I Audited Eight 'AI Outreach' Tools Against Their Own Websites. Here's the Body Count.

TL;DR: The "we ran the same leads through five tools" genre has a receipts problem, so I did the audit that can actually be checked: eight LinkedIn outreach vendors, graded against their own homepages and docs on 2026-08-06. Five run in the cloud, two are still a browser extension or desktop app, six market "AI," and zero of eight publish a methodology-backed benchmark on the homepage. Only two publish any aggregate number with a sample size anywhere, both buried in blog posts, and exactly one admits on its own site that LinkedIn forbids what the product does.

I do outreach for a living and I'm cheap with my time, so once a year I re-check the market instead of trusting last year's impression. This year I did it the way this site does everything: no vibes, no unverifiable reply-rate screenshots, just what each vendor's own website and docs actually commit to. Full disclosure up front: I run Reachium and build on it, so treat me as biased and check every claim, which is the entire point of auditing things that are public.

Why not just publish my own reply-rate bake-off?

Because unverifiable "I ran 500 leads through five tools" tables are precisely the disease the AI SDR listicle audit diagnosed: numbers that could only exist if someone ran a test, presented without one artifact. If I show you a comparison, you should be able to check it. You can't check my inbox. You can check a homepage. So the audit below grades the eight best-known tools (Expandi, Dripify, Waalaxy, Linked Helper, Dux-Soup, Octopus CRM, HeyReach, La Growth Machine) on five things their own sites reveal.

What is actually under the "AI" paint?

Architecture first, because it decides whether you bought a product or a macro. Five of eight run fully in the cloud: Expandi, Dripify, HeyReach, La Growth Machine, and (only since July 2026) Waalaxy, which finally retired its Chrome extension. Octopus CRM is still a Chrome extension, Linked Helper is a desktop app, and Dux-Soup sells both a browser mode and a cloud tier. Close the laptop on the wrong plan and your "automation" stops. In 2026, a tool that needs your browser open is not a product, it's a macro with a subscription.

The AI labeling is nearly universal and nearly weightless. Six of eight market "AI" on the homepage: AI writers, AI icebreakers, "AI-Powered Campaigns." Only HeyReach goes full 2026 and sells "AI agents" running outreach "end-to-end." The two that don't say AI at all, Expandi and Octopus CRM, sell the same category of automation. The word is decor either way; the homepage token-count post shows how cheap that decor has gotten.

Where are the receipts?

This is the finding that should change how you shop. Not one of the eight publishes a methodology-backed benchmark where buyers actually look. The homepages carry testimonials, review-site badges, and case-study cherry-picks ("57% acceptance rate!") with no denominators.

Exactly two vendors publish any aggregate number with a stated sample size, and both bury it in a blog post: HeyReach analyzed 96,051 of its own campaigns and reported a 21% average acceptance rate with 22% of accepted replying, and Waalaxy cites a study of almost 10 million connection requests whose actual percentages have since vanished from the live page. Credit where due: HeyReach's number is a real receipt, and it is not a flattering one, which is exactly why I believe it.

For calibration, the only other public dataset in the category with a stated cohort rule is Reachium's outreach benchmarks: 180,155 matured connection requests, 27.11% accepted, 27.55% of accepted replying, 1.99% of accepted booking. Two vendors publishing denominators out of a whole category selling "results" is the body count that matters.

What does LinkedIn itself say about all of this?

The part every homepage whispers past: LinkedIn's own help pages say it doesn't allow "third-party software or browser extensions that scrape, modify the appearance of, or automate activity" on the site, and that a restricted account gets re-enabled only after the offending software is disabled. Every tool in this audit automates activity. Seven of the eight respond to that reality with the word "safe" and no specifics. Exactly one, La Growth Machine, states plainly on its own safety page that "LinkedIn does explicitly forbid" automation tools and warns what triggers flags. One honest sentence in eight websites.

That asymmetry is your buying signal. A vendor that documents limits and risks (Expandi and Dripify at least publish concrete daily caps in their docs) is planning for your account to survive. A vendor that just says "100% safe" is selling you a countdown. Reachium's approach, for disclosure, is the boring one: verified API, accounts calibrated to roughly 25 invites a day, and the worst failure mode observed in its data being a temporary rate limit rather than a permanent ban.

What actually separates the tools, then?

The inbox. Six of eight have some reply-management surface; Octopus CRM has none and Dux-Soup tells you to "follow up in LinkedIn." But feature-checkbox inboxes and an inbox-first product are different things, and this is the half of the funnel the category systematically underbuilds, which is also the half the AI SDR vendors are still hiring humans to cover. The send is a commodity. The reply is where the money is. A tool that gets a prospect to answer and then dumps you into LinkedIn's raw notification pile has handed you a part-time job, not a solution.

The send is a commodity. The conversation is the product. Price tools accordingly.

How should you actually buy one?

Grade any outreach tool on four questions its pricing page won't volunteer:

1. Does it run in the cloud? If it needs your browser open, it's a toy. (Three of eight still fail or half-fail this in 2026.)
2. Where is the vendor's denominator? Ask for one aggregate number with a sample size. Six of eight have never published one anywhere.
3. What happens after the reply? If the answer is "check LinkedIn," you bought half a tool.
4. Does it document limits, or just promise "safe"? The vendors that publish caps are the ones expecting your account to still exist next quarter.

And if you want the head-to-head reply-rate answer, run it yourself: one clean 500-lead list, one offer, matched send windows, 30 days, and count what happens after the reply, not just the reply. Demand the same artifact from anyone who claims they ran that test for you. The listicle audit found zero of ten could produce one.

Verdict

Two of eight tools are still browser-bound macros cosplaying as platforms. Six of eight wear "AI" as paint. Zero of eight put a real benchmark where buyers look, and the one vendor honest about LinkedIn's rules is the exception that indicts the category. The differentiator isn't send volume or how many times "AI" appears on the site. It's whether the vendor shows you denominators and respects that outreach is a conversation, not a broadcast.

The receipts don't lie, mostly because the category barely prints any.