AI Tools Recommend Brands, Cite Other Sites: Data

Search Engine Journal recently reported on an analysis from Shopify agency Shero Commerce showing that AI assistants regularly name brands in their answers while linking almost anywhere else. Across 1,851 sources cited by Google AI Mode, ChatGPT and Perplexity, only 2.8% were brand-owned pages.

The full write-up by Matt G. Southern is here: AI Tools Recommend Brands But Cite Other Sites, Data Shows. It is worth reading before you touch your content plan, because it describes a gap most founders have not budgeted for.

What the data actually says

Shero looked at purchasing questions across 60 product categories and collected every source the three AI tools returned. Per SEJ's summary:

• The largest source category accounted for 59% of citations, including publications like Good Housekeeping, Verywell Fit and Reviewed.com.
• Brand-owned pages made up 2.8% of all citations.
• Shero counted 159 brand recommendations across the platforms. In 31% of those cases the brand's own page was cited. In the other roughly two thirds, the brand got named and someone else got the link.
• In Google AI Mode specifically, brands from the sample were cited or recommended in 9.5% of relevant store checks across the 60 categories. In about a third of those categories, none of the sampled brands showed up at all.

SEJ is careful about one thing, and this article will be too: the 9.5% Google figure is not comparable to ChatGPT or Perplexity, because Shero did not publish an equivalent recommendation rate for those two. Treat it as a standalone number.

The report also notes a possible contributing factor without claiming causation. In a stratified sample of Shopify product descriptions, 20% were identical or very similar to text on other sites, mostly among retailers and marketplaces reselling the same products rather than between competing brands. Shero suggests syndication may muddy attribution. The analysis does not demonstrate that duplication caused third parties to get more citations, and it does not explain why AI systems picked the sources they picked. That is the honest boundary of this data set.

One more detail that stings if you run a thin site: out of 173 stores Shero could measure cleanly from raw HTML, 27 had fewer than 50 words of product-specific content.

Why AI tools recommend brands but cite other sites

If you are a solo founder, the mental model has probably been: rank my page, get the click, get the signup. AI answers already weakened that, in the way AI Overviews changed blog traffic over the past year. This data suggests something more specific and more annoying.

You can win the recommendation and lose the click.

An assistant tells someone "look at Gymshark, Alo or Beyond Yoga" and links a review roundup. The brand got the mention. The roundup got the traffic. If you are bootstrapped and your entire acquisition model depends on the click, that is a broken loop, and no amount of on-site optimization fixes it by itself, because the assistant was never planning to link you in the first place.

The reframe is uncomfortable but simple. Your own blog is no longer the only surface that matters. The sources AI tools actually cite are also part of your distribution: comparison articles, category roundups, forums, directories and the handful of niche publications your buyers read.

A caveat worth stating plainly. Shero's data covers ecommerce product queries. It does not prove the same 2.8% ratio holds for B2B SaaS questions or local services. But the underlying behaviour, assistants preferring third-party aggregators over vendor-owned pages, matches what anyone who has typed "best CRM for small teams" into ChatGPT has watched happen. Directionally, plan for it.

What to actually do this week

Photo: Rawpixel Ltd · BY

Five things, roughly in order of effort.

1. Audit what the assistants say about you. Open ChatGPT, Perplexity and Google AI Mode. Ask the questions your buyers would actually ask, not "what is my company." Ask "best AI blog automation tool for solo founders," "cheapest way to publish SEO content without a writer," "SEObot alternatives." Write down two columns: were you named, and what got cited. That two-column table is your gap analysis. If you are named but never cited, your problem is off-site. If you are not named at all, your problem is that you do not exist in the sources these systems read.

2. Fix the thin pages. Twenty-seven stores in Shero's sample had under 50 words of product-specific content. If your pricing page, your comparison pages and your core feature pages are three sentences and a screenshot, there is nothing there for a model to lift. This is not word-count worship. It is about whether your page contains a single specific, quotable fact nobody else has published: your actual price, your actual publishing cadence, what your tool refuses to do.

3. Kill duplicated copy. Shero found 20% of sampled product descriptions were near-identical to other sites, mostly from resellers. The SaaS equivalent is the boilerplate you pasted into ten directory listings, your Product Hunt page, your G2 profile and your homepage. The same paragraph everywhere signals that your page is not the canonical source of anything. Rewrite each one. Nobody has tested whether this changes citation behaviour, so treat it as reasonable hygiene rather than a guaranteed lever.

4. Get into the sources that do get cited. The 59% category in Shero's data was third-party publications. In your niche that probably means the comparison roundups that already rank, the subreddit and Indie Hackers threads where your category gets discussed, the directories your competitors are listed in, and any newsletter covering your space. You do not need PR. You need to be a findable, described entity on pages other people wrote. The mechanics of that are in how to get cited by ChatGPT, Perplexity and AI Overviews.

5. Publish the comparison content yourself, honestly. If assistants prefer roundups, be a roundup. A genuinely fair "X vs Y vs us" page, with real prices sourced from each vendor's own pricing page, tends to get cited because it is the structured, comparative content these systems reach for. Our own roundup of AI blog automation tools is written to that standard, prices linked to each vendor. The catch: if you fake it, you lose the only thing that makes it useful.

What the two-column audit looks like filled in

The audit in step one takes about forty minutes and is the only part of this list that produces a decision rather than a chore. Run five to eight buyer questions through each assistant and record what came back.

Three patterns fall out of a completed table.

Named but never cited is the good problem. The assistants know you exist; they just prefer somebody else's page to describe you. The fix is off-site: accurate listings, presence in the roundups that already rank, and a page of your own worth linking to.

Not named at all is the entity problem, and it is slower to fix. You are missing from the pages these systems read, so no amount of on-site work registers. Directories, comparison sites and community threads come first.

Cited for nothing means you appear only on queries with no purchase intent. That is a keyword selection problem rather than a citation problem, and the answer is the same as it has always been: pick queries a site your size can actually win, as in long-tail keywords a brand new site can rank for.

Whichever pattern you land on, change one thing, then rerun the same questions in ninety days. Assistants update, your table becomes a time series, and you learn more from your own two data points than from anyone's framework.

Where this leaves the publishing cadence question

There is a temptation to read this data as "content does not work anymore, go do PR." That is the wrong lesson.

What it says is that a single site publishing once a quarter has almost no chance of being the source an AI reaches for. Citation is a volume-and-specificity game. The sites winning that 59% share publish constantly, cover categories exhaustively and keep structured, comparable information on every page. You are not going to out-publish Good Housekeeping. You can absolutely out-publish the three-person competitor in your niche who last blogged in March.

That is the boring version of the strategy: publish steadily on the specific questions your buyers ask, keep your own pages factually dense enough to be worth quoting, and spend some of your time making sure your brand is accurately described on the third-party pages that already get cited. If publishing steadily is the part that keeps slipping, that is the job Rankody does on autopilot, drafts to your approval and straight to your CMS. The off-site work is still yours, and it always was.

It is also worth separating this from the older worry about whether AI-written pages are safe to publish at all, which is a different question with a clearer answer: see what Google actually says about AI-generated content. More reporting on how AI search is reshaping traffic sits in the AI content and Google hub.

The last thing worth repeating from the SEJ piece: nobody has yet tested whether rewriting duplicated copy or restructuring pages actually changes citation behaviour. Shero mapped where citations landed, not why. So run your own two-column audit, change one thing, and check again in ninety days. That is more signal than any framework someone sells you this quarter.