Google’s AI Cites Your Website Without Naming You 43% of the Time

GEOforge Research · Citation Data

When Google’s AI answers cite a brand’s own website, they leave the brand out of the text about four times in ten. ChatGPT almost never does. A citation is not a mention.

By the GEOforge TeamGEOforge6 min read

Many AI visibility reports count a brand as “visible” when its website appears in an answer’s source list. We tested how often that same answer actually names the brand. On ChatGPT it nearly always does. On Google it often does not.

What is the short answer?

• On Google, 43% of the answers that cited a brand’s own website did not name the brand in the text. That is 20,914 of 48,964 answers from Google AI Mode and AI Overviews, from 13 July to 4 October 2026.
• On ChatGPT it was 3%. 224 of 8,496 answers. When ChatGPT cites your site, it almost always talks about you.
• It holds brand by brand: 15 of 15. In every brand with at least 30 site-citing answers on both engines, Google left the brand out more often than ChatGPT.
• The size of the gap varies a lot between brands. Google’s share ran from under 20% in 5 brands to above 90% in 2 brands, so treat the pooled figure as a typical case, not a promise.
• It has been stable since mid-July. It was a little lower in June. Since then weekly figures move around, but not in one direction, and the 24 August change in Google’s source lists did not shift it.

What did we measure?

GEOforge runs each brand’s tracked prompts against ChatGPT, Google AI Mode and Google AI Overviews and stores two things for every answer: the list of sources the engine attached, and a grade of the answer text. For this piece we kept 271,841 answers that came with a source list, from 25 brands that have a website on file, collected 13 July to 23 August and 31 August to 4 October 2026. (We skipped the week of 24 August, when Google’s source lists changed. More on that below.)

We asked two separate questions of each answer. First: does any cited source sit on the brand’s own website? Second: does the answer text name the brand? “Named” follows our grading rule: the brand appears in the prose as a specific organization, product or person tied to it. A source link on its own does not count. An answer that cites your site but does not name you is what we call cited but not named.

How often is a brand cited but not named?

On Google, 43% of the time. Combining AI Mode and AI Overviews, 20,914 of 48,964 answers that cited a brand’s own website did not name the brand. On ChatGPT it was 224 of 8,496, or 3%.

Google was not always this way in our data. In June (1 June to 12 July) the share was 37% in AI Mode and 31% in AI Overviews. Across 13 July to 23 August it was 43% and 42%, and it has stayed there since. ChatGPT stayed between 1% and 5% in every week.

Is this just one or two brands?

It is concentrated, so we checked. The two brands with the most such answers account for 55% of Google’s cases. Without them, Google’s share is 30% rather than 43%: lower, but still more than 1 in 4 and still far from ChatGPT’s 3%.

Brand by brand, comparing only brands with at least 30 answers citing their site on both engines, Google left the brand out more often than ChatGPT in 15 of 15. The typical brand sees 38% on Google and 1% on ChatGPT (medians). The spread is wide, though, and two brands sit near the top of it for reasons we have not investigated.

Did Google’s late-August change cause this?

No, as far as we can tell. In the week of 24 August the number of sources Google AI Mode attached to each answer in our data roughly halved, and we cannot tell from our side whether Google or our data provider changed. If the shorter lists were driving this result, the share would have moved with them. It did not: Google was at 42.7% before and 42.8% after. ChatGPT, whose source lists did not change, works as the control.

What about brands that are named but not cited?

The reverse is much more even. Among Google answers that named the brand, 22% did not cite the brand’s own site. For ChatGPT it was 17%. So the engines differ mainly in one direction: Google will lean on your content without crediting you in the text, while ChatGPT mostly cites the sites it talks about.

What should you do with this?

• Do not read a citation as a mention. If your reporting treats “site in the source list” as visibility, then on Google roughly 4 in 10 of those wins are answers where the reader never sees your name.
• Track the two separately, by engine. On ChatGPT the two move together, so one number can stand in for both. On Google they do not.
• On Google, make the page say who you are. We have not tested what makes Google name a brand. Stating the organization and what it offers plainly in the content Google is already citing is a reasonable, cheap experiment.
• Treat this as “what the answers show”, not “why”. Our data cannot tell you whether an unnamed citation sends traffic or builds recall. That is a separate question.

How did we measure this?

Data. 271,841 answers with source lists (209,902 from Google AI Mode and AI Overviews, the rest from ChatGPT) across 25 brands, 13 July to 23 August and 31 August to 4 October 2026. Only brands with a website on file are included. The week of 24 August is excluded from totals because collection volume and source-list length changed during it.

Classification. An answer “cites the brand’s site” when any source URL is on the brand’s own domain or a subdomain of it. An answer “names the brand” when our automated grader, which reads the answer text, judges that the brand is referred to as a specific organization in the prose. A source link alone does not count.

Checks. We split the data by week and before/after, compared brand by brand, removed the two largest brands, ran each engine separately, checked that the collector and grader code did not change, and compared against a June baseline. As a check on the grader, we took two weeks of Google answers from the seven brands with the most such cases, removed the source links from the answer text, and searched what was left for the brand’s name. For five of the seven, the name never appeared in any answer the grader had marked as not naming the brand. The other two have names that are also ordinary words or phrases, so a plain text search also catches everyday uses of those words (in 2% and 30% of their answers). The grader’s verdict holds up: in these answers the brand’s site is a source, and the brand is not in the prose.

Limitations. The prompts are the ones our clients track, so they lean B2B and are not a random sample of searches. The pooled figure is dominated by a few brands; the brand-level spread is wide. We cannot see why Google behaves differently from ChatGPT here. Every figure is aggregated and no brand or prompt is identified. See our earlier work on social platforms in AI citations and LinkedIn posts and articles for how the engines differ in other ways.

Sources & method. All figures come from the GEOforge production citation database: 271,841 AI answers with source lists, 25 brands, 13 July to 23 August and 31 August to 4 October 2026; June baseline 1 June to 12 July 2026. Brand matching uses each brand’s website on file.