Who gets cited when AI helps people buy?

Review and editorial websites had the largest share.

164 of 277 link occurrences—59.2%—pointed to editorial and review websites. Brand-owned sites accounted for 26 occurrences, or 9.4%. Marketplaces, retailers, and comparison or listing services together supplied 78, or 28.2%.

Three buying questions, without naming a brand.

On 28 September 2026, we sent each question independently to ChatGPT, Gemini, Claude, Perplexity, Grok, and Meta AI. Each received the same context: shopping in Germany, the date, and prices in EUR. We did not explicitly request citations.

01Choosing a type

“What kind of coffee machine should I buy for two cappuccinos a day if I don't want much cleaning?”

02Comparing costs

“For two cappuccinos a day, is a bean-to-cup or capsule machine cheaper over two years?”

03Finding an offer

“Which coffee machine under €400 suits two cappuccinos a day, needs little cleaning, and can reach Berlin within a week?”

This produced 18 completed answers. We classified the websites appearing in their links and source lists.

Who appeared at each buying stage?

Choosing a type: review and editorial sites accounted for 64 of 87 occurrences. Brand-owned websites appeared nine times; marketplaces and retailers together appeared six times.

Comparing costs: editorial and review sources led with 55 of 88 occurrences. Marketplaces appeared ten times, while community and social sites appeared five times.

Finding an offer: marketplaces and comparison or listing sites together accounted for 41 of 102 occurrences—40.2%. Editorial and review sites were the largest single category, with 45 occurrences.

The percentages pool link occurrences, so assistants returning more links contribute more weight. Rounded column percentages may not add to exactly 100%.

The websites that appeared most often.

Coffeeness led with 13 occurrences, followed by billiger.de with 11. Brewmance and De’Longhi each appeared nine times. This is a count of appearances, including repeats—not a count of distinct pages, independent endorsements, or assistants recommending a brand.

A site can also contribute several kinds of material. The archive includes a Nespresso manual, a Testberichte review roundup, and a billiger.de product comparison. The website category describes who hosts the material; it does not imply that every link is a product listing.

How we classified the websites.

• Brand-owned: official manufacturer websites, including their shops, guides, and manuals. Examples: Philips, Nespresso, and De’Longhi.
• Retailer: stores and equipment suppliers selling or supplying multiple brands. Examples: Coffee Friend and Coolblue.
• Marketplace: platforms hosting multiple sellers, including hybrid stores. Examples: Amazon, Galaxus, MediaMarkt, and Wolt.
• Comparison / listings: price comparisons, product directories, and business listings. Examples: billiger.de, idealo, and Justdial.
• Editorial / reviews: publications, buying guides, reviews, and review aggregators. Examples: Coffeeness, Brewmance, and Testberichte.
• Community / social: discussions, social and video platforms, and community-contributed data. Examples: Reddit, YouTube, and Numbeo.
• Other / unclear: destinations outside these groups or with insufficient classification evidence. This run contains one IndexBox market-data link.

These are editorial classifications based on website role, saved titles, destination URLs, and spot checks. A retailer’s buying guide stays a retailer link. Affiliate links alone do not turn a review site into a retailer. For hybrid platforms, we use the marketplace category; for social platforms, we classify the host rather than infer the author’s affiliation.

What this means for a business.

Our interpretation: your own website is one part of the source landscape around a buying decision. In this sample, editorial sites supplied the largest share. Marketplaces and comparison sites accounted for 40.2% of occurrences in the purchase question.

• Inspect the review pages that appear. Identify which buying guides and comparisons discuss your category and how they describe your products.
• Check commercial listings for purchase questions. Review the product information on marketplaces, retailer pages, and comparison services appearing in the source lists.
• Keep your official product information useful. Brand websites appeared through product pages and supporting material, including manuals.

These are places to investigate, not proven tactics for gaining citations. We did not test whether changing any page increases future appearances, traffic, or sales.

Every occurrence counts. Nothing is deduplicated.

We extracted every HTTP(S) link occurrence from the original answer text and every URL entry in its returned source list. If a URL appears twice in the text and once in the source list, it counts three times. Repeats within a list, across answers, and across questions also remain in the total.

Query parameters and fragments are preserved. For website summaries, subdomains are grouped under their classified domain: for example, acc.philips.de belongs to philips.de. Country domains such as amazon.com and amazon.de remain separate. This grouping never removes occurrences.

Meta’s saved answer text and returned source arrays contain no URLs. Its separate search-tool results do contain links. Raw search-tool results are outside this study’s answer-and-source-list count for every assistant; Meta’s zero does not mean it used no sources. Numbered markers and bare website names do not add URL occurrences.

Returned source lists can include retrieved pages whose use in the answer is unconfirmed. We retain them in the count and label their origin, without guessing citation-marker mappings. This measures the recorded source mix of these configured API models, not consumer-app citation displays. Search settings and provider source capture differ; unresolved Google source redirects may already have been omitted by the collection helper.

One product category, three questions, one run. These shares describe this dataset, not a general benchmark. We did not test repeatability or independently audit every source’s accuracy.

See which Citations appear around your business.

doping.ai connects buyer Probes with original answers and Citations. Use Evidence to inspect what appeared, then examine the source pages behind it. The occurrence percentages in this study are an editorial analysis, not a product metric.