Do AI Answers That Cite External Pages Also Name Brands?

Key findings

1. 15.3% vs 45.2%: from June to September, 15.3% of ChatGPT answers had "external sources only, no brand name"; 45.2% of Gemini answers had "brand names only, no external source."
2. 84.2% vs 95.6%: among ChatGPT answers with external sources, 84.2% named a brand; among answers without external sources, the share was actually higher, at 95.6%. Including external links does not make brand names appear more often.
3. 52.0% → 4.3%: among Google AI Mode answers that named a brand, the share with no external source fell from 52.0% in June to 4.3% in August. Over the same period, answers with external sources rose from 49.2% to 95.2%, while the share naming a brand stayed between 90% and 92%.
4. 25.8% vs 7.3%: in September, 25.8% of ChatGPT answers to awareness questions had external sources only and no brand name; for decision questions, it was only 7.3%.

When an article shows up among the reference links in an AI answer, teams often log it as a win. Then they read the full answer and find they are not on the recommendation list. The opposite also happens: AI names the brand often, but monitoring tools can't find a matching source page.

The Dageno Team analyzed 10,045,768 answers in the Dageno AI answer database from June to September 2026, across 5 AI platforms: ChatGPT, Google AI Mode, Google AI Overviews, Gemini and Copilot. For each answer, we recorded two things: whether it had external sources, and whether it named a brand.

01 | In one answer, how many combinations of external sources and brand names are there?

Four, and their distribution varies widely by platform.

June to September combined, out of every 100 answers:

Table: each row adds up to about 100; rounding may cause small differences.

• ChatGPT and Google AI Overviews include external sources in almost every answer. The difference is whether a brand gets named: 15.3% of ChatGPT answers included sources but named no brand at all.
• Gemini and Google AI Mode often name brands without giving external sources. 45.2% of Gemini answers had brand names only; for Google AI Mode, it was 35.3%.

So if you only track "whether our pages get cited," you miss a large number of answers that name brands without links. If you only track "whether our brand gets mentioned," you count citations that explain without naming anyone as failures.

02 | Do answers with external sources name brands more often?

No. The two are barely related.

June to September combined, the share of answers naming a brand, with and without external sources:

• ChatGPT: 84.2% with sources, 95.6% without.
• Google AI Mode: 91.1% vs 90.3%.
• Google AI Overviews: 91.9% vs 92.7%.
• Gemini: 91.2% vs 85.4%.
• Copilot: 89.5% vs 93.0%.

On 3 of the 5 platforms, answers without sources actually named brands more often, and the gap was largest on ChatGPT. Note that ChatGPT and Google AI Overviews had very few answers without sources, only about 124,000 and 24,000 respectively, so these two figures only show direction.

This means getting AI to cite your pages and getting AI to say your brand name are two different outcomes. An answer can cite an industry explainer and name no brand at all; it can also give no external links and simply list a few brands.

03 | Why does the share of "brand named, no external source" differ so much by platform?

It mainly depends on whether the platform gave external links that month, not much on the brand names themselves.

Among answers that named a brand, the share with no external source:

Table: the denominator is answers that named a brand in that month.

Put it next to the "share of answers with external sources," and the pattern is clear:

• Google AI Mode: in June and July, only about half of answers had external sources (49.2% and 50.2%); in August, it rose to 95.2%. Over the same period, "brand named, no external source" fell from about 50% to 4.3%. The share of answers naming a brand stayed between 90.4% and 91.6% and did not move with it.
• ChatGPT: answers with external sources fell from 98.8% in June to 90.0% in September, and "brand named, no external source" rose from 1.3% to 11.6%. The share naming a brand was 83.4%, 86.1%, 83.9% and 84.0%; June and September differed by only 0.6 percentage points.
• Gemini: each month, only 45.7% to 50.2% of answers had external sources, and "brand named, no external source" stayed between 48% and 54%.

So if external pages citing you dropped over some period, that alone doesn't show your brand is doing worse in AI; the platform may simply have given fewer links while still naming your brand. Likewise, more links don't mean your brand is chosen more often.

The questions and collection volume differ by month, so not all of these changes come from the platforms, and we do not attribute them to any single platform update.

04 | Which questions most often have "sources, but no brand"?

Awareness questions, meaning questions where people "want to understand a concept first."

In September, on every platform, the share of answers with "external sources only, no brand name" was highest for awareness questions and lowest for decision questions:

• ChatGPT: awareness 25.8%, consideration 11.3%, decision 7.3%.
• Copilot: 16.4%, 5.8%, 3.9%.
• Google AI Overviews: 16.1%, 7.1%, 5.6%.
• Google AI Mode: 10.4%, 4.2%, 3.5%.
• Gemini: 7.0%, 2.7%, 2.2%.

This shows that answers explaining a concept often need sources but no brand names. If someone asks "what's the difference between web scraping and web crawling," an accurate explanation is enough; there's no need to recommend a particular service. When a knowledge article gets cited without bringing a brand along, that isn't necessarily a failure. It may simply be doing its job of explaining.

What's usually missing is the step from knowledge to choice: why this feature matters to which users, which products have it, what it costs, and when it doesn't apply.

05 | Within the same industry, do platforms still differ?

Yes. Even in the same industry, the gaps between platforms are large.

In September, among answers that named a brand, the share with no external source:

Table: for web scraping APIs, Google AI Overviews was 7.0% and Copilot 19.4%. The other industry question sets did not cover Gemini.

Gemini was highest in both industries. That matches how often it gives external links: in September, only 39.3% of Gemini answers to charging and power bank questions had external sources, and only 32.9% for web scraping API questions. Meanwhile, the share of these industry answers naming a brand was very high; for Gemini it was above 98% in both industries.

So brands can't judge every platform by a single "external sources" target. A brand that Gemini names often should first check whether the product facts and use cases AI states are correct. A brand that ChatGPT cites often but rarely names should check whether its own pages give reasons to choose it.

The industry question sets differ in questions and regional mix, so this table is not a comparison of platform capabilities.

06 | If a source and a brand name appear together, is it the same brand?

Not necessarily. This article counts "any source" and "any brand": if AI cites site A and names brand B, that still counts as "both."

To know whether your own pages bring along your own brand name, you need to match answers one by one: whose pages did this answer cite, whom did it name, which product did it pick, and why.

Research has already done this kind of matching. Semrush and Kevin Indig's "ghost citations" study, published in June 2026 (Why 62% of AI citations don't lead to brand mentions), recorded 3,981 brand web page appearances across 4 AI platforms and 115 questions. It found that in 61.7% of cases, "the page was listed as a source, but the answer text didn't mention the brand." The platform differences they saw point the same way as ours: Gemini names more and links less; ChatGPT links more and names less. They count by "brand × page," while we count by answer, so the two sets of numbers can't be compared directly.

For a brand, matching answers one by one can follow 4 layers:

1. Who owns the page: is the cited page yours, a competitor's, or an unrelated third party's?
2. What gets named: the company name, a product line, or a model the user can actually buy?
3. Why it was chosen: what conditions and reasons does the answer give, and does it mention trade-offs or cases where it doesn't fit?
4. Where to go next: can the user check those conditions and find the matching product page or contact point?

These 4 layers can't guarantee a recommendation, but they help a team find which layer the gap is in.

07 | What should marketing teams do next?

1. Split "cited" and "named" into two metrics. One going up doesn't mean the other does. Report them separately in monthly reports, and don't use one in place of the other.
2. Look at each platform separately. ChatGPT and Google AI Overviews include external sources in almost every answer, so focus on whether they name you. Gemini and Google AI Mode often name brands without links, so focus on whether what they say when naming you is correct.
3. When sources drop, first check whether the platform's link rate changed. In August, Google AI Mode answers with external sources went from about half to 95.2%; in September, ChatGPT fell to 90.0%. Neither had anything to do with brand names. Rule out this kind of shift before deciding whether your content has stopped working.
4. Set different goals for knowledge content and choice content. Awareness questions often get citations without names, so the goal of a knowledge article is to explain accurately. If you want to be chosen, you need separate pages that cover fit, comparisons and limits, with internal links connecting the two kinds of pages.
5. Check "who was cited, who was named" answer by answer. Spot-check the questions you care about using the 4 layers above, and find whether the gap is in page ownership, product names, reasons to choose, or next steps.
6. Retest with the same set of questions. Fix the questions, platforms and time window, and record "cited," "named" and "reasons for being chosen" separately. Don't look at a single total.

Methodology

The data in this article comes from the Dageno AI answer database. From June to September 2026, we analyzed 10,045,768 answers from ChatGPT, Google AI Mode, Google AI Overviews, Gemini and Copilot. The questions come from Dageno's historical monitoring and cover different countries, languages and industries. It is a fixed question set, not a random sample of user queries.

"External sources" means the answer's sources still include at least one link after identified brand websites are excluded; this is not the same as all links, and the remaining sites are not necessarily independent of brands. "Brand names" are detected automatically and may include tools, institutions and other entities. For the 4 combinations, the denominator is all answers; for "brand named, no external source," the denominator is answers that named a brand. The buying-stage comparison uses September only, because over 95.7% of that month's answers carry a stage label. The industry question sets total 450,570 answers and cover different months and platforms.

This article measures whether two things appear together in the same answer. It does not match any brand to its own pages, and it does not measure the effect of citations on recommendations or purchases.

References

1. Semrush, "Why 62% of AI citations don't lead to brand mentions [Study]", June 9, 2026.