AI assistants do not just recommend products. Sometimes they steer you away from one by name. In our last 28 days of sampled answers, that happened 614 times across 189,375 product appearances, a 0.32% rate, touching 389 different products.
Here is the part nobody would guess. The products AI warns people away from most often are not scams, zombies or also-rans. They are the biggest names in software.
| Product | Warnings | Appearances | Warning rate |
|---|---|---|---|
| Kubernetes | 14 | 667 | 2.1% |
| AWS | 13 | 671 | 1.9% |
| Grafana | 7 | 307 | 2.3% |
| WordPress | 6 | 645 | 0.9% |
| Shopify | 6 | 728 | 0.8% |
| Notion | 5 | 823 | 0.6% |
| Sentry | 5 | 146 | 3.4% |
| Stripe | 5 | 768 | 0.7% |
| Grafana Loki | 5 | 102 | 4.9% |
| Slack | 5 | 1,150 | 0.4% |
Products are shown only where we hold 100 or more sampled appearances in the window.
Why the giants
We read the answers behind every warning in this table. The warnings are almost never "this product is bad." They are "this product is wrong for YOU," aimed at a specific kind of asker, and four reasons dominate.
Overkill. Kubernetes warnings cluster on questions from small teams: the assistants steer solo developers and startups toward simpler deployment platforms before they inherit a cluster.
Cost and complexity drift. AWS warnings appear when the asker says budget, indie, or "I do not want to become a cloud engineer." The assistants suggest platforms with flat pricing instead.
Maintenance burden. Self-hosted-tooling warnings target small DevOps teams, with lighter stacks recommended in the same breath.
Lock-in. Vendor lock-in language shows up around managed platforms when the question mentions open source or portability.
What it means
Being the category default cuts both ways in AI answers. The same visibility that gets a product recommended thousands of times also makes it the thing assistants steer certain buyers around. A warning rate of 2% on 667 appearances is not a reputation problem. It is evidence the engines segment buyers, and the default answer loses exactly where a lighter competitor fits better.
If you sell against a giant, this is the open lane made measurable: the assistants are already arguing your case to small buyers, by name, in the answers where the giant is the wrong fit.
If you are the giant, the warnings are a map of the segments where you leak.
Method
Every product appearance in our sampled answers is classified into one of four roles (recommended, listed, context, warned against) by the extraction pipeline behind the public AI Recommendation Index. Warned against means the answer steered the asker away from the product for their stated situation. All answer excerpts are paraphrased, never quoted. Per-product rows require 100 or more appearances in the 28-day window ending August 26, 2026. Full methodology and the open dataset: orbator.io/data.