Technical & Structure · 9 min read · July 15, 2026
How technical procurement now pre-selects suppliers with AI
Technical procurement in mechanical engineering now asks ChatGPT, Gemini, or Perplexity before a request for quotation ever goes out. Someone searching for "supplier for hardened gears module 4" or "special-machine builder for assembly automation" gets handed a short list of named candidates. If your shop isn't on it, you never see the inquiry — no matter how good your production actually is.
The pre-selection happens before you ever hear about it
Supplier search in mechanical engineering used to run through trade fairs, existing vendor lists, and the colleague who 'knows someone there'. Today the technical buyer opens an AI window first. The question is concrete: 'Which German manufacturers can supply milled parts from 1.2379 at 60 HRC in small batches?' The AI doesn't hand back ten blue links — it names three to five specific companies, with reasoning attached. That pre-selection decides who gets an inquiry at all and who doesn't.
The tricky part is that it's invisible. You won't notice it happening. There's no click, no form submission, no analytics trail. If your company isn't named for 'supplier for hydraulic blocks, batch size 50', you're simply missing from the shortlist — and the buyer never even learns you exist. That absence leaves no trace in your data, which is exactly what makes it dangerous for new business.
Ask yourself honestly: when a design engineer at an OEM in Baden-Württemberg searches today for a partner for precision turned parts, does your name come up in the AI's answer? Most mechanical engineering firms genuinely don't know, because they've never tried it. That's the first step — ask the question yourself and see who gets named instead of you.
Why the AI names these particular companies
Language models don't guess — they draw on text they've found across the web. For mechanical engineering that means technical data sheets, reference reports, trade articles, association directories, and company profiles with concrete production parameters. A company that states on its own site 'We mill tool steel up to 62 HRC, components up to 1200 mm edge length, tolerances down to IT6' hands the AI exactly the building blocks it needs to recommend them.
Whoever leans on marketing language instead — 'highest quality', 'decades of experience', 'your strong partner' — gives the machine nothing to work with. Sentences like that are empty for a language modelbecause they could describe any business. The AI needs materials, processes, dimensions, batch sizes, certificates, and industries. The more concrete your digital footprint, the sooner you get matched to the right inquiry.
On top of that comes the role of third parties. When a trade portal, a procurement directory, or a supplier catalogue describes your business using the same hard facts, that reinforces the match. The model sees consistent information from several independent sources and gains confidence in it. Visibility in AI answers isn't a solo effort — it's a web of consistent mentions.
The concrete questions procurement asks today
To understand where you need to be visible, look at the real search queries. In mechanical engineering, they read as technical and specific. Examples: 'Who welds stainless steel frames to EN 15085 CL1?', 'Contract manufacturer for laser-cutting mild steel sheet up to 25 mm near Stuttgart', 'Provider for nitriding transmission shafts with DIN certification', 'Special-machine builder for final assembly of electric motors with test bench'.
What stands out is that these questions almost never search for a product alone — they search for a combination of process, standard, material, and region. You need to cover exactly those combinations on your website. If you don't mention EN 15085 CL1 anywhere, you'll never surface for that question, even if you weld those seams every day.
Test this yourself with your ten most important services. Phrase each one as a buyer's question and put it to ChatGPT, Perplexity and Gemini. Note whether you're named, who's named instead, and which wording the competition uses. That list is the most honest market analysis you'll get — and it costs about an hour.
The difference between a Google ranking and an AI mention
Many mechanical engineering firms assume a good Google ranking is enough. It isn't anymore. On Google, the buyer clicks through the results and forms their own opinion. With AI, the model does the pre-selection and hands over only the result. The intermediate step — where a good website could still save you — disappears. Either you're in the answer, or you don't exist for that inquiry.
On top of that, AI answers often arrive without a visible source. The buyer reads 'For hardened gears in small batches, suitable options include Company A, Company B, and Company C' and moves on from there. They rarely check why those three were chosen. The trust your website used to have to earn is now granted by the model instead — but only to the companies it names.
This doesn't mean classic search engine optimization is dead. Both channels rely on similar signals: clear structure, concrete facts, evidence. The difference is in the preparation. For AI, you have to state your capabilities so they can be quoted in a single sentence. That's the core of Generative Engine Optimization.
How to make your production machine-readable
The most important lever is your own website. Build a dedicated page for every manufacturing process and describe it in hard numbers. Instead of 'We offer machining', write: 'CNC milling, 5-axis, components up to 800 x 600 x 500 mm, materials from aluminium to tool steel 1.2379, tolerances down to IT6, batch sizes from 1 to 500 pieces.' Specs like these are directly usable by the AI and match cleanly against a buyer's need.
Spell out standards and certificates in plain text: ISO 9001, IATF 16949, EN 1090, DIN EN ISO 3834. Name the industries you produce for — drive technology, packaging machinery, machine tools, conveyor technology. Build an honest reference list with concrete tasks, without exposing customers who'd rather stay unnamed. Every one of these entries becomes a docking point for a later AI recommendation.
Technically, structured markup helps machines read your specs unambiguously. A clear heading structure, clean tables of production parameters, and a well-maintained FAQ section do double duty — they help the human reader and the language model at the same time. Consistency is the key: the same numbers and terms everywhere, so no contradiction creeps in.
Consistency across every source
A language model gets suspicious when it finds contradictory information. If your profile on a supplier portal says 'sheet up to 15 mm' but your website says '25 mm', the AI can't tell which is right — and when in doubt, it leaves you out. So check every platform where your business appears and bring the technical figures into line.
This covers industry directories, procurement platforms like Wer liefert was, association listings, your Google Business Profile, and trade portals. Everywhere, your company name, location, processes, and core parameters should match exactly. This kind of consistency is unglamorous, but it's one of the strongest trust signals you have. It costs mostly diligence, hardly any money.
Also watch for staleness. When you bring a new 5-axis machine or an additional hardening process online, update it everywhere. Outdated information leads the AI to recommend you for work you no longer do, or to overlook you for capabilities you've since added. A fixed review cycle — twice a year is enough — keeps your digital footprint clean.
Evidence and third parties who vouch for you
Your own website is necessary, but it isn't enough on its own. Language models weight statements more heavily when independent sources confirm them. In mechanical engineering, that means trade articles in industry media, talks at technical conferences, mentions in case studies from machine manufacturers or material suppliers, and entries in reputable supplier catalogues. Every genuine mention improves your odds of being named.
An effective and often underused route is a joint reference story with a customer or tooling partner. When a manufacturer of machining centres presents you as a user achieving tight tolerances with their equipment, that's a strong, credible signal. Content like that tends to get linked and quoted, and from there it works its way into the models' knowledge.
Be patient and honest about this. It's not about flooding the system with volume — it's about building a coherent picture over months. A business that shows up online consistently, backed by facts and confirmed from multiple angles, becomes the obvious recommendation over time. That's real work, but it's exactly the kind of work competitors without deep technical knowledge tend to skip.
What you can concretely do this week
Start with the visibility test. Take your ten most important services, phrase them as buyer questions, and put them to three AI services. Write down where you're missing and who shows up instead. That gap is your action plan — it shows you, in black and white, which inquiries you're invisible for today even though you're fully capable of doing the work.
Then rewrite your most important service pages so each one speaks in hard facts: processes, materials, dimensions, tolerances, batch sizes, standards, industries. Add an honest FAQ that answers exactly the questions procurement is asking. In parallel, check your listings across portals and directories for contradictions and align them. None of this needs a big budget, but all of it has a direct impact.
Treat AI visibility as an ongoing discipline, not a one-off project — the same way you maintain your machine park. Check twice a year whether you show up for your core questions, and follow up wherever you're missing. Mechanical engineering runs on trust, and increasingly that trust gets formed in the AI's answer, long before the first phone call.
Common questions
We produce almost exclusively to drawing for existing customers. Is AI visibility even worth it for us?
Yes, especially then. Existing customers get acquired, switch suppliers, or relocate procurement. When a new buyer looks for a second source for your drawing parts, they ask the AI. If you're not present there with your concrete processes and tolerances, the inquiry goes to a competitor instead. Visibility protects your future business without turning you into a marketing operation.
Do we have to make confidential production details or customer names public for this?
No. This is about technical capability, not trade secrets. You can name processes, materials, dimensional ranges, tolerance classes, standards, and industries without giving away a single job. References can be anonymized — 'transmission shafts for a drive technology manufacturer', for example. That level of detail is more than enough for the AI to match you to the right inquiry.
How often should we check whether the AI recommends us correctly?
At least twice a year, and always after a major change to your machine park or service offering. The models get updated regularly, and your competitors are working on their own visibility too. A fixed cycle with your ten core questions shows you whether you still appear, whether the facts cited about you are correct, and where new gaps have opened up. The check itself takes under an hour.
Read on