ChatGPT and Google's AI Overviews cite different sources even when they get the same question: AI Overviews lean heavily on YouTube and on company websites, ChatGPT on the public bodies that set the rules and on big brands. In our test on 28 October 2026 across 24 Italian searches in 6 sectors, YouTube appeared in 16 of 18 AI Overviews and in none of the 24 ChatGPT answers, while ChatGPT cited a public body in 13 answers out of 24 and AI Overviews in 3 out of 18.
In practice: if you want to be cited by both, you are working on two almost separate lists of sources. On the same search, the two engines shared at least one domain only 8 times out of 18.
We see the same logic in classic search, where sources are won one page at a time. Macropix, which makes custom LED screens, reached first position in Italy for "ledwall" (4,400 searches a month) with one page for each product line (the full story is in our piece on manufacturing SEO).
Sunscape, a solar and agrivoltaic developer, went from 329 to 5,809 organic impressions a month in six months from March 2026 (we cover it in how long SEO takes).
And AI answers?
There the question changes. Knowing whether your site appears is not enough: you need to know who gets cited instead of you, and by which engine.
The two mistakes we see most often:
- assuming the overall rankings of cited sites apply to your sector (across almost 30 million Italian keywords Wikipedia is the second source in AI Overviews; in our 24 business searches it appeared twice);
- tracking one engine only and assuming the other cites the same pages.
Below you'll find the test on 24 Italian searches with a table by sector, the three differences between ChatGPT's sources and AI Overviews' sources, who gets cited sector by sector, and a five-step method to check the sources in your own market.
The test: 24 Italian searches, 6 sectors, two engines (on a single day)
On 28 October 2026 we ran 24 searches in Italian, 4 per sector: energy, manufacturing, business software, tax and business finance, health, digital marketing. They are informational questions real buyers ask, such as "difference between CRM and ERP" or "how factoring works".
For each one we used DataForSEO to read the Google Italy results page (desktop, Italian) and ChatGPT's answer with web search on, located in Italy. Then we counted cited pages and domains.
Google showed an AI Overview on 22 of 24 searches. In 4 cases the box loaded asynchronously and we couldn't read its sources, so the AI Overview comparisons are based on 18 answers. The two searches without one were both technical: "heat pump or condensing boiler" and "laser cutting sheet metal maximum thickness".
| Sector (4 searches) | AI Overviews shown | Pages cited per AI Overview | AI Overviews citing YouTube | Pages cited per ChatGPT answer | ChatGPT sources from public bodies |
|---|---|---|---|---|---|
| Energy | 3 of 4 (2 readable) | 7 | 2 of 2 | 6.3 | 6 domains of 15 |
| Manufacturing | 3 of 4 (1 readable) | 9 | 1 of 1 | 5.5 | 0 of 17 |
| Business software | 4 of 4 | 9 | 4 of 4 | 5.8 | 7 of 18 |
| Tax and finance | 4 of 4 (3 readable) | 8.3 | 3 of 3 | 4.8 | 6 of 15 |
| Health | 4 of 4 | 7 | 3 of 4 | 3.5 | 11 of 14 |
| Digital marketing | 4 of 4 | 9 | 3 of 4 | 4 | 0 of 13 |
| Total | 22 of 24 (18 readable) | 8.2 | 16 of 18 | 5 | 30 of 92 (33%) |
One warning before the numbers. Answers change from one call to the next: repeating the same search a few minutes apart, we saw organic results in a different order. Treat this as a snapshot, not a stable ranking (we tested variability separately in how accurate AI visibility data is).
AI Overviews cite YouTube almost every time (16 answers out of 18)
The most cited source in Italian AI Overviews is YouTube, present in 16 answers out of 18. No other domain appears in more than 3.
That includes Shorts in Portuguese, French or English under searches made in Italian. We found the same pattern in our test on YouTube videos in AI Overviews, where ChatGPT cited one video out of 152 sources.
Other social platforms barely feature: Facebook in 3 answers, Instagram and TikTok in one each. Reddit: zero. Wikipedia: twice.
Didn't the big studies say Reddit and Wikipedia?
Yes, for English-language search. Ahrefs' study of 76.7 million AI Overviews (June 2025 data) puts YouTube at 9.5%, Wikipedia at 8.4% and Reddit at 7.4% of mentions, and Wikipedia at 16.3% for ChatGPT. The Pew Research Center, looking at 68,879 searches by 900 US adults in March 2025, found Wikipedia, YouTube and Reddit together made up 15% of sources.
Large Italian datasets also favour YouTube and Wikipedia. SEOZoom's study of almost 30 million Italian keywords (December 2025) ranks YouTube first with 875,914 mentions, Wikipedia second with 864,517 and the fitness site my-personaltrainer.it third with 403,934.
In our 24 business-topic searches, though, Wikipedia stopped at 2 citations and Reddit at zero, while my-personaltrainer.it appeared 3 times, all on the health questions. Their place went to company and practitioner websites: of 136 domains cited by AI Overviews, 111 were different.
ChatGPT goes to whoever writes the rules (public bodies in 13 answers out of 24)
When a topic has an authority that regulates it, ChatGPT cites that authority. In the test 30 of the 92 domains it cited (33%) were public bodies, in 13 answers out of 24. In AI Overviews: 3 out of 136.
- On "what is the NIS2 directive" ChatGPT cited Italy's cybersecurity agency (ACN), the Ministry of Enterprise, EUR-Lex and Normattiva; the AI Overview cited 7 security and telecoms companies, plus YouTube.
- On "Transizione 5.0", an Italian tax incentive, ChatGPT cited only the energy services operator (GSE) and the ministry.
- On "what is SEO" ChatGPT cited two pages, both from Google; the AI Overview cited 7 pages, none from Google: three Italian agencies, a telecoms provider, a Swiss host and YouTube.
Health had a surprise. In 3 answers out of 4 ChatGPT cited at least one foreign source (the NHS, NICE, Mayo Clinic, MedlinePlus, the European Heart Journal) for a question asked in Italian, alongside the Italian health ministry and the National Institute of Health. AI Overviews on the same searches stayed mostly with Italian hospitals, clinics, online pharmacies and health product brands.
ChatGPT's second favourite group is the big brands of the sector: Microsoft in 3 answers, SAP in 2, IBM and Salesforce in the CRM vs ERP answer.
Two engines, two lists (11 shared domains out of 136)
On the same search, ChatGPT and AI Overviews shared 11 domains in total. On 10 searches out of 18 they shared none.
The overlaps are nearly always the "owner" of the topic: the Italian Revenue Agency on e-invoicing, GSE on renewable energy communities, the National Institute of Health on flu, Google on Google Ads, SAP and IBM on business software, two Italian banks on factoring and leasing.
The number of sources differs too. AI Overviews cited 8.2 pages on average, ChatGPT 5. In the US, Pew found 88% of Google's AI summaries cite three or more sources; in our Italian sample every readable AI Overview cited at least 5.
Across the whole sample, of the 111 domains cited by AI Overviews and the 78 cited by ChatGPT only 15 appear on both lists. Track one engine only and you see less than half of the picture.
If you only check Google, you miss the sources ChatGPT reads (and the other way round).
In the test the two engines shared 11 domains out of 136. We ask your customers' questions to ChatGPT and Google's AI, open every source cited in your sector and tell you which pages, yours or third parties', are worth working on.
Book a strategy callWho gets cited in your sector (and where small companies get in)
The sector changes the type of source more than the engine does. Where no public body sets the rules, the sources are the websites of the companies doing the work, small ones included.
Manufacturing: workshops and machine makers, no public bodies
In ChatGPT's 4 answers on milling, laser cutting, 3D printing and CNC suppliers, none of the 17 domains cited was a public body or a publisher: machine shops, machine builders, 3D printer makers such as Formlabs and EOS, and one trade portal. In the sector's only readable AI Overview, 7 of 8 domains were machining companies, plus YouTube. It matches what we saw in why ChatGPT recommends your competitors: whoever explains the process well on their own site becomes the source.
Tax and finance: banks explain their own products
Of ChatGPT's 15 sources in this sector, 8 were banks or factoring companies (Banca Ifis, Intesa Sanpaolo, UniCredit, Banco BPM, Banco di Sardegna), all on the leasing and factoring questions. On the tax incentive and company-type questions, only GSE, the ministry, the chamber of commerce and the notaries' council.
Software and marketing: big brands on one side, agencies on the other
On ERP and CRM ChatGPT picked Microsoft, SAP, IBM and Salesforce; AI Overviews mixed TeamSystem, Oracle, SAP, smaller software houses, Wikipedia, Facebook, Instagram and TikTok. In marketing the split is even sharper: on "how Google Ads works" ChatGPT cited only Google pages, while the AI Overview cited 9 domains: three of Google's, plus a popular Italian tech blog, an online university, consultants' sites and YouTube.
In energy, on "how solar with battery storage works", the AI Overview cited 5 installers and sellers (Enpal, Otovo and three smaller firms), plus a technical publisher and YouTube; ChatGPT cited Enel, Eni Plenitude and two other sites. For a solar business like Sunscape, a technical guide on its own site can get into AI Overviews, while ChatGPT, on that question, preferred the large energy suppliers.
Cited sources are not Google's first page (19 out of 120)
Ranking on page one helps, but in our sample it doesn't explain citations. Excluding YouTube, only 19 of the 120 domains cited by AI Overviews were also among the organic results on the same page (7 to 9 per search). For ChatGPT, 10 of 92.
Google explains why. Its Search Central documentation (updated 10 December 2025) says AI Overviews and AI Mode may use "query fan-out": they run several related searches across subtopics and sources and pick up more supporting pages from there. We cover it in query fan-out. Ahrefs' March 2026 study of 863,000 keywords, an international sample, found only 38% of citations came from the top 10 (the numbers are in AI Overview ranking factors). On our Italian sample the share is lower still.
What it changes for your visibility in AI answers (no promises)
Nobody can guarantee a citation: answers change between calls, and Google says a page meeting every requirement may still not be shown. What the data does show is where it makes sense to work for each engine.
- For AI Overviews, pages that explain a process or a difference matter, and so do videos: YouTube was in 16 answers out of 18.
- For ChatGPT, being consistent with official sources matters where the topic is regulated. If your page contradicts the regulator, the regulator gets cited.
- For both, a page dedicated to the exact question, with numbers and conditions in plain words. It is also where the AI citations and brand mentions we track for clients converge.
How to find which sources get cited in your sector (five steps)
1. Start from 10-20 real questions
Take them from Search Console, quote requests and what your sales team gets asked. How to choose them is in our guide to prompt tracking.
2. Read both engines' sources, not just the text
Open the links under each answer. In our test the same page often appeared several times with slightly different URLs, so count pages, not links.
3. Sort sources into four groups
Public bodies, companies in the sector (competitors included), video and social, publishers. The biggest group tells you where to act: your own pages if companies dominate, video if YouTube does, consistency with official sources if public bodies do.
4. Check the engines can read you
For Google, the page must be indexed and eligible for a snippet, with no extra technical requirements. For ChatGPT, OpenAI's documentation says OAI-SearchBot is used to surface websites in ChatGPT search, and sites that block it are left out of search answers. Details are in our guide to AI crawlers.
5. Repeat monthly and watch the traffic
One reading is a snapshot; monthly readings show who comes in and who drops out. Also, 115 of the 148 links ChatGPT cited in the test carried utm_source=chatgpt.com, which is what lets you see those visits in Google Analytics (see ChatGPT traffic in Google Analytics 4).
Beyond 20 questions, doing it by hand gets heavy. That's why we built ScryRank, the monitoring tool developed by Visilay: it measures how ChatGPT, Claude, Gemini, Perplexity and AI Overviews talk about your brand and shows which domains and pages the AI treats as authoritative. If you'd rather we do the reading and the work, there's our AI Monitoring service, and GEO for the pages themselves. The full measurement framework is in our guide to AI visibility.
FAQ: sources cited by ChatGPT and AI Overviews
It depends: it only shows links when it searches the web. In our test we turned search on for all 24 questions and all 24 answers had sources, 5 pages on average. When it answers from the model's memory there are no links, and your page cannot be cited.
It depends on the question. Across almost 30 million Italian keywords Wikipedia is the second source in AI Overviews according to SEOZoom, but in our 24 business-topic searches it was cited only twice and Reddit never. Their place went to YouTube, company websites and, for ChatGPT, public bodies.
No. Google Search Central says there are no additional requirements and no special structured data: the page must be indexed and eligible for a snippet. Even then a citation is not guaranteed.
No. According to OpenAI, sites that block OAI-SearchBot are left out of ChatGPT search answers and can at most appear as navigational links. GPTBot is a different setting: it covers model training.
Summary
YouTube in 16 AI Overviews out of 18, in no ChatGPT answer.
ChatGPT cites a public body in 13 answers out of 24.
11 domains shared by the two engines, out of 136.
Only 19 cited domains out of 120 were in the organic results.
Where there's no regulator, the sources are the companies in the sector.
One last thing. For twenty years the question was "where do I rank?". With AI answers there's one before it: which sites does the engine read when someone asks about my line of work? In our sample, in manufacturing, they were workshops and machine makers just like your competitors. So the source missing from that answer may be a page you haven't written yet.
Speak soon!
