TL;DR
- Search Engine Journal reported on 26 August 2026 that Shero Commerce found only 2.8% of 1,851 sources cited across Google AI Mode, ChatGPT and Perplexity were brand-owned pages.
- Shero counted 159 brand recommendations across the three tools, and the brand's own page was cited in 31% of them, so most named brands got no source link.
- Publisher and review sites such as Good Housekeeping and Verywell Fit made up the largest source category at 59% of citations.
- Shero did not test whether rewriting syndicated copy or changing page structure improves citation odds, and Shero is a Shopify SEO and AI agency.
- Of 173 stores measurable from raw HTML, 27 carried fewer than 50 words of product-specific content.
Being recommended by an AI assistant and being cited by one are two different outcomes, and new data suggests most brands are getting the first without the second. Search Engine Journal reported on 26 August 2026 on an analysis by Shero Commerce which found that only 2.8% of the 1,851 sources cited across Google AI Mode, ChatGPT and Perplexity were brand-owned pages, even as those same tools named brands directly in buying answers.
That gap is the whole story. A shopper can ask an AI tool which leggings to buy, get a named brand back, and follow a link that goes to a magazine review rather than to the brand. If you report AI visibility as a single number, those two very different events collapse into one, and the number stops telling you what is actually happening.
What Shero measured
Shero examined purchasing questions across 60 product categories and collected every source the three AI tools returned. The total came to 1,851 sources. The largest single source category accounted for 59% of citations and was made up of publisher and review sites such as Good Housekeeping, Verywell Fit and Reviewed.com. Brand-owned pages accounted for 2.8%.
The citation work sat inside a broader review of product content on Shopify stores. Shero collected 8,573 product descriptions from 883 stores with live product data, which is where the second half of the report, on duplication and thin pages, comes from.
One thing to state plainly before reading any of the numbers: Shero Commerce is a Shopify agency that specialises in SEO and AI services. The analysis is competent and the figures are specific, but the firm sells services in the area the report describes as a problem. That does not make the data wrong. It does mean the framing deserves the same scepticism you would apply to any vendor-published research, and it is a reason to hold the interpretation loosely even where the counts are solid.
Recommended and cited are two separate outcomes
Across all three platforms Shero found 159 brand recommendations, and in 31% of those cases the brand's own page was cited alongside the recommendation. In the remaining cases the brand was named while the source link pointed somewhere else. Shero's example: both ChatGPT and Perplexity recommended brands including Gymshark, Alo and Beyond Yoga in response to a squat-proof leggings query, while citing third-party sites as sources.
For a marketer this is the practical takeaway, and it changes how you read your own AI-visibility reporting. A tool that counts brand mentions in AI answers and a tool that counts citations to your domain are measuring different things, and a healthy mention count can sit on top of a citation count of roughly zero. Neither number is wrong. They answer different questions, and only one of them is about traffic.
The table below sets out the counts Shero reported, as reported by Search Engine Journal. Every figure comes from the same analysis, so read them as one sample rather than as separate studies.
| What was measured | Figure Shero reported |
|---|---|
| Sources cited across Google AI Mode, ChatGPT and Perplexity | 1,851 |
| Share of those sources that were brand-owned pages | 2.8% |
| Share held by the largest source category | 59% |
| Brand recommendations found across all platforms | 159 |
| Recommendations where the brand's own page was cited | 31% |
The Google AI Mode figure is not a cross-platform comparison
Shero reported that in Google AI Mode, brands were cited or recommended in 9.5% of relevant store checks across the 60 categories, and that in around one third of those categories no sampled brand appeared at all. That is a useful figure on its own terms.
It is not a figure you can put next to ChatGPT or Perplexity. Search Engine Journal states this explicitly: Shero did not publish an equivalent recommendation rate for the other two platforms, so there is nothing to compare the 9.5% against. If you see that number quoted as evidence that Google names brands less often than ChatGPT does, the comparison has been manufactured after the fact, not measured.
The duplication finding is a hypothesis about attribution, not a demonstrated cause
In a stratified sample of the product descriptions it collected, Shero found 20% were identical or very similar to text on other sites. The duplication mostly appeared among retailers and marketplaces reselling the same products, rather than between competing brands writing about their own goods. The authors suggest that syndicated text spread across many domains could make attribution harder for an AI system deciding which page to credit.
That is a reasonable idea and it is worth taking seriously. It is also, at this stage, only an idea. The analysis does not show that duplicated text caused third-party sites to receive more citations. The two findings sit in the same report; no causal link between them has been established, and Search Engine Journal is careful to say so.
The related finding on page thinness is more concrete. Shopify's description field can understate how much text a page actually carries, so Shero also measured the raw HTML where it could. Of the 173 stores it could measure cleanly that way, 27 had fewer than 50 words of product-specific content. That is a straightforward content problem regardless of what any AI system does with it.
What the report did not test
The honest boundary of this research matters as much as its headline number. Shero did not test whether rewriting syndicated descriptions increases the chance of being cited. It did not test whether adding original copy changes citation behaviour. It did not test whether changing page structure makes a difference. And it does not explain why the AI systems favoured the sources they favoured.
What the report shows is where citations landed in one sample, at one point in time, across one set of 60 product categories. Establishing cause would take a deliberate before-and-after test on the same pages, which nobody in this report ran. Any advice presented as a proven route to AI citation is going beyond the evidence.
What a marketer can honestly act on
Two checks follow from this without requiring the causal claim to be true, and Relevant Audience would frame both as content hygiene rather than as citation tactics.
The first is whether your own product pages carry original, product-specific copy or the manufacturer's syndicated text. If you sell goods you did not make, there is a fair chance the description on your page also sits on a dozen other domains. That is worth knowing whether or not it affects AI citations, because it affects ordinary organic search too. The second is how thin those pages actually are. Shero's 27 stores under 50 words is a reminder that a page can look full in a template while carrying almost no text that is specific to the product. Both checks are cheap and both are defensible on their own merits, which is the point. Neither is a promised route to being cited, and anyone selling it that way is overreaching.
The third change is to your reporting rather than your pages: track brand mentions in AI answers and citations to your domain as two separate lines. Teams working on AI search visibility and on ecommerce marketing are usually looking at the same dashboard for both, and this data is a good argument for splitting them.
What this means for Thai marketers
The source names no market. Shero's sample is drawn from Shopify stores and the report gives no country breakdown, no language breakdown and no statement about where the purchasing questions were framed. There is no Thailand figure in this research, and presenting the 2.8% or the 9.5% as a Thai-market number would be inventing a claim the data does not carry.
What can be said is narrower. The structural condition Shero describes, where resellers and marketplaces publish the same manufacturer copy on many domains at once, is a familiar shape in any market with a large marketplace layer, and Thailand has one. Whether AI answers in Thai behave the way this English-language sample behaved is untested, by Shero and by anyone else quoted here. A Thai brand that wants to know should measure its own AI answers for its own category rather than importing a US-flavoured percentage. The original report was published by Search Engine Journal on 26 August 2026 and can be read in full at Search Engine Journal.
FAQ
Does this mean AI tools ignore brand websites?
No. Brand-owned pages made up 2.8% of the 1,851 sources Shero counted, which is small but not zero, and brands were still named in answers 159 times across the platforms tested. The finding is that brand pages are cited far less often than publisher and review sites, not that they are excluded.
If ChatGPT names my brand, is it linking to my site?
Not usually, on this evidence. Shero found the brand's own page was cited in 31% of the 159 recommendations it identified, so in most cases the recommendation and the source link went to different places.
Will rewriting my product descriptions get my pages cited?
The report does not say, and it did not test this. Shero found that 20% of sampled descriptions were duplicated across sites and suggested syndication may complicate attribution, but it ran no test of whether rewriting, adding original copy or restructuring pages changes citation behaviour.
Can I compare the 9.5% Google AI Mode figure to ChatGPT and Perplexity?
No, because there is nothing to compare it to. Search Engine Journal notes that Shero gave no equivalent recommendation rate for ChatGPT or Perplexity, so the Google figure stands alone.
Does the research say anything about Thailand or Thai-language search?
It does not. Shero's analysis covers Shopify stores and 60 product categories with no country or language breakdown stated, so any Thai-market reading is inference rather than data.
