Amazon has made a major move that many sellers may not have noticed yet.
COSMO algorithm is fully launched.
In simple terms, this is an AI-driven product recommendation system running in parallel with the original A9/A10 algorithm. A9 determines if you can enter the candidate pool; COSMO determines if you get recommended to users.
This means whether your product gets recommended depends not only on sales and keywords, but also on user feedback.
And the most direct expression of user feedback is reviews.
What Exactly Is COSMO Doing?
The previous algorithm logic was simple: user searches “Bluetooth earphones,” system sorts by sales volume, conversion rate, keyword matching, etc., putting “most likely to convert” products at the top.
COSMO is different. It’s trying to “understand” user intent and provide personalized recommendations.
For example, when a user searches “earphones for running,” A9 looks for “Bluetooth earphones + high sales + high conversion.” COSMO first understands “what does running scenario need” — maybe waterproof, lightweight, not easy to fall out — then finds products matching these characteristics.
This is why your product might have higher ratings and bigger sales but less exposure than a competitor — because COSMO is looking for “the right product,” not just “best-selling product.”
How Do Negative Reviews Affect COSMO Recommendations?
Here’s the key point.
COSMO and Rufus (Amazon’s AI shopping assistant) analyze your reviews, extracting product characteristics and user feedback.
If reviews cluster around certain negative keywords — like “breaks easily,” “not waterproof,” “falls out easily” — COSMO will determine your product has problems in these areas and reduce recommendation weight.
More seriously, research found: when 5-7 negative reviews about the same defect gather, Rufus will stop recommending the product.
5-7 reviews, doesn’t sound like much, right? But the concentration of negatives in many products is quite high — maybe just a few problems repeated over and over.
Rating Doesn’t Matter Anymore? Trends Do
Another key change: the importance of static ratings is declining, rating trends are becoming more critical.
Previously you could rely on accumulating positive reviews to maintain 4.5 stars, then coast. That doesn’t work anymore.
COSMO looks at your review “sentiment trend” — is it improving or declining. If your rating is still 4.5 but has been consistently dropping over the past three months, COSMO will determine your product is “deteriorating” and reduce recommendation weight.
Rufus will tag products with “rising negatives but still acceptable ratings” as “risk products.” A 4.4 with increasing negatives is more dangerous than a 4.2 with decreasing negatives.
A Fact That Should Make You Think
Previously, negatives affected conversion rate — people who saw negatives might not buy.
But conversion rate can be “controlled”: brushing orders, running ads, promotions — you can always pull up conversion rates.
Now negatives directly affect exposure. COSMO doesn’t recommend you, users can’t even find you in search, high conversion rate is useless.
This is like negatives used to be “internal injury” — visible but not necessarily fatal. Now it’s “external injury” — written all over your face, impossible to hide.
How Should We Respond?
Honestly, the situation is more severe than before. But it’s not without solutions.
1. Regular Review “Physical Exams”
Previously maybe you checked ratings once a month; now you need to pay more frequent attention to review changes.
My suggestion: use tools for incremental comparison, analyze review changes weekly. See if negative keywords are increasing, if any new problems are emerging.
2. Focus on “Problem Concentration” Not “Total Negative Reviews”
5-7 reviews about the same problem triggers Rufus’s “stop recommending” mechanism.
So not only focus on negative review quantity, but also focus on: are problems concentrated in one point? 100 negatives distributed across 20 issues might not be as dangerous as 10 negatives concentrated on 1 issue.
3. Proactively “Fix” Problems, Not “Fix” Ratings
Previously was “when negative reviews come, please delete them.” Now “if the problem isn’t solved, please deletion doesn’t help.”
COSMO reads review content — deleting reviews but problems remain, COSMO will still reduce weight.
The truly effective approach: solve product/service problems at the root, let new reviews improve.
4. Focus on Three-Dimensional Problem Classification, Address Based on Symptoms
Negatives aren’t a monolithic block — there are product issues, service issues, and experience issues.
COSMO handles different problem types differently — for example, “safety hazards” might directly trigger product removal, while “poor packaging” might only reduce recommendation weight.
Figure out which type of problem it is to determine priority and handling method.
A Real Warning Case
I previously analyzed a smart band, rating 4.3, looked okay. But review trends showed:
- Negative review rate in last 30 days 15% higher than previous 90 days
- Negative keywords concentrated on “battery life overstated” and “App connection unstable”
- One 1-star review got 50+ “helpful” votes
This situation is very dangerous under COSMO framework — problems concentrated, repeatedly mentioned, with diffusion trend.
If I were this seller, I would immediately:
- Confirm if the problem truly exists (contact customer service to check complaint records)
- Contact factory to investigate battery life overstatement cause
- Contact App team to investigate connection issues
- Reply to reviews explaining fixes are in progress
By the next algorithm update, this product would likely be reduced in ranking.
Tool Recommendations
I currently use AstrMap for review monitoring — several features are quite practical:
- Three-dimensional problem classification: can distinguish product issues from experience issues, prioritize COSMO high-weight issues
- Problem concentration statistics: see the count and proportion of each problem type, judge danger level
- Incremental comparison: regularly fetch reviews incrementally, compare trend changes
- Hidden negative mining: discover hidden negative signals in positive reviews before they explode
Of course, tools are just auxiliary — the real importance is realizing: the logic of negative review management has changed.
Previously was “handle when negatives come,” now is “proactively monitor problem signs and handle before they affect exposure.”
One question in the comments: Do you currently have a habit of regularly analyzing reviews? How often do you check review changes?