What AI search engines and shopping agents actually cite: Azoma analysed 38 million citations
Last Updated:
Our study examined citation data across Azoma's commerce client dashboards, breaking results down by engine and by source category. The differences between engines are far larger than most brands assume.
How AI engines choose their sources
The five engines resolve a query in fundamentally different ways, and the citation data makes those differences visible.
ChatGPT answers from learned knowledge plus retrieval, invoking web search for queries that need current information. Google's AI Overviews runs against Google's live index, breaking a question into subtopics and searching each in parallel, then composing a summary with links. Alexa for Shopping and Walmart Sparky are closed retail assistants, answering inside a shopping context rather than the open web.
Those architectures produce measurably different citation profiles. One engine takes over half its citations from publishers. Another takes nearly 40% from retailers. A third takes nearly 40% from affiliate content.
brand.com throughout this piece means any manufacturer-operated domain — a brand's own site or a competitor's.
What do AI engines and shopping agents cite most?
| Category | ChatGPT | Google Gemini | Google AI Overviews | Alexa for Shopping | Walmart Sparky |
|---|---|---|---|---|---|
| Earned media | 22.5% | 51.9% | 7.6% | 48.0% | 16.6% |
| brand.com | 21.5% | 12.6% | 11.8% | 0.8% | 21.6% |
| Institutional | 37.2% | 11.2% | 3.6% | 4.3% | 10.1% |
| Retailer | 17.5% | 15.5% | 39.2% | 0.6% | 34.9% |
| Social / UGC | 0.5% | 3.0% | 12.2% | 0.6% | 6.8% |
| Affiliate | 0.0% | 0.0% | 0.2% | 39.6% | 1.8% |
| Other | 0.9% | 5.9% | 25.4% | 6.0% | 8.3% |
Read across any row and the disagreement is obvious. Earned media ranges from 7.6% of citations on Google AI Overviews to 51.9% on Google Gemini. Retailer content ranges from 0.6% on Alexa for Shopping to 39.2% on AI Overviews. Affiliate content is effectively zero on three engines and 39.6% on a fourth.
What sources does ChatGPT cite?
ChatGPT leans harder on regulators, clinical literature and government sources than any other engine measured:
Institutional and primary sources dominate at 37.2% of citations, reflecting a clear preference for regulatory bodies, clinical literature and government health sources
Earned media accounts for 22.5%
brand.com sits at 21.5%, close behind publisher content
Retailer sources contribute 17.5%
Social and UGC is effectively absent at 0.5% — a rounding error
Affiliate content registers 0.0%, with a further 0.9% unclassified
ChatGPT is also the least concentrated engine in the study. No single domain accounts for more than 2.4% of its citations, which makes it the hardest engine to influence through any single relationship.
| Category | Share of citations |
|---|---|
| Institutional | 37.2% |
| Earned media | 22.5% |
| brand.com | 21.5% |
| Retailer | 17.5% |
| Other | 0.9% |
| Social / UGC | 0.5% |
| Affiliate | 0.0% |
What sources does Google Gemini cite?
Gemini is the most publisher-led engine measured, with a markedly different profile from ChatGPT:
Earned media leads at 51.9% — more than half of everything Gemini cites is editorial content
Retailer sources account for 15.5%
brand.com contributes 12.6%
Institutional sources sit at 11.2%, a third of ChatGPT's weighting
Social and UGC accounts for 3.0%
Affiliate content registers 0.0%, with 5.9% unclassified
| Category | Share of citations |
|---|---|
| Earned media | 51.9% |
| Retailer | 15.5% |
| brand.com | 12.6% |
| Institutional | 11.2% |
| Other | 5.9% |
| Social / UGC | 3.0% |
| Affiliate | 0.0% |
What sources does Google AI Overviews cite?
AI Overviews is the most retail-led engine in the study, and the only one that cites social content at meaningful scale:
Retailer sources lead at 39.2%, roughly twice ChatGPT's weighting
Social and UGC contributes 12.2%, around twenty times ChatGPT's rate
brand.com accounts for 11.8%
Earned media drops to 7.6%, roughly a seventh of Gemini's weighting
Institutional sources sit at 3.6%
Affiliate registers 0.2%, with a further 25.4% unclassified
| Category | Share of citations |
|---|---|
| Retailer | 39.2% |
| Other | 25.4% |
| Social / UGC | 12.2% |
| brand.com | 11.8% |
| Earned media | 7.6% |
| Institutional | 3.6% |
| Affiliate | 0.2% |
What sources does Alexa for Shopping cite?
Amazon's shopping assistant is the outlier of the study, with a profile unlike any other engine measured:
Earned media leads at 48.0%
Affiliate content follows at 39.6% — by far the highest of any engine
Institutional sources account for 4.3%
Publisher and affiliate content together account for 87.6% of everything Alexa for Shopping cites, while brand-operated sites account for under 1%. It cites almost nothing from retail domains — including Amazon's own.
Alexa for Shopping cites no social platforms at all
It cited none of YouTube, Reddit, Instagram, LinkedIn or Wikipedia — not once, across three months and more than 250,000 citations in its heaviest month.
| Category | Share of citations |
|---|---|
| Earned media | 48.0% |
| Affiliate | 39.6% |
| Other | 6.0% |
| Institutional | 4.3% |
| brand.com | <1% |
| Retailer | <1% |
| Social / UGC | <1% |
What sources does Walmart Sparky cite?
Sparky shows the most balanced citation profile of the five, with no single category above 35%:
Retailer sources lead at 34.9%
brand.com accounts for 21.6% — the highest of any engine measured
Earned media contributes 16.6%
Institutional sources sit at 10.1%
Social and UGC accounts for 6.8%
Affiliate registers 1.8%, with 8.3% unclassified
Retailer, brand.com and earned media all carry meaningful weight here, which is what separates Sparky from the more lopsided profiles elsewhere in the study. Along with ChatGPT, it is one of only two engines where a brand's own website does substantial work, at over a fifth of citations.
| Category | Share of citations |
|---|---|
| Retailer | 34.9% |
| brand.com | 21.6% |
| Earned media | 16.6% |
| Institutional | 10.1% |
| Other | 8.3% |
| Social / UGC | 6.8% |
| Affiliate | 1.8% |
Walmart Sparky figures are August 2026; September data was not available at time of publication.
Key takeaways: building visibility across every engine
1. There is no single AI search strategy
Publisher relationships drive Gemini and Alexa for Shopping. Retailer content drives Google AI Overviews and Walmart Sparky. Institutional and first-party sources drive ChatGPT. These are separate programmes, not one programme with five outputs.
2. Owned content matters
brand.com carries over a fifth of citations on both Walmart Sparky and ChatGPT, at 21.6% and 21.5% — yet under 1% on Alexa for Shopping.
3. Affiliate content is a channel, not a side effect
It accounts for 39.6% of Alexa for Shopping citations and under 2% on every other engine. Brands with no view of which affiliate publishers cover their category are invisible to the agent answering on Amazon's product pages.
4. Social platform weighting is volatile and engine-specific
Social and UGC accounts for 12.2% of Google AI Overviews citations and 3.0% on Gemini, where it has been stable. On ChatGPT it currently sits at 0.5% — down from roughly a fifth of citations three months ago, following a change in how the engine retrieves sources. That kind of swing can reverse, so a snapshot is a poor basis for a budget decision.
The exception is Alexa for Shopping, which cited no social or reference platform at all in any month measured. That one looks structural rather than volatile.
5. Measure per engine, always
A blended "AI visibility" number averages five profiles that have almost nothing in common, and tells you nothing actionable about any of them. The volatility above makes the case sharper: a single figure that moves twenty points on one engine while four others hold steady is worse than no figure at all.
👉 Book a call with our team to see how your brand is being cited across ChatGPT, Gemini, Google AI Overviews, Alexa for Shopping and Walmart Sparky.
Methodoloy
Figures are drawn from Azoma platform citation data across commerce client dashboards. Category shares are calculated across each engine's top-cited domains, with every domain classified by source type.
brand.com combines brand-operated and competitor-operated manufacturer domains.
Affiliate content is reported as a separate category in this analysis. Azoma's Q2 2026 figures grouped affiliate sources within earned media; on that basis the combined figure for Alexa for Shopping is 87.6%, consistent with the 86.5% previously published.
Naming. Alexa for Shopping is Amazon's shopping assistant, previously branded Rufus. Third-party research may use the earlier name.
Prompt set. Azoma's prompt set is weighted toward commerce and category queries — the questions buyers actually ask before a purchase. This matters when comparing against citation studies built on general or informational queries, which produce materially different results.
Figures are September 2026 except Walmart Sparky, which is August 2026.
Article Author: Max Sinclair