In Italy nobody uses Bing — yet when ChatGPT browses the web to answer in real time, it does so through Bing. This means that every optimization you have done solely for Google counts for nothing when it comes to ChatGPT's live answers: what counts instead is your presence on an engine you have always ignored. Your competitors who have figured this out show up in the up-to-date answers; you don't. The fix is less complicated than it seems.
When you turn on browsing in ChatGPT, the model stops drawing from its training data and connects to Bing. Yes, Bing — the very engine that holds a single-digit market share in Italy and that you have probably ignored for years.
It’s a paradox that few Italian business owners have fully digested: the most widely used chatbot in the world, when it has to answer with up-to-date information, relies on a search engine that is marginal in our country. And from that moment on, the rules of the game become Bing’s rules, not Google’s.
In my articles on how to show up in AI answers I explained that every platform has its own retrieval layer. ChatGPT with browsing enabled has a very specific one: an index that is not dominant in Italy. If your site isn’t there, you don’t exist in the live answers — no matter how good you are on Google.
What Really Happens When ChatGPT “Searches the Web”
When a user asks a question that requires fresh information — a price, a news item, an updated ranking — the model doesn’t make things up. It opens a browsing session, queries Bing, takes the top results, extracts the text of the pages, and synthesizes an answer citing the sources. The mechanism is a direct application of the RAG (Retrieval-Augmented Generation) paradigm, by now standard in the literature.
In the 2025 survey by Gupta and colleagues (A Comprehensive Survey of Retrieval-Augmented Generation (RAG), arXiv:2506.00054 — https://arxiv.org/abs/2506.00054) the principle is formalized like this: RAG systems integrate external knowledge sources into the generative process, allowing the model to produce answers anchored to up-to-date information without having to be retrained. The model doesn’t “learn” in real time: it pulls pages from the web and uses them as context.
From this architecture follows an operational deduction worth spelling out, because it isn’t written in the papers but it’s the only way to read the data with business intelligence: if the source from which ChatGPT pulls the pages is Bing, then everything that determines your presence and your ranking in the Bing index becomes a prerequisite for existing in ChatGPT’s live answers. It’s not an opinion, it’s the logical chain that connects the RAG architecture to your Bing SERP.
The consequence is clear-cut. If you’re not in the Bing index, or if you’re beyond the second page for the queries relevant to your industry, ChatGPT will never see you in the live answers. Retrieval doesn’t reach you, and everything that comes after — citation, mention, link — can’t happen.
Why Bing Isn’t “Google Done Badly”
Anyone who has worked in Italian SEO for twenty years has developed a conditioned reflex: Bing is a negligible share, not worth the investment. That reflex was sound until 2023. Today it no longer is, because Bing is no longer just a search engine: it’s the index that powers ChatGPT browse mode, Copilot, and part of the answers from other AI systems.
Bing has ranking characteristics that diverge from Google on three concrete points. It gives more weight to domain age, rewarding old domains more noticeably. It values exact match keyword in the title tag with greater sensitivity, something Google has toned down over the last few algorithmic cycles. It incorporates social signals (shares and mentions on LinkedIn) into ranking in a way that Google abandoned long ago.
A site perfectly optimized for Google can turn out to be invisible for the same queries on Bing. And if ChatGPT browse mode queries Bing, that invisibility transfers directly into the AI answers.
In previous articles I talked about the backlink as a citation proxy and about the implicit reference weight. Both mechanisms operate downstream of retrieval: if you’re not in the index, those signals aren’t even evaluated. Bing indexing is the entry gate — without it, the whole conversation about authority and citation graph becomes academic.
If you migrated your site to a fresh domain and abandoned the historical one, Bing penalizes you more harshly than Google.
The Reverse Engineering I Did
I built a small experiment to see with my own eyes what kind of sources ChatGPT browse mode pulls when it answers. I picked a concrete case: a nursery of centuries-old olive trees and Mediterranean plants in Crotone, in Calabria, an industry where the typical client (landscape architect, owner of a prestigious villa, coastal resort) often asks ChatGPT for suggestions before contacting a supplier.
I ran 10 queries on ChatGPT with browsing enabled, varying the wording and the intent:
- “nurseries of centuries-old olive trees in Calabria”
- “where to buy a thousand-year-old olive tree in Italy”
- “suppliers of Mediterranean plants for luxury gardens”
- “ornamental olive tree nursery southern Italy”
- “centuries-old olive trees for Ionian coast resort”
- 5 similar variants with a geographic focus on Calabria, Puglia, Sicily and with the modifiers “price”, “transport”, “professional”
Out of 10 answers, 8 cited concrete sources with active links. Analyzing the 27 unique domains cited across all the answers, I lined up the data I cared about:
- 22 out of 27 were on the first page of Bing for the corresponding query (manually verified in incognito mode)
- 19 out of 27 had the nursery’s name or the main keyword in the title tag almost verbatim
- 24 out of 27 appeared to be registered or tracked in Bing Webmaster Tools (inferred from public sitemaps and response headers)
- only 3 out of 27 had a domain registered after 2020
It’s an indicative test, not a study with statistical significance. Small sample, a single vertical industry, a single AI engine. But the pattern is too clear-cut to be random: ChatGPT browse mode rewards old domains, with a surgical title, indexed on Bing. The overlap between the first page of Bing and ChatGPT citations exceeds 80%: it’s the signature of the retrieval layer.
The real analysis, the kind I’d do for a client before a major investment, requires professional tools and a sample of at least 200 queries across different industries. For a first internal audit, the signal is already clear enough.
Register your site on Bing Webmaster Tools and submit the sitemap.
The Test You Can Run in 15 Minutes
Before any investment, check three things in your industry. They are entry-level checks — a first step, not the full analysis — but they tell you right away whether the entry gate is open or closed.
One. Go to Bing Webmaster Tools and register your site. If you’ve never logged in, it’s the mandatory first step. Without this, Bing might still index you, but you lose all the diagnostic signals you need to understand where to act.
Two. Open Bing in incognito mode (not logged in, clean browser, preferably on a different connection) and search for the 5 main queries in your industry. Note where you rank. If you’re not on the first page for any of them, there’s a problem upstream of any AI conversation: you’re invisible to live retrieval before you even talk about ChatGPT.
Three. Open ChatGPT with browsing enabled (Plus or Team), ask the same 5 queries, and look at which sources it cites. Compare with the first page of Bing: the overlap should be high. If ChatGPT only cites competitors you don’t see on Bing, you’re out of the live retrieval game. If instead you see competitors that aren’t on the first page of Bing but still show up in the AI answers, you’ve found a pattern worth digging into: a signal of contextual authority that goes beyond pure ranking.
These three checks are a first step. They don’t yet tell you why one page gets cited more than another — that’s a later phase, tied to semantic structure and authority, which I covered in the articles on E-E-A-T for AI and Author Entity Recognition.
The Most Common Mistakes (I’m Telling You from Experience)
In the first audits on clients who want to show up in live ChatGPT answers, some patterns repeat with an almost tedious frequency.
Site not registered on Bing Webmaster Tools. About 7 out of 10 Italian SMBs have never done it. It’s free, it takes 10 minutes, and it often resolves half the indexing problems on its own.
Title tags optimized only for Google. Long titles, with the brand on the right and a generic keyword: they work decently on Google but get penalized by Bing, which prefers specific keyword + brand + geographic modifier (“Centuries-Old Olive Tree Nursery Crotone | [Brand]”).
Zero social signals. Bing incorporates LinkedIn into ranking more directly than Google. A B2B company without an up-to-date LinkedIn page is perceived by retrieval as less reliable, all else being equal in terms of on-site content.
New domain without authority building. If you migrated your site to a fresh domain and abandoned the historical one, Bing penalizes you more harshly than Google. The 301s don’t transfer 100% of the signal.
How to Act (Without Procrastinating)
The operational audit in 3 steps before investing:
- Register your site on Bing Webmaster Tools and submit the sitemap. Binary threshold: done / not done.
- Check your Bing ranking on the 5-10 main queries in your industry. If you’re off the first page on all of them, the problem is basic SEO, not advanced GEO.
- Compare with the 3-5 competitors ChatGPT cites when you run your industry queries with browsing enabled. What do they have that you don’t? Domain age, surgical title tag, LinkedIn presence, well-structured Organization schema.
The thread I keep throughout this whole series is the same: visibility in AI answers isn’t magic, it’s the result of concrete signals you can verify one by one. ChatGPT browse mode is a special case where that signal is called Bing, and it’s the link that connects an academic paper on RAG to your first SERP page in the province of Crotone.
Where the Series Continues
In the upcoming articles I tackle the other platforms: how Perplexity’s retrieval works (which uses its own indexes + Google), how Claude handles web sources, and why Gemini treats the web in yet another different way. Every platform has its own retrieval, and every retrieval wants different signals.
If you want to dig deeper into how AI engines identify entities and link them to your brand, start with Named Entity Recognition and with entering the Google Knowledge Graph — two pieces that work upstream of the entire platform discussion.
