Hi everyone, this is Neo.

If you run an independent ecommerce site, you’ve probably spent the last two years staring at one metric: is my brand being cited in AI search? Two fresh studies just dropped that put a painful truth on the table — there’s a widening gap between who AI recommends and who AI cites.

Both landed around the same time in late August:

• Shopify agency Shero Commerce analyzed 1,851 cited sources across Google AI Mode, ChatGPT, and Perplexity, spanning 60 product categories.
• AI platform Trendos crunched 107 million AI answers to map the citation source structure of three industries across four engines (ChatGPT, Perplexity, Gemini, and Google AI Overviews).

Different angles, different samples, same verdict: brand-owned websites are nearly invisible in AI citations, while community content and third-party media own the cited slots.

Let me merge the two datasets into one picture, then get into what you should actually do about it.

What Shero found: brand pages are just 2.8% of citations

Shero’s analysis (titled “Duplicate Content and AI Citations”) asked purchase-type questions across 60 product categories in Google AI Mode, ChatGPT, and Perplexity, then logged every source the tools cited. The headline numbers are brutal:

• Of all 1,851 cited sources, only 2.8% were the brand’s own page; 59% were publishers and review media — Good Housekeeping, Verywell Fit, Reviewed.com, Reddit, that crowd.
• The subtle part is the recommend-vs-cite mismatch: of 159 times a brand was recommended by name, its own page was the cited source only 31% of the time. In plain terms, roughly two thirds of favorable mentions name the brand — and hand the citation to a reviewer.
• A concrete example: ask about squat-proof leggings, and both ChatGPT and Perplexity will name Gymshark, Alo, and Beyond Yoga. The links they cite point to Good Housekeeping and Reddit.
• In Google AI Mode “store check” queries, established DTC brands were cited or recommended just 9.5% of the time across 60 categories — and in a third of those categories, not a single well-known brand was named at all. The answer went to a Forbes list, The Good Trade, CNN Underscored, OutdoorGearLab, or a marketplace.

Shero also checked a technical angle: 88% of stores emit product structured data, yet on 59% of them, the AI-readable product description is under 50 words or missing entirely. The report puts it bluntly: a 40-word description is fine for a human who’s already looking at five photos and a price, but it gives a text-only AI model almost nothing to reason about.

One more stat worth knowing: 20% of sampled product descriptions were identical or very similar to text on other websites (mostly a brand’s own copy scattered across channels it sells through). That said, Shero itself admits the data doesn’t prove duplication directly causes lost citations — so don’t over-read that one.

What Trendos found across 107M answers: citations are industry-specific

The second study comes from Trendos (full disclosure: it ran on SEJ as a sponsored piece — the data is worth referencing, but read the conclusions with a vendor’s lens on).

Trendos analyzed 107 million AI answers across three industries — IT & solutions services, consumer goods, and communication services — and mapped the top citation sources for each engine. Two findings stand out:

Finding one: community and user-generated content owns about half of all leading citations.

Across all three industries, the share of community/UGC sources (Reddit, YouTube, forums, Q&A communities) among leading citations was strikingly consistent — close to 50% everywhere. Whether you sell enterprise software or car wax, real people talking in public accounts for half of the sources AI reaches for. If your brand is absent from those threads, you’re invisible in half of the citation pool.

Finding two: the other half is completely industry-dependent.

• IT & solutions services: third-party proof wins. Your homepage won’t carry you — review directories and B2B listings are the citation workhorses.
• Consumer goods: brand retail and marketplace listings (Amazon product pages, price comparison sites) lead.
• Communication & media: Wikipedia and press coverage feed the citations.

Trendos’ action items are refreshingly concrete: claim the two or three directories your buyers actually shortlist from; fill every field (categories, descriptions, features, integrations, pricing, screenshots — half-empty profiles get skipped by buyers and engines alike); match your category and feature wording to how prospects phrase questions to AI; collect recent reviews on the directories that actually surface in AI answers, steadily rather than in one big batch; and reply to reviews, including the critical ones — your responses add fresh, quotable text to the page.

Putting the two studies together: three conclusions

1. Being recommended ≠ being cited. This is the core takeaway. An engine can recommend your brand by name (brand mindshare working) while handing the citation to a review site (credibility working). Recommendation drives conversion; citation drives “looks trustworthy” — two separate machines.
2. Community content is the biggest winner in AI citations. Reddit keeps showing up in Shero’s data, and community/UGC hits ~50% in Trendos’. That’s not a coincidence — the density of discussion-type content in training data shapes citation preferences.
3. Brand-owned pages are starving themselves on machine-readability. 59% of stores ship product descriptions under 50 words in their structured data. It’s a problem brands can fix themselves — and almost nobody has.

What to do about it: a 5-point checklist

1. Write product descriptions as text for AI. The description field in your structured data should be a 50-word-plus “information packet” with real specifics — materials, use cases, fit, applicable scenarios — not a one-liner. Lowest cost, most direct win, available right now.
2. Find your category’s citation list. Don’t just track brand keywords. Ask the AI engines a few typical purchase questions in your category and see who they cite — that’s your benchmark list.
3. Community strategy means getting users to talk about you — not talking about yourself. Reddit threads, forum answers, YouTube comments: real user voices are the raw material AI cites. Self-promotional brand accounts aren’t on that list.
4. Pick your battlefield by industry. IT/software → attack review directories. Consumer goods → attack marketplace listings and reviews. Media/content → attack Wikipedia and industry press.
5. Make reviews targeted and continuous. Build up recent reviews on the directories that appear in AI answers, and keep replying — including to negative ones — so the page stays a source of fresh, quotable text.

Neo’s take

The most valuable thing in these two studies isn’t the low brand-citation rate itself. It’s the mechanism underneath: AI search is splitting “brand awareness” and “information attribution” into two separate ledgers.

In the old SEO world, your brand site was both the information source and the conversion page — one fish, two dishes. In the AI era, the recommendation list and the citation list are diverging: the recommendation list runs on brand mindshare (how much you’re discussed, how often authorities mention you), and the citation list runs on information quality (which page can actually support the answer).

Two implications:

• The good news: independent sites don’t need to beat Reddit and Good Housekeeping at the citation game. But you can earn a seat on the recommendation list by raising the density of discussion around your brand — a relatively fair playing field for small brands.
• The bad news: if your product data is a lazy 40-word description, you don’t even have a ticket to the citation list. That’s not the platform’s fault — you made your own data thin.

One caveat before you take these numbers to your boss: both studies have limits. Trendos is sponsored research; Shero is an agency’s own analysis (sample skews to Shopify stores). Don’t trust any single study’s numbers — but when two independent studies point the same direction, trust the direction: in the world of AI citations, community and third-party content are eating brand-owned territory, and the one card brands can play immediately is making their own pages thicker for machines to read.

Being recommended is luck. Being cited is substance. Being cited consistently is brand. Might be worth sticking that on the whiteboard.