Why AI Names Your Store But Links a Review Site

Ask ChatGPT for the best walking boots for wide feet and it may name your brand out loud. Then it links Outdoor Gear Lab. The advice sitting on page one for this problem says that gap is entirely off-page, that your own site barely counts anymore, and that the fix is a months-long campaign to get listed and reviewed and mentioned somewhere else.

My position is narrower and a lot more useful to you on a Tuesday. If you sell products or local services and your pages carry copy the manufacturer handed you, or copy your web guy syndicated across forty similar stores, that duplicate text is a specific and now measurable reason the machine credits a review site instead of you. It is on your side of the fence. You can fix it this quarter without sending a single outreach email.

The number that changed my mind on how big this is: across 1,851 sources cited by Google AI Mode, ChatGPT and Perplexity, only 2.8% were pages owned by the brands being discussed. In the same body of work, 20% of a sampled slice of the product descriptions were duplicate or near identical. Those two facts sit next to each other for a reason.

The copy on your product page already exists on forty other sites

Open your best-selling product page. Copy one full sentence out of the description. Paste it into Google inside quote marks. If you sell anything you did not manufacture yourself, you will get back a wall of results: the distributor, three marketplaces, a dozen competitors, all running the same paragraph the supplier emailed everyone in the channel. Local service sites do the same thing in a different costume, where the "Emergency Plumbing" page is the template text the web agency has shipped to sixty plumbers with the town name swapped.

I have run search for more than 300 businesses since 2008, a good share of them ecommerce stores and local service companies across the US and UK, so rewriting supplier-fed copy is a job I have done with my hands rather than diagrammed on a slide. It is also a standard I hold my own site to, which is why I will not publish two thin pages for two spellings of the same query. That rule used to be about Google penalties. It is now about whether a language model has any reason to quote you rather than the outdoor magazine that wrote something original.

The Shero Commerce work makes the shape of the problem concrete. They looked at 8,573 product descriptions from 883 stores. One in five in their sample was duplicate or near identical. Of 173 stores they could measure cleanly, 27 had fewer than fifty words of product-specific content on the page. Fifty words is a long tweet. That is the entire evidence base a retrieval system has to decide whether your page answers the question better than a review roundup with 2,000 words of testing notes.

That gap costs money in a specific way. A model asked to recommend something has to attach the recommendation to a source it can quote. If your page says the same forty words as thirty competitors, quoting you is arbitrary, and arbitrary loses to a page that says something only that page says. You get named in the sentence and the click goes to the review site. Your brand did the work of being memorable. Someone else collected the traffic, and then the affiliate commission on your own product.

Owners hear "off-page" and stop looking at their own pages

The advice that ranks for this question is not wrong, it is incomplete in an expensive direction. Maria Dykstra, who works on citation authority, frames it cleanly: your own pages are claims, and off-page mentions are evidence, and a model weighs evidence over claims.

"This is why publishing more content makes the problem worse."

Maria Dykstra, in her guide on fixing citation authority.

She is right about volume. Publishing forty more thin pages does make it worse, and SEMrush's page on the same question argues the same direction, that the work sits off your own site. Where I part company is the conclusion an owner draws from it, which is almost always "so my site is not the lever, go buy off-page." That conclusion quietly protects the one thing the owner fully controls from ever being examined.

Publishing more content and rewriting existing content are opposite moves. One adds claims. The other makes the claims you already have worth quoting. An owner who reads "publishing more makes it worse" and concludes "so leave the product pages alone" has taken a true statement about volume and applied it to quality, which is how a business ends up with 400 pages of supplier text and a twelve-month outreach retainer running at the same time.

There is a second version of this problem that runs through your blog rather than your catalog, where eight posts you wrote about the same topic split your own authority and your own pages end up suppressing your AI citations. Same disease, different organ. Today's is the external kind, where your text is identical to other companies. That one is internal, where your text is identical to itself.

Both columns are real work. Only one of them starts on Monday and finishes when you decide it finishes. The correct order for a store with supplier copy is the left column first, because the off-page campaign you eventually run will send readers to pages that still have nothing distinctive on them.

Findable is not the same as understandable, and the audit data shows the split

Reza Moaiandin of SALT.agency audited fifty major sites on June 12, 2026, scoring them on how ready they were for AI systems. Mean readiness came in at 56.6%. Split into layers, the picture gets specific: retrievability, meaning can the machine reach and load your page, averaged 74.4%. Attribution and meaning, whether it can work out what your page actually says and who it belongs to, averaged 38.5%. Seventy percent of homepages had JSON-LD markup, which is the box most agencies tick and call the job done.

Read that as a business problem rather than a technical one. Being reachable is table stakes. Being understandable means the page contains something a machine can attribute to you and only you. Manufacturer copy fails that test by design, because it was written to be attributable to nobody in particular so that every reseller could paste it. Moaiandin's closing line on the audit is the part I keep quoting to clients: "These are all solvable problems – but only if you decide to solve them."

Shero did not test whether rewriting duplicate copy changes citations. They measured duplication and they measured citations and found the two patterns sitting together. Nobody in that study ran a rewrite and watched the citation rate move. So the claim I am making is not "rewrite and the model will cite you," it is that duplicate copy is the failure mode most consistent with what was measured, it is the cheapest thing on the list to fix, and it improves your pages for humans whether or not a model ever notices.

You will also see two numbers thrown around in this conversation: a three-times higher ChatGPT citation rate for domains with review-platform profiles, and Reddit accounting for roughly 45% of Perplexity citations. Both are Dykstra's and the field's claims rather than findings from a primary study I can point you at. Treat them as directional. Do not sign a retainer on them.

Rewrite the twenty pages that carry your money, in this order

You are not rewriting the catalog. A store with 4,000 SKUs rewriting everything is a project that dies in month two. You are rewriting the pages where a customer makes a decision, which is usually a small slice of the catalog carrying most of the revenue.

What goes in the rewrite is the part people get wrong. They swap the supplier's adjectives for their own adjectives and publish the same empty page in a new voice. What a model can attribute to you is not adjectives. It is facts nobody else has: the measured weight, the actual fit against a competitor, the two situations where this product is the wrong choice, the failure rate you see in returns, the question your support team answers eleven times a week. Treating product data as decision data is the whole shift, and it is why the person who has handled the product should write the page, not the person who has handled the CMS.

If you outsource this, the question to put to your agency is short and it will tell you a lot. Ask them to name the five product or service pages on your site that currently contain the most non-original text, and to show you the evidence. An agency that has looked will answer in a day. An agency that has been billing you for content while shipping template pages will explain why the question is more complicated than it sounds.

Count citations, not mentions, and count them monthly

The measurement most owners set up here flatters everybody and informs nobody. A dashboard reports that your brand was mentioned in 60% of AI answers this month, everyone nods, nothing changes. Mention is not the metric. The metric is whether the answer linked your URL, because a citation is not the same thing as a recommendation and a mention is not the same thing as either.

Build the scoreboard by hand before you buy a tool. Write down fifteen questions a customer would actually type before buying from you, in their words, not your category words. Run all fifteen in ChatGPT, Perplexity and Google AI Mode. For each answer record two things only: were you named, and was your URL cited. Do it before the rewrites, then monthly. Fifteen questions across three systems is forty-five checks, one sitting rather than a project.

Two traps to avoid while you watch that number. The first is judging it weekly, since these systems recrawl and re-rank on their own schedule and week-to-week movement is mostly noise you will mistake for signal. The second is treating an AI-visibility percentage score from a vendor as an outcome. Nobody buys because your score went from 42 to 51. Track the named-but-not-linked rate specifically, because that is the exact population this work targets: answers where you already won the recommendation and lost the click.

Give it a quarter. The mechanism you are hoping for is slow by construction, since a rewritten page has to be recrawled and re-read by the AI systems before anything downstream can change, and no honest source can tell you the exact lag. If nothing moves after three months of rewrites on your top twenty pages, you have learned something real and cheap, and you have twenty product pages that convert better for humans regardless.

Frequently Asked Questions

Will rewriting my product descriptions actually make ChatGPT cite my site?

Nobody can promise that, and be suspicious of anyone who does. The Shero Commerce study found that only 2.8% of cited sources were brand-owned and that 20% of sampled product copy was duplicate, but it did not test whether rewriting changes citations, so that link is correlation and not proven cause. What I can tell you is that duplicate supplier copy gives a machine no reason to quote your page over a review site, that fixing it is entirely within your control, and that original pages with real specifics convert better for human buyers whether or not an AI system ever notices.

What actually counts as duplicate content on a product page?

Any text that appears word for word, or close to it, on another company's site. The usual sources are manufacturer-supplied descriptions pasted straight into your store, distributor feeds, and template pages an agency built once and resold to businesses in other towns. Boilerplate you repeat across your own site, like shipping terms or a size chart, is normal and is not the problem, but it does not count toward the page being about that product. Strip the boilerplate and if fewer than fifty words remain that are specific to that one product, treat the page as thin. In the measured sample, 27 of 173 stores fell below that line.

Should I fix my own pages first or build review-site presence first?

Fix the pages first if you sell products or local services and your copy came from a supplier or an agency template, because it is faster, it costs you nothing but attention, and you control the timing. Off-page authority matters and the people arguing for it are not wrong, but it moves on editors' and platforms' schedules, not yours, and it will eventually be sending people to pages that still say nothing distinctive. Run the off-page work in parallel if you have the budget. Just do not let it be the reason your product pages never get read by anyone on your own team.

Most owners reading this will do the search I described in the first section and find their best-selling page duplicated across a dozen sites, and their honest reaction will be that it never seemed to hurt rankings much. That was true for a decade. It stopped being the only question the day a machine started deciding, on the evidence of a single page, whether you are the source or just one of the resellers. If you want a second pair of eyes on which of your pages are quotable and which are somebody else's words in your font, book an AI visibility call and we will look at your top twenty together. Your competitor did not outrank you on that answer. They just wrote something you did not.