Why a worse-reviewed competitor still beats your best-reviewed product in AI answers

A brand has 4.6 stars on 900 reviews. A direct competitor sits at 4.2 stars on 200. Ask ChatGPT or Perplexity which one to buy, and the assistant names the competitor. The owner pulls up the review dashboard, confused, because by every number on it they are clearly ahead. This isn’t a scoring glitch. It’s a real mechanism, and recent research gives a surprisingly specific account of how it works.

Why this keeps happening even when your numbers are better

Answer engines aren’t comparing two spreadsheets of ratings. When a model has no strong signal to distinguish two products, it tends to fall back on whatever brand associations it picked up during training, which skew toward whichever brand is more established, more written-about, or more familiar, independent of which product is actually better that week. A 2026 study out of Trine University and Texas A&M tested this directly: it built sets of identical skincare products that differed only by brand name, one real (the incumbent) and nine fictional, and ran them through GPT-4o-mini, Claude and Gemini. Across 670 valid trials, the real brand was recommended every single time, and not one fictional brand was ever chosen when the products looked the same on paper (arXiv:2606.17443). That’s the baseline a worse-reviewed, better-known competitor starts from: a model that has learned to reach for the name it recognizes when nothing else tells it not to.

A real review advantage has to be big enough for the model to notice

The same study then gave the underdog brand a measurable edge and asked how much was needed to flip the recommendation. The answer is: not much, but not nothing either. When the unfamiliar brand’s specs were clearly better, the real brand kept winning only 1.7% to 4.6% of the time across subcategories, meaning the model picked the objectively stronger product in roughly 97% of trials. Testing five graduated advantage levels, the researchers found the switch wasn’t gradual, it was a step: at an identical rating, the unfamiliar brand won just 3.6% to 6.0% of the time, and at the smallest advantage tested, its win rate jumped to 64% to 80%. The threshold where it won half the time worked out to roughly a +0.075-star rating edge, a 1.6x review count, or a 7.3% price discount, smaller than the gap between a 4.3 and a 4.4 star rating (arXiv:2606.17443). So a brand that is genuinely ahead on reviews should usually win the comparison once that advantage is visible to the model. The catch is “visible to the model”: an edge that’s real in your own analytics but never shows up in the text an engine actually reads, because it’s buried in a dashboard rather than stated on the page or carried by a review platform the engine cites, functions the same as no edge at all.

A worse-reviewed competitor can buy a quality signal it doesn’t have

The same research then tested what happens when a brand’s product description uses authority-style language, citing credentials, named studies or expert endorsements, without changing the underlying product at all. That language broke the brand-recognition advantage 50% to 73% of the time, well above sales-pressure tactics like urgency or scarcity claims, which barely moved the outcome at 9.6% to 12.9%. The researchers quantified the effect as a “Bias Surplus Value”: authority-style language was worth roughly the same recommendation boost as a real +0.17-star rating improvement, a 15.3% price discount, or 1.9 times the review count, at no cost to produce (arXiv:2606.17443). That’s the second, less obvious way a worse-reviewed competitor wins: not by having better reviews, but by stating real credentials, certifications or named evidence plainly enough that an engine treats it as proof, while the better-reviewed brand’s page says nothing beyond “trusted by thousands of customers.” Worth being precise about the boundary here: the study’s most extreme version of this tactic was fabricated clinical claims, which its own authors flag as a false-advertising risk, not a tactic to copy. The legitimate version is using the real certifications, named studies or expert credentials you already have instead of leaving them off the page, which is the same ground building the entity signals AI engines trust covers from the brand-identity side.

Presence on a source the engine cites outweighs a few points of rating

Even when the rating gap is real and visible, where that rating lives matters as much as what it says. An 804,491-response study across ChatGPT, Google AI Mode, Gemini and Perplexity, run by Seer Interactive on behalf of Trustpilot, compared brands with no active review-platform profile against brands with even a minimal one, as few as a handful of reviews, and found the median AI citation rate jumped from 1% to 53.5% just from having an active profile at all, before accounting for review volume or star rating (Seer Interactive, 2026). Going from a minimal profile to a heavily managed one added only another 6 to 25 points on top of that. Read against the brand in our opening example: if the 4.2-star competitor’s reviews sit on a platform the engine actually cites for that category, and the 4.6-star brand’s reviews sit mostly on its own site or a platform the engine rarely pulls from, the competitor’s weaker rating can still be the only one the model has legible evidence for.

Diagnose which of these is actually yours

Three causes, three checks, worth running in this order because each one is cheaper to rule out than the next. First, confirm the rating or review-count gap you believe you have is actually stated somewhere an engine would read it, not just true in your internal reporting; tracking competitors in AI answers is how you see what the model is actually citing for both of you side by side. Second, check whether the competitor’s product page states real credentials, certifications or named evidence that yours doesn’t, since that’s a content gap you can close directly rather than a reviews problem at all. Third, check where each of you is reviewed: the sources AI engines trust covers how to find which review and directory sites are actually feeding your category’s answers, and a rating advantage sitting on the wrong platform won’t help you until it’s also on the right one. If the losing product is one SKU inside your own catalog rather than a competitor’s, the mechanism is closely related but the fix is different; why an AI shopping assistant picks one of your products over another covers that case specifically, and why AI recommendations make reviews matter more goes deeper on how engines read review text rather than just the star count.

Check it instead of guessing it

A better dashboard number doesn’t settle which of these three is costing you the recommendation, and guessing wrong means fixing the wrong thing for a quarter. Run a free Visibility Check on the exact query where you’re losing to see which source the engine is actually citing, and if the pattern holds across more than one prompt, Monitor tracking will show you whether it’s a one-off or a standing gap worth the content or outreach work to close.