How SaaSNet.work works

We don't collect user reviews and we don't take payment from products. We take what a product says about itself, check it against what the wider web says, and tell you what we found — including when we found nothing.

The pipeline

Discovery. Every day we pull the newest launches from Product Hunt in AI, SaaS, fintech and productivity topics. Products are picked by this feed, not by submission or sponsorship.
Claim extraction. We read the visible text of the product's own homepage and use an AI model to pull out a handful of specific, checkable claims — concrete statements like “24/7 support” or “10,000+ users”, not vague marketing tone. If the homepage can't be read, we fall back to the product's Product Hunt tagline.
Outside check. We run a web search for what people say about the product (user reviews and opinions), then an AI model judges each claim against only the results that search returned. Each claim gets one of three verdicts, defined below.
Pricing scan. We fetch the product's own pricing page directly rather than trusting marketing copy. A price is shown only when it appears on that page; otherwise the tier is listed as “not disclosed”. Tracked products are re-scanned daily and each scan is stored, so the price on a review is the most recent one we saw, and history builds up over time.
Draft. An AI model drafts a one-line verdict, three pros, three cons and a star rating from 1 to 5 using only the pricing data and claim verdicts from the steps above.
Human approval. Nothing is published automatically. A person reads the draft next to the evidence, can edit the wording, fix the category, or drop claims that were extracted badly, and only then approves it. Verdicts themselves are not hand-edited.

What the verdicts mean

Corroborated — the outside sources we found support the claim.
Contradicted — the outside sources we found conflict with the claim.
Unverifiable — the sources we found neither support nor conflict with it, or we found nothing relevant. We never treat silence as a contradiction, and we never treat it as confirmation either.

What the star rating is (and isn't)

The rating reflects the evidence we gathered: how the product's claims held up and how openly it shows its pricing. It is not the result of hands-on testing — we don't sign up for or use the products — so it says nothing about how good a product feels to use. Treat it as a measure of how well a product's own story checks out.

5 stars — nearly all claims corroborated, none contradicted, pricing clearly shown.
4 — mostly corroborated or unverifiable with no red flags, pricing mostly clear.
3 — mixed: many claims unverifiable and/or pricing vague, but nothing contradicted.
2 — at least one claim contradicted, or pricing largely hidden.
1 — several claims contradicted, or its central claims fail.

Ratings are in half-star steps. A brand-new product with little outside coverage will usually land around 3 to 4, because unverifiable is not a penalty. A person can adjust the rating before approving.

Limits you should know about

Newly launched products often have little outside coverage yet, so many claims will honestly come back unverifiable. That is expected, not a failure.
We run one search per product and judge against the result snippets, not by reading every page. The results can include review sites, forums, blogs and social posts, but we don't guarantee any particular site is covered.
Claim extraction and verdicts are produced by AI models and can be wrong. The human approval step catches some errors, not all.
Pricing pages that load prices with scripts, hide them behind a login, or use unusual layouts may show as “not disclosed”.
A review is a snapshot. The date it was approved and the date pricing was last checked are shown on each page.

Our crawler

We fetch a product's homepage and pricing page, at low volume, using the user agent SaaSNetBot/0.1. If you'd like us not to check your product, tell us.

Something wrong?

If a verdict, price or fact is wrong, request a correction. We re-check and update the review, and unpublish it if it can't be supported. Read more about how we operate on the About page.

Last updated September 19, 2026.