citability, answer structure and presence in the sources that feed ChatGPT and Claude. Bliss manages the two systems as a single investment plan.
How is visibility in AI systems measured?
No standardised tools exist yet. Bliss uses a proprietary framework based on: AI Brand Mentions (frequency of unprompted citations), Sentiment Bias Analysis (how AI systems describe the brand), SGE Feature Rate (percentage of queries in which content appears in Google’s AI Overviews), and tracking of AI-referred sessions in Google Analytics 4.
The monthly report combines these metrics with traditional SEO data in a single dashboard.
How long does it take to be cited in AI answers?
Timescales vary by system. Google Gemini’s AI Overviews follow the logic of organic ranking: three to six months for the first significant changes. ChatGPT with web browsing updates its answers in real time: the results of editorial work can be seen within weeks.
The knowledge embedded in the models (training data) is updated with each new model version. Ongoing oversight is necessary because models are updated and the competition evolves.
What is the Knowledge Graph and why does it matter for GEO?
Google’s Knowledge Graph is the database of entities (companies, people, places, products) and their relationships. When Gemini or Google Search generates an answer about a brand, it draws on the Knowledge Graph to understand who that entity is and what it does. A brand that is absent or poorly represented is read less accurately and appears less often in generative answers.
Integration is achieved through correct Organization schema markup, Wikipedia, Wikidata, Google Business Profile, a presence on authoritative platforms (LinkedIn, Crunchbase) and absolute consistency of information across all web sources.