Who does AI read?

A founder in Thingtesting's recent AI-shopping piece asked the question every brand is circling: “Does AI become Instagram, or does AI become Google?” Our citation data has a partial answer: inside shopping answers, AI is already part creator platform, and the creator economy it runs on doesn't look like the one brands are used to buying.

The data. Every month we run frozen prompt sets through four AI assistants across consumer categories and resolve every cited source. This note rolls up 14,374 citations across beauty, streaming, live TV, and credit cards: for each, what share of the sources behind the answer is a creator or community (YouTube, Reddit, TikTok, X, Substack) rather than an indexed website.

Shares are of all resolved citations within each monthly study (grounded responses only). All four assistants run through one shared retrieval layer, so these shares describe that layer's supply; native vendor search layers differ, especially on social content (measured here).

Finding one: the creator share is structural, and it varies by category. Between 15 and 34 percent of the sources behind AI shopping answers are creators and communities. Live TV is the heaviest at 34 percent, almost all YouTube: cord-cutting is a how-to decision and the how-to lives on video. In credit cards, YouTube alone out-cites any single issuer's own website. If AI answers shape your category, a third of the shelf may be people, not pages.

Finding two: the reach economy inverts. Of the 2,263 YouTube citations where we resolved channel size: channels under 100K subscribers take 54% of citations (336 channels); channels over a million subscribers take 5.6% (38 channels). The mid-tier takes the rest (40.4%). Retrieval rewards specific, recent, question-shaped videos, not audience size. The influencer pricing model, which is priced on reach, is mispriced for this channel in both directions.

Finding three: the supply moves month to month. In May, our streaming study's raw responses contained zero Reddit citations. In June, same category, same prompt families, Reddit was 3.5% of all citations, led by community threads. We verified the May absence against the stored raw payloads: Reddit genuinely entered this retrieval layer's streaming supply between the two runs. The shelf AI reads from is not static, which is why we re-measure monthly instead of auditing once.

So: Instagram or Google? Inside the answer itself, neither, exactly. AI assembles shopping answers from a mixed shelf where creators and communities hold a structural minority share that varies by category, where tiny channels out-cite mega-channels, and where whole platforms can enter the supply in a month. For a brand, the practical reading: the creators worth knowing are the ones AI actually cites in your category, and that list does not match the one reach-based influencer tools produce.

Conversational Commerce Intelligence

mapou measures how AI assistants alter commercial decisions inside high-intent consumer categories. The per-creator citation detail behind this note lives in mapou's Source Intelligence.