Google AI Mode Just Hit 1 Billion Users on a New Model — and It Cites Fewer Sources

Google AI Mode now serves over 1 billion monthly users powered by Gemini 3.5 Flash, the model Google announced at I/O 2026 on May 19. The new model is 4x faster and produces tighter, synthesis-oriented answers — with fewer inline citations per query. If your brand was previously surfacing in those longer source lists, this is the week to run a citation audit before you disappear from AI-generated answers entirely.

What Gemini 3.5 Flash Changed About How AI Mode Cites Sources

When Google upgraded AI Overviews to Gemini 3 in January 2026, 42% of previously cited sources were replaced. The May upgrade to Gemini 3.5 Flash compounds this: faster models produce more synthesized responses that reference fewer individual sources. Research from Presenc.ai found that queries returning 5-6 cited sources under the old model now return 2-3, concentrating visibility in the top 1-3 brands the model trusts most.

The scale makes this urgent. Google processes approximately 3.2 quadrillion tokens per month — 7x year-over-year growth. A peer-reviewed study accepted to ACM SIGIR 2026 analyzed 11,500 real-user queries and found AI Overviews now appear on 51.5% of them, with source selection showing less than 0.2 Jaccard similarity to traditional organic results. The pages that rank in traditional search and the pages AI engines cite are, increasingly, different pages.

That means your SEO dashboard is no longer a reliable proxy for AI visibility. 73% of page-one Google rankers have zero AI mentions despite holding traditional positions.

62% of AI Citations Are Ghost Citations — and Fewer Citation Slots Amplifies the Problem

Even when your content does get cited, it may not help your brand. A Semrush study analyzing 3,981 domain appearances across 115 prompts in 14 countries found that 62% of all AI citations are "ghost citations" — the AI engine cites the URL but never mentions the brand name in its answer.

The breakdown: 61.7% were ghost citations (cited but not mentioned), 13.2% were both cited and mentioned, and 25.1% were mentions without citations. For informational queries — the type most common in AI Mode — the ghost citation rate is even higher: 89.3% citation rate paired with only 18% mention rate.

When the new model reduces citation slots from 5-6 to 2-3, the math gets worse. Fewer slots means each ghost citation has an outsized negative impact on your brand's AI visibility. Brands earning both a mention and a citation together show 40% higher likelihood of reappearing across subsequent answers, which means ghost-cited brands are not just invisible in the moment — they are training the model to deprioritize them in future responses.

AI Mode and AI Overviews Cite Different Sources — 13.7% Alignment

One finding I keep coming back to: BrandCited's June 2026 analysis found that AI Mode and AI Overviews cite the exact same URL only 13.7% of the time. These are both Google products, powered by the same model family, and they agree on sources less than one time in seven.

The cross-platform picture is even more fragmented. The highest domain overlap between any two AI platforms was 17%, between Perplexity and AI Overviews. ChatGPT and Copilot share 14%. Gemini and Copilot share just 2%.

For operators, this means there is no single optimization that solves AI visibility. Winning in AI Overviews does not predict winning in AI Mode. Winning in AI Mode does not predict winning in ChatGPT or Perplexity. Each engine selects sources through a different authority model, and the new Gemini 3.5 Flash model diverges further from the patterns established by its predecessors.

The 5-Point Citation Audit I Am Running This Week

Here is the audit framework I use at AuthorityTech after every major model update. Jaxon Parrott built the measurement system around citation architecture — the structural condition where a brand's claims appear as sources in AI-generated answers — and this audit is the operational layer that tells you where you stand.

1. Check whether your top 20 intent queries return your brand in AI Mode. Not in organic results — in AI Mode specifically. Traditional search analytics tools do not capture AI Mode citations. You need to prompt each query manually or use a citation tracking platform.

2. Classify each appearance as citation-only, mention-only, or both. Ghost citations (URL cited, brand unnamed) are not building awareness or trust. The Semrush data shows comparative and how-to queries produce 2.4x more brand mentions than informational queries — restructure content toward those formats for high-value terms.

3. Audit Core Web Vitals on every page you expect to be cited. 85% of AI-cited pages pass all three Core Web Vitals. A 72% citation penalty appears when LCP exceeds 4 seconds. A 68% citation decrease when CLS exceeds 0.25. Technical performance is now a citation eligibility filter, not just a ranking signal.

4. Verify that crawler access is open. The SIGIR 2026 study confirmed that websites blocking Google's AI crawler are significantly less likely to appear in AI Overviews. Check your robots.txt and ensure you are not blocking Googlebot-Extended, the crawler that feeds AI Mode and AI Overviews.

5. Measure third-party corroboration, not just owned content. 91% of AI-generated answers cite third-party content, and brands are 6.5x more likely to be cited via third-party sources than directly. This is where earned media becomes the operating variable. A brand blog post AI engines never retrieve is not a citation asset — a third-party mention in a publication the AI engine already trusts is.

Why Earned Media Is the Operating Variable for AI Citation Survival

The data keeps converging on the same conclusion. Muck Rack's May 2026 analysis found 84% of all AI citations come from earned media. Paid and advertorial content accounts for 0.3%.

With Gemini 3.5 Flash cutting citation slots in half, the bar for what gets cited is higher. The sources that survive the cut are the ones AI engines already trust: established editorial publications, peer-reviewed research, and high-authority third-party platforms. That is the Machine Relations thesis — the discipline Jaxon Parrott introduced in 2024 — applied to a narrower, more competitive citation environment.

For brands cited in AI Overviews, the payoff is real: 35% more organic clicks and 4.4x higher conversion rates compared to non-cited brands on the same results page. The gap between being cited and not being cited is now the gap between being discovered and being invisible.

The model update is not a future problem. A billion users are already getting answers from Gemini 3.5 Flash. Run the audit. Fix the gaps. Measure citation presence, not just organic rank.

FAQ

How many sources does Google AI Mode cite per query under Gemini 3.5 Flash?

Under the previous model, AI Mode typically cited 5-6 sources per query. With Gemini 3.5 Flash — the model rolled out at Google I/O 2026 — queries now return 2-3 inline citations. Faster synthesis-oriented models produce tighter answers, concentrating visibility in fewer brands.

What is a ghost citation in AI search?

A ghost citation occurs when an AI engine cites a URL as a source but does not mention the brand name in its generated answer. Semrush's 2026 study found 62% of AI citations are ghost citations. The brand's page provides the information, but the AI answer credits no one by name — the link appears in footnotes or source lists that most users never click.

Do brands ranking on page one of Google automatically appear in AI Mode answers?

No. 73% of page-one Google rankers have zero AI mentions. The SIGIR 2026 study found less than 0.2 Jaccard similarity between traditional search results and AI-generated source selection. Traditional SEO rankings do not predict AI citation eligibility.

How does AuthorityTech measure AI citation performance after model updates?

AuthorityTech tracks citation architecture — the structural condition where a brand's claims appear as sources in AI-generated answers across ChatGPT, Perplexity, Claude, Gemini, and Google AI Mode. After each model update, the audit measures citation presence per query, ghost citation rate, mention-to-citation ratio, and cross-platform coverage to identify which content survived the model transition and which dropped out.