ChatGPT Search and Bing: Why Brand Visibility Depends on Bing's Index

87% of ChatGPT Search citations come from Bing's index, not Google's. If your brand ranks perfectly on Google but is absent from Bing, ChatGPT doesn't know you exist. This is the main finding from a study by Seer Interactive analyzing over 500 SearchGPT citations, and it flips much of what Italian SEO professionals take for granted.

In Italy, Bing has a 3.24% overall market share, 8.22% on desktop (Statcounter, 2025). For years it was dismissed as a marginal channel. But ChatGPT already has 8.8 million monthly active users in Italy, equivalent to 20% of connected Italians, with 35% penetration in the 18-24 age group (Statista/DataReportal estimates, April 2025). When these people ask ChatGPT for a hotel recommendation, consulting firm, or business software, the answers are built by pulling from Bing's index.

In this article I explain how ChatGPT Search works technically, what the April 6, 2026 Search Engine Land study says about the Baccarat vs Fifth Avenue Hotel case, which signals actually determine visibility on these systems, and what you can concretely do if you run a business in Italy.

How ChatGPT Search works under the hood

ChatGPT Search combines two knowledge sources: the model's parametric memory (training data) and real-time search via Bing.

The first layer is the model's parametric knowledge, accumulated during training on huge text corpuses up to the cutoff date. The second layer is real-time search via Bing, activated whenever the question needs current information or requires identifying specific brands and products. OpenAI calls this second system "Prometheus": it's an integration layer that queries the Bing API, collects results, and passes them to the model as additional context.

How query fanout works

The most relevant mechanism for understanding visibility on ChatGPT is query fanout. When a user asks something like "what's the best luxury hotel in New York?", ChatGPT doesn't pass it directly to Bing as-is. It breaks it down into multiple parallel sub-searches, each formulated differently to maximize coverage of results.

Real data: on average 1.78 sub-searches per prompt, with 75% of questions generating exactly 2 fanouts. Some complex searches can reach 20, but the "15-25 sub-searches" numbers circulating online are exaggerated. (Nogami & Tannenbaum, Search Engine Land, April 6, 2026, 68 iterations with GPT-5.2 Instant, 25 unique queries)

The key point for visibility: each sub-search queries Bing, not Google. The results returned by Bing are synthesized by the model to build the final response, with selective citations. On average ChatGPT cites only 3-4 brands per response (Onely, 2025). Everyone else, even if excellent on Google, doesn't exist for that particular user.

To understand how this mechanism intersects with Google's AI Overviews, which follow different logic, read my analysis on GEO and AI Overviews in content strategy 2026.

The complete flow of a ChatGPT response

The path of a response with citations follows these steps. The user's question is analyzed by the model, which decides whether it needs real-time data. If so, it generates the fanout sub-searches and sends them to Bing. The results (titles, snippets, URLs, partial page content) are passed as context to the model. The model synthesizes a response and selects which sources to cite. Source selection is not random: it depends on relevance to the specific fanout query, position in Bing's SERP, and page content.

The study that flipped the certainties: Baccarat vs Fifth Avenue Hotel

On April 6, 2026 Search Engine Land published an analysis by Nogami and Tannenbaum that precisely documents the paradox of AI visibility. The case involves two luxury hotels in New York, but the lesson applies to any business category.

Baccarat Hotel dominates Google for relevant queries. It has nearly six times more reviews. It has a higher price and a more established history. Yet ChatGPT mentions it in fewer than 2 out of 100 iterations. Fifth Avenue Hotel, opened in 2023, with a third of the reviews and 40% lower price, is cited 19% of the time.

The explanation is precise: Fifth Avenue Hotel dominates Bing results for queries generated by fanouts. Baccarat dominates Google, which ChatGPT doesn't consult.

The detail that changes everything: URL specificity

The analysis by Nogami and Tannenbaum highlighted another aspect that usually doesn't emerge in high-level discussions of AI search. It's not enough to be cited by an authoritative outlet: which specific section of that outlet cites you matters.

In the Baccarat case, Forbes Travel Guide positions it fourth, which contributes to visibility on Google. But Forbes Personal Shopper, which is likely queried by fanouts for queries like "best luxury hotel for shopping in NYC", doesn't mention Baccarat. The model receives context from Forbes Personal Shopper, not Forbes Travel Guide. Same publication, different section, opposite result.

This has concrete practical implications for digital PR strategy. It's not enough to get a generic mention on an authoritative outlet: the mention needs to appear in the section or type of content that answers the specific queries generated by fanouts for your industry.

The researchers also tested three different ChatGPT memory states (no memory, active memory, reset memory): no significant difference in results. Visibility depends on Bing's index, not the user's conversation history.

Numbers every marketer should know

Seer Interactive analyzed over 500 SearchGPT citations comparing them with Bing and Google results. The main finding is already known: 87% overlap with Bing's top 20, only 56% with Google. But there are other numbers worth attention.

The most counterintuitive finding is about pages with no Google visibility: 28% of the pages most cited by ChatGPT don't appear in Google's organic results. This means entire content segments exist that Bing indexes and ranks well but Google overlooks or ranks differently. For a brand, this opens an opportunity window that exclusive Google presence doesn't allow.

The value of a visitor from AI search

Why invest in this channel? Onely data shows that a visitor from an AI search averages 4.4x more valuable than a visitor from traditional organic search, with a conversion rate 2.3x higher. The explanation is plausible: someone arriving from ChatGPT has already received a synthetic answer and a specific brand recommendation. The click isn't exploratory, it's intentional.

On the cost side: the CAC (customer acquisition cost) via GEO is $559 versus $612 for traditional SEO (Onely, 2025), with an average time to reach results of 89 days versus 127 days for classic SEO. The numbers aren't universal, but the trend is consistent across multiple independent studies.

For a broader analysis of how AI is redefining search and online visibility, AI and SEO in 2026 provides the complete picture.

Why traditional SEO signals no longer suffice

The Onely study is explicit on this point: traditional SEO signals like backlinks, domain authority, and keyword density have "near-zero" influence on ChatGPT visibility. Not that they don't matter at all. It's that they matter far less than signals that determine position on Bing, and Bing weighs them differently than Google.

The signals that correlate most strongly with AI visibility, according to available data, are different. Attention: these are observed correlations, not proven causal relationships.

• Mentions in authoritative "best of" lists: 41% of overall signal. Being included in "Best Legal Firms in Milan according to Corriere", "Top 10 SEO Agencies in Italy according to Trustpilot" or similar lists from recognizable outlets is the most powerful signal. It's not just the list site that counts: the outlet's editorial authority and the specific section where the mention appears matter.
• Industry awards and credentials: 18% of the signal. Certifications, recognitions from industry associations, documented awards on third-party pages help build what researchers call "topical authority" in Bing's index.
• Online reviews: 16% of the signal. Not absolute numbers, as the Fifth Avenue vs Baccarat case demonstrates, but distribution and recent frequency. Relevant platforms vary by industry: Google Business Profile for local businesses, Trustpilot for B2B services, TripAdvisor for hospitality.
• Wikipedia and Wikidata: 47.9% of ChatGPT's top 10 sources include Wikipedia content (Profound, 2025). For brands with relevant presence in their industry, having a verifiable Wikipedia entry is an authority signal that the model uses both in training and in real-time synthesis.

URL specificity matters more than domain

The Forbes case that emerges from the Nogami & Tannenbaum study introduces a principle that applies to the entire AI-oriented link-building strategy. A backlink or mention from an authoritative domain is not equivalent if they come from sections with different editorial purposes.

A mention in a "best of" guide on a general outlet is worth more than a mention in a news article on the same outlet, even though the domain is identical. The reason is that ChatGPT's fanout queries are specific: "best SEO consultant Milan 2025" intercepts a "best of" page, not the company homepage or a news article about it.

Structured data play an important role in this context. To understand how to implement them correctly, the guide to structured data and Schema.org is the most practical technical reference.

Structured data: +40-50% citation probability

Onely reports that pages with FAQ schema have a 67% citation rate in AI responses, versus much lower for pages without markup. More broadly, structured data presence increases citation probability by 40-50%.

The mechanism is logical: Schema.org markup helps Bing (and consequently the ChatGPT model) understand precisely the type of content, the described entity, and semantic context. A page with Organization schema, FAQ schema, and sameAs links to Wikipedia, Wikidata, and LinkedIn is interpreted with less ambiguity than plain text.

Practical checklist for the Italian market

How do you appear on ChatGPT, in practice? The actions that follow are ordered by expected impact and implementation ease. You don't need to do all at once: the first four are the minimum foundation for starting from zero.

1. Bing Webmaster Tools and IndexNow

The first step is verifying your site on Bing Webmaster Tools, which is free and takes less than 30 minutes. If you already use Google Search Console, you can import verification directly.

IndexNow is an open protocol (not proprietary to Microsoft) that allows you to notify Bing (and other engines that support it) of page publication or update in real time. As of April 2026, according to Microsoft data, it processes about 5 billion URLs per day across over 80 million sites. Google doesn't support it, but Bing, Yandex, and others do. Implementing it reduces indexing delay on Bing from weeks to hours.

To implement: generate an API key from Bing Webmaster Tools, place a validation file in the site root, and add a POST call to the IndexNow endpoint every time you publish or update content. Most CMSs (WordPress, Webflow, Next.js) have native plugins or hooks to do this automatically.

2. Bing Places for Business

If you have a local business or a business with a physical location, Bing Places for Business is the Bing ecosystem equivalent of Google Business Profile. It's free and, for those with a complete Google Business Profile, allows direct data import with a few clicks. The Bing Places card feeds Bing's local results and, consequently, the data source on which ChatGPT builds its responses for queries with local intent.

3. Structured data: Organization, FAQ, and Product schema

Three types of markup have the most direct impact on ChatGPT visibility.

• Organization schema: includes name, URL, logo, description, and most importantly, the sameAs field with links to Wikipedia, Wikidata, LinkedIn, and verified social profiles. The sameAs field helps AI systems connect scattered brand mentions to a single, verified entity.
• FAQ schema: pages with FAQ schema have a 67% citation rate in AI responses (Onely). Implement it on service pages with the most frequent questions in your industry. Not customer service FAQs, but ones that answer the informational questions potential clients ask ChatGPT before choosing a provider.
• Product/Service schema: if you sell products or services with quantifiable features (price, duration, geographic area served), specific markup helps Bing correctly categorize your offer in comparative queries.

4. Digital PR on relevant Italian outlets

This is the action with the greatest long-term impact but also the most labor-intensive. The goal is to appear in "best of" lists on Italian outlets that Bing indexes with authority for your specific industry.

For an SEO consultant or digital agency in Italy, relevant outlets include Ninja Marketing, Corriere Comunicazioni, Il Sole 24 Ore Technology, PMI.it, digital.corriere.it. For a hotel or restaurant, Gambero Rosso, TripAdvisor, Identità Golose. For a professional firm, Italia Oggi, Il Denaro, industry guides from professional associations. The logic is always the same: authoritative outlet, specific section, comparative context ("the best X in Y").

5. Managing AI crawler directives in robots.txt

Here there's a decision to make consciously. Available data show that 69% of top-ranking sites block ClaudeBot and 62% block GPTBot in robots.txt (Paul Calvano, analysis on July 2025 data). The motivation is often protecting content from model training.

But there's an important technical distinction: blocking GPTBot prevents crawling for OpenAI's training, not Bing crawling. ChatGPT Search uses Bing's index (via Bingbot) for real-time responses. Blocking GPTBot in robots.txt doesn't reduce ChatGPT Search visibility. Blocking Bingbot does. It's also worth noting that OpenAI introduced OAI-SearchBot, a crawler dedicated to real-time search distinct from GPTBot: if you allow OAI-SearchBot but block Bingbot, your pages might still appear via OpenAI's direct crawler. Before modifying robots.txt, verify which bots you're actually blocking and what the overall impact on visibility is.

6. The llms.txt file: reality vs hype

Some guides suggest creating an llms.txt file (a text file in the site root that declares how AI models should use the site's content). It's honest to be clear on this: adoption rate is estimated around 2% (Search Engine Land, April 2026) and there's no verified evidence that it has measurable impact on ChatGPT or other AI system visibility. It's not a priority. If you've already optimized structured data and Bing presence, you can add it as a low-cost experiment, but don't expect concrete results at this stage.

7. Monitoring brand visibility on ChatGPT

How do you measure whether your brand appears in ChatGPT responses? There are two main paths. The first is manual monitoring: define a set of 10-20 questions that your potential customers might ask ChatGPT and regularly test the responses. It's not scalable, but it's free and gives immediate qualitative insights.

The second is using specialized tools. DataForSEO has introduced endpoints for AI mention monitoring, Semrush has added AI visibility features to its suite. They're still in development but work for those needing to monitor a large set of keywords.

For those who want to integrate SEO data into automated workflows, the article on how I use DataForSEO via MCP with Claude Code shows a practical approach to programmatic monitoring.

What changes for Italian SMBs

In Italy only 3.24% of searches go through Bing (Statcounter, 2025),

but ChatGPT usage has reached 8.8 million monthly active users. This creates an asymmetry: your traditional SEO efforts are optimized for 90%+ of traffic (Google), while an emerging 20% of connected Italians use a system that's 87% powered by Bing. For an SMB, this means the cost-benefit calculation changes.

If your website generates €100,000 per year and 5% comes from organic search, that's €5,000 per year from SEO. If a fraction of that traffic—even 5-10%—starts coming from ChatGPT, it's not enough to overhaul your entire strategy. But if your competitive set on Bing is much thinner than on Google because nobody has optimized for Bing yet, the ROI of tactical Bing work can be immediate.

The barrier isn't technology. The barrier is that Italian digital marketers aren't yet thinking about Bing as a channel separate from Google, and they don't distinguish between traditional SEO visibility and AI search visibility. Once that distinction becomes clear—which it will when ChatGPT usage reaches 25-30% of Italy's connected population—the competitive advantage of being early is real.

If you want a complete perspective on how to adapt your strategy across Google, Bing, and AI answer engines, get in touch via the contact page. I analyze which channels matter most for your market and which actions have immediate ROI.