When AI gives shopping advice it cites YouTube

• Five systems, 2,013 questions, 22 companies, 1 July–9 September 2026.
• Most-cited domain: 4,581 citation occurrences, covering 21 of 22 companies. 3,178 videos and seven channel pages.
• 651 questions for five D2C stores: 319 cited a YouTube video at least once.

Until recently, purchase research meant search results, several tabs and eventually a store. Increasingly, shoppers ask AI. Google displays AI Overviews for many German searches, Perplexity offers shopping from answers and ChatGPT is expanding shopping features, despite discontinuing its own Instant Checkout in March 2026.

The key issue is where research happens and which sources AI uses, rather than where payment takes place. Purchases still reach the store, but an answer names a few sources, not dozens. Being absent can mean missing the recommendation.

We wanted to identify those sources and creator content's role in them, using 44,562 actual answers.

As of 9 September 2026. Author: Moritz Lambrecht, Co-Founder & CEO. Our own measurement and analysis.

Citation occurrences for domains appearing across at least five of the 22 companies. Clients' own domains and Google grounding redirects are excluded.

Source: Ad Specialist AI monitoring: 439,236 resolved citation occurrences from 44,562 answers, 1 July–9 September 2026.

shopify.com and otto.de rank highly because several companies in the sample are Shopify merchants or Otto partners. YouTube and Reddit are the only domains appearing almost everywhere.

A single client can push a domain up the ranking. shopify.com is second because seven companies in the sample are Shopify merchants. YouTube and Reddit appear across 21 and 20 of 22 companies respectively, spanning workwear, lighting, tax advice, dentistry and software.

That's the central finding. There is no short list of universally cited specialist sources, but two platforms recur across product topics: video and user discussions.

Across all 22 companies and 2,013 questions. Each answer is counted once, regardless of how many videos it cites.

Source: Ad Specialist AI monitoring: 43,132 answers containing at least one source, 1 July–9 September 2026.

The effect is concentrated in systems using live web search. Perplexity and Google AI Overviews retrieve current sources, including videos. In our measurement, Claude and Gemini rarely cite platforms, while ChatGPT's API returns almost no sources.

For ChatGPT, this is a measurement limitation rather than a conclusion about the platform. Our API-based monitoring returns few citations. Analyses of the web interface rank Reddit first for ChatGPT and YouTube first for Perplexity, at around one-third of citations. Our figures are lower but show the same ordering.

The practical takeaway: YouTube is a leading source in the systems where AI answers are becoming a prominent part of search and purchase research.

651 purchase questions for five D2C brands in workwear, lighting, lifestyle, farm retail and accessories. Share of questions where the platform was cited in at least one answer.

Source: Ad Specialist AI monitoring: five D2C companies, 13,215 answers to 651 questions, 1 July–9 September 2026.

This is the other half of the finding, and it matters for influencer budgets. Instagram appears as a source for 5% of purchase questions and TikTok for fewer than 2%. Product discussion exists on both, but Stories expire after 24 hours, Reels offer less indexable text and platform restrictions limit crawler access.

YouTube videos offer titles, descriptions, transcripts and chapters in a lasting, machine-readable format. Of 4,581 YouTube citations, 3,178 point to individual videos and only seven to channel pages. The source is the video featuring the product, rather than the creator's profile.

Example questions included “Carhartt or Engelbert Strauss: which lasts longer on a building site?”, “How should I light a dining table to make it feel cosy?” and “Which lamps offer good quality at a fair price?” Perplexity cited videos for all three. These are comparisons, reviews and guides — formats suited to creator integrations. See the Duschbrocken YouTube case study for an example.

Share of questions with at least one YouTube video cited as a source.

Source: Ad Specialist AI monitoring: five D2C companies, 1 July–9 September 2026.

YouTube integrations are usually assessed through views, clicks and new customers. This study points to another effect: videos remain online as sources and can be retrieved when someone asks about the category months later.

In the Influencer Performance Report 2026, 85% of YouTube bookings have tracking links and 85% have reported actual views. Instagram's figures are 27% and 13%. In these datasets, YouTube combines stronger measurement coverage with content that AI systems cite most frequently.

For channel planning, Instagram buys attention over days; a YouTube integration can also leave a source online for years. Explore measurable influencer marketing and our article on AI's role in influencer content.

How to put this into practice

Collect purchase questions before briefing creators. What do customers ask AI before buying in your category? Comparisons, durability, care, price. Put those questions in the brief so the video answers them.

Make integrations easy to cite. Include the product category in the title, description and chapter marker, alongside the brand. A video titled “best work trousers 2026” is easier to match to that question.

Plan usage rights and whitelisting. Creator content reused in Meta Ads has value. So does content that stays on YouTube. Cover both in the contract.

Measure whether it happens. Identify your category's twenty most important purchase questions, then check monthly which videos AI cites. Make it a metric someone owns.

How we measured

2,013 questions for 22 companies. Questions customers might ask AI before buying: purchase advice, comparisons, provider searches and how-to questions. Test and demo accounts were excluded.

Five systems, ten weeks. Each question was regularly submitted to ChatGPT, Claude, Gemini, Google AI Overviews and Perplexity from 1 July to 9 September 2026. We collected 44,562 answers, of which 43,132 contained at least one source.

Sources resolved to domains. 439,236 citation occurrences. Google grounding redirects without a destination domain — 57,486 occurrences — were excluded. YouTube's share for Gemini and AI Overviews may therefore be understated.

Three distinct ways of counting. Citation occurrences count every mention. Answer shares count each answer once. Question shares count a question once if any answer to it cites the platform.

Our own monitoring and analysis. We run the monitoring ourselves. The study's analysis is documented in a script and can be repeated for each new period.

• Dataset: 44,562 AI answers to 2,013 questions, 22 companies, five systems
• Period: 1 July–9 September 2026
• D2C subset: five stores, 651 purchase questions, 13,215 answers
• Metrics: citation occurrences by domain, answer share by system, question share by platform, video versus channel links
• As of 9 September 2026. Author: Moritz Lambrecht, Co-Founder & CEO

These limitations are essential to interpreting the study.

• The 22 companies are our clients, not a random market sample.
• ChatGPT is measured through an API that returns few sources. A zero describes this measurement, not the behaviour of the web interface.
• A citation isn't a recommendation. We measure which sources AI uses, not whether it describes the product positively.
• The period covers ten weeks. Citation behaviour changes with model updates.