How this works

This is a benchmark score, not a performance rating. We have not run production traffic through these products, so we cannot tell you which works better. Anyone publishing a performance score without testing is guessing, and the guess usually correlates with who pays them.

What we can score is what is verifiable from published material. Five inputs, weighted, shown as five separate bars on every tool page so you can disagree with one without discarding the rest. A single number is unfalsifiable; five are auditable.

Three things that make a score defensible

• Nothing reaches 5. The scale is capped at 4.6 so it keeps headroom for a theoretical ideal none of these products is.
• Some things are unrated. 70 of 264 entries carry no score, because there is too little published material to judge them without guessing. Stating the absence beats inventing a number.
• The score cannot see the money. Commission data lives in a module the scoring code does not import. That is a checkable boundary rather than a stated intention — if affiliate.ts is not in rating.ts's import graph, commission cannot have reached the score.

We earn commission on some links. It never affects what we recommend or the order things appear in — nobody can pay to appear here, or to rank higher.

The distinction that matters is between commission and placement. Taking a cut when a reader buys costs nothing editorially. Selling a position in a list costs everything, because the ordering is the only thing a ranked list is for.

Take commission. Never sell position. Every competitor this site is trying to beat lost on exactly that distinction — on the main commercial query for AI SDR tools, five of seven page-one results rank themselves, two of them first.

What we cannot claim

We cannot claim to take no money from vendors, because that will stop being true. What stays true, and is checkable, is that nobody can pay to appear here or to rank higher.

The evidence is in the list. Of 264 tools covered, 0 currently have a commission arrangement with us. PolyAI, Sierra, Amazon Connect, Whisper and every enterprise contact centre platform here pay us nothing, and several of them outrank products that would. Products with no programme are included prominently and recommended where they are the right answer, because that is the only evidence of independence that means anything.

At some point a product that pays well will be worse than one that pays nothing, and we will write that down. The commercial argument, if one is needed: a site that recommends the right product builds an audience that returns; a site that recommends the paying one converts once.

These are encoded in the data model rather than kept as intentions, so the system cannot express a claim we would not stand behind.

• null means unknown, and renders as unknown.Where a vendor publishes no price we print “Custom” — never an estimate, a competitor's number, or a range we inferred.
• Every price and every rate carries a date. Anything older than 90 days is flagged by our own audit and re-verified.
• Computed is not measured. The all-in figures on this site are arithmetic over published rates. The data model has a separate field for measured cost and it is empty everywhere. When we start running real traffic, un-measured entries will stay visibly distinguished rather than the distinction being retrofitted.
• Pending is not closed. An announced acquisition and a completed one are different states with different consequences for a buyer, and the schema can express both.
• Every entity says what it is bad at. The field is required, not optional, so a page cannot duck the question. It is the sentence no vendor-authored page can write.
• Where sources disagree, we publish the disagreement rather than picking one and hiding the other.

Prices change, companies get acquired, and we will get things wrong. If something here is inaccurate, tell us and we will fix it and note what changed: corrections@ai-sdr.cc.

Vendors are welcome to correct factual errors about their own products. What we will not do is change an ordering, remove a criticism, or soften a “look elsewhere” on request.

See the cost model for the component rates behind every figure, or browse all 264 tools.