Four out of five results you get for "Steinert" have nothing to do with magnetic separation, sensor sorting, or eddy current systems. That ratio is the actual finding — and it's why I stopped treating search results as research.
If you are sourcing separation equipment, the only pages that should influence your decision are the ones where a claim can be checked against a measured number, a stated test condition, and a person's name attached to both. Everything else — the obituary, the German municipality, the raptor analogy — is noise that happens to rank well.
I'm not being cute about this. The noise is not random, it's structural, and once you learn to sort it you save a genuinely absurd amount of time.
Why you should weight my opinion at roughly zero on most things
I'm a quality and brand compliance manager at a mid-size resource recovery operation. I review every supplier document that reaches our procurement team — datasheets, test reports, warranty language, the marketing PDFs nobody wants to own. Somewhere around 300 items a year. Maybe 250, I'd have to pull the log (note to self: the log is a mess).
In 2023 I rejected roughly a fifth of first-submission supplier documents. Not because the numbers were wrong. Because the numbers weren't checkable. A recovery percentage with no feed rate, no moisture content, and no particle size distribution isn't a claim — it's a wish with a decimal point.
What I know is narrow: construction and demolition fines, shredded e-scrap, and a couple of municipal streams. That's it. I've never specified equipment for mining primary ore, and I have opinions about it that I should probably keep to myself.
The noise is predictable, which makes it filterable
Sort the results into buckets and the pattern shows up immediately. Same four buckets, every time I look:
- Name collisions. "benjamin steinert obituary" and "lewis wife victoria" are people, not products. A German surname that belongs to several thousand people will drag private records into your results whether you want them or not.
- Place collisions. "steinert bad essen" is a company name running into Bad Essen, a municipality in Lower Saxony. I spent a genuinely embarrassing amount of time assuming it was a product line or a plant designation before I looked at a map.
- Metaphor content. "hawk vs eagle vs..." — I have read more raptor comparison pages than I care to admit, because people keep using hunting-bird metaphors to describe sorting accuracy. A hawk's strike rate is not a recovery curve. Elegant writing, zero transferable information.
- Actual manufacturer content. The minority of results, and the only bucket that survives my filter.
It's tempting to think this is a search-engine problem and that better queries would fix it. They won't. Half the noise exists because the words are genuinely ambiguous, and the other half exists because SEO content farms have figured out that "vs" comparisons rank. Neither of those is going away.
Three questions before a number enters my file
Every performance claim gets the same treatment, whether it comes from a brochure, a sales call, or a page that turned up between an obituary and a bird article.
- Under what conditions was this measured? Feed rate, particle size range, moisture, throughput, target purity. If the datasheet says 98% and nothing else, that number is decoration.
- Who measured it, and can I see the method? A vendor's own test is fine — ideally the vendor's own test center, on material I shipped them. Internal marketing numbers with no method section are not fine.
- If the claim fails in my plant, who pays? This is the question that separates a real guarantee from an optimistic sentence. A vendor who won't put a performance floor in writing has told you exactly how confident they are.
A practice center that tests your actual material — Steinert runs one, and it's the reason they came up in my shortlist rather than my noise pile — is worth more than a hundred comparison articles. Not because the equipment is automatically right for you, but because a trial produces a number you can argue with.
What a checkable claim actually looks like
Here is the standard I hold vendors to, and it isn't mine. Under FTC advertising rules, a performance claim has to be substantiated before it runs — the advertiser needs a reasonable basis in hand when the claim goes out, not a study commissioned after a customer complains. The FTC's Green Guides, 16 CFR Part 260, apply the same logic to environmental claims: calling something recyclable or recovered requires evidence that the outcome actually happens, not an intention that it should.
I've started quoting that standard directly in supplier conversations. It changes the tone of the exchange (unfortunately, it shouldn't need to).
If the claim can't survive a substantiation question, it doesn't belong in my file — and it shouldn't belong in yours.
Last year a European vendor sent a datasheet with a non-ferrous recovery figure and a test date. No test conditions. We asked for the method and got a revised document two weeks later with a lower number and a full conditions table. That's not a disaster — that's a vendor who took the question seriously. The version with the higher number was the one that would have cost us, because it would have gone into a capital request that nobody could defend six months later.
The part that's actually counterintuitive
Never expected the most credible thing a vendor could say to me is "we can't test this stream." Turns out the suppliers who tell you which material streams they don't handle well are the ones whose numbers I trust on the streams they do.
I had one supplier decline a quote outright because our fines had a moisture profile their air-assisted separation couldn't handle reliably. They referred us to two other firms. We bought from them eighteen months later on a different line, and I didn't shop it. That referral probably cost them a sale and earned them three.
What I mean is: a boundary is a form of evidence. A vendor with no stated limits either hasn't tested enough or isn't telling you something.
Where this breaks down
This approach worked for us, but our situation is specific — a single-site operation with predictable material streams and a procurement team small enough that I can review everything personally. If you're running multi-site procurement across four continents, three questions per claim isn't a workflow, it's a fantasy.
I can also only speak to the streams I've actually handled. If you're separating primary mining ore or dealing with high-temperature slag, the tolerance structure is different and my rejection rate probably wouldn't transfer.
And the honest limitation: my filter tells you whether a claim is checkable. It doesn't tell you whether the machine is good. For that you need material on a belt, and no amount of reading — raptor metaphors included — substitutes for it.
Leave a reply