The deal you never knew you lost
Most of the deals a sales team loses are visible. There was a meeting, a sample, a quotation, a pilot — and then a competitor was chosen, or the project was cancelled, or the buyer went quiet. You can analyse those losses, argue about them in pipeline reviews, and learn from them.
There is a second category of lost deals that never appears in any CRM. The engineer had exactly the problem your product solves. They researched it, built a shortlist, contacted two suppliers and qualified one. You were never on the list, so you never heard about it. No lead, no lost opportunity, no post-mortem. Just a design-in that went to someone else.
This category has always existed. What is changing is how the shortlist gets built. A growing share of early technical research now starts not with a search engine and ten blue links, but with a question typed into ChatGPT, Claude, Copilot, Gemini or Perplexity. The answer comes back as a few paragraphs, often with a handful of named suppliers. If you are not in that answer, you are not on the list — and you will never know.
What the data actually says
There is a lot of noise around this topic. Agencies selling "AI visibility" services publish headline numbers claiming that nine out of ten buyers now research with ChatGPT. Most of those figures come from small, self-selected samples or from software buyers, who are not representative of engineers specifying components. The more reliable data tells a more nuanced — and, for OEM suppliers, more useful — story.
- Business buyers have adopted generative AI broadly. Forrester reported in January 2026 that 94% of business buyers use AI somewhere in their buying process, up from 89% a year earlier, and that buyers naming generative AI or conversational search as their most important information source outnumber those naming any other single source by two to one.
- Buyers engage sellers late. 6sense's 2024 Buyer Experience Report found that buyers are nearly 70% of the way through their purchasing process before engaging with sellers, and that they initiate first contact more than 80% of the time. Whatever shapes the shortlist happens before you are in the room.
- Engineers are slower — but moving. The 2026 State of Marketing to Engineers report by TREW Marketing and GlobalSpec, based on more than 1,000 technical professionals, found that 69% of technical buyers now use generative AI during the purchasing process. The share who never use it to evaluate vendors or products fell from 42% to 31% in one year, and the share who use it often almost doubled, from 6% to 11%.
- Engineers don't trust it — so they verify. In the same study, engineers rated their trust in AI-generated answers at 4.7 out of 10, and only 10% rated AI tools as very or extremely trustworthy. Among those who saw AI summaries in search results, 69% read the summary and then kept looking through the results.
- Search is losing share, not disappearing. Gartner predicted in February 2024 that traditional search engine volume would drop 25% by 2026 as AI chatbots and virtual agents take over part of the research. The direction matters more than the exact number.
Put together, this is the important insight for deep-tech OEM markets: engineers use AI as a shortcut for discovery, and then verify what it tells them against sources they trust. That means AI doesn't make the final decision. It decides which suppliers get verified in the first place.
Why this is a sales problem, not just a marketing problem
It's tempting to file this under "website and SEO" and hand it to marketing. That would be a mistake, because the consequences show up directly in sales performance.
When a supplier drops out of the early research stage, three things happen to the sales team. First, fewer qualified inbound opportunities arrive, and the ones that do arrive are later in the process, when requirements are already written around someone else's product. Second, the opportunities that do come in increasingly carry an AI-shaped frame — a comparison, a set of assumptions, sometimes outdated or simply wrong information about your product — that the salesperson has to discover and correct. Third, the win rate on the remaining deals looks normal, which hides the problem: you are converting the same share of a shrinking pool.
Classic sales effectiveness work focuses on what happens after first contact: qualification, stakeholder mapping, pilots, proposals. All of it assumes the buyer found you. In a market where the first filter is an AI answer, positioning becomes part of the sales process — the very first stage of it.
How AI decides which suppliers to mention
Generative assistants don't rank websites the way a search engine does. They synthesise an answer from what they have learned and, increasingly, from pages they retrieve in real time. Several things make a supplier more likely to appear in that answer:
- Content that matches how the problem is described. Engineers don't ask "best InGaAs camera". They describe a problem and its constraints: wavelength, speed, size, environment, integration, cost. Suppliers whose content explains how they solve that problem are far easier for an AI to connect to the question than suppliers whose content only lists product specifications.
- Information an AI can actually read. Key specifications locked inside PDF datasheets, configurators that need a login, or pages built entirely in JavaScript are harder for AI systems to use than plain, well-structured web pages.
- Independent confirmation. AI answers lean on what multiple credible sources say. Application notes in trade publications, conference papers, standards work, distributor listings and industry association content all reinforce the association between your name and a problem.
- Specificity and evidence. Academic work on what its authors call generative engine optimisation — a 2023 paper from Princeton and collaborators, accepted at KDD 2024 — found that content with clear statistics, quotations and cited sources could increase visibility in generative engine answers by up to 40%. The specific techniques will evolve, but the principle is stable: precise, evidenced content gets used; vague claims don't.
There is also a simple technical point that many industrial websites miss: if your website blocks the crawlers used by AI providers, you are making yourself invisible by choice. That may be a deliberate decision for some companies, but it should be a decision, not an accident.
A photonics example: the SWIR camera that never made the list
Consider an illustrative case, simplified but built on mechanics I see in the market. An engineer at a semiconductor equipment manufacturer needs to inspect bonded silicon wafers for voids at the interface. They know visible light won't penetrate silicon. They open an AI assistant and write something like: "What's the best approach to detect voids in bonded silicon wafers inline, at high throughput, in a compact inspection module? Which camera and light source suppliers should I consider?"
The answer explains that silicon becomes transparent beyond roughly 1.1 µm, recommends short-wave infrared imaging with an InGaAs sensor and a suitable illumination wavelength, discusses resolution and frame rate trade-offs, and names a few camera suppliers.
Now imagine two specialised SWIR camera companies with comparable products.
Supplier A has an excellent product and a website built around it: a product family page, a downloadable PDF datasheet for each model, a contact form and a news section with trade-show announcements. Nowhere does the site explain wafer inspection, bonded-wafer voids or the throughput and resolution trade-offs an engineer worries about. Its one application note on the topic was presented at a conference years ago and lives behind a registration form.
Supplier B has a similar product, but publishes open application pages written the way an engineer asks the question: "Inspecting bonded silicon wafers with SWIR imaging", with the physics, recommended wavelengths, achievable resolution at given working distances, typical frame rates and the integration constraints of an inline module — including, honestly, the cases where its camera is not the right choice. Key specifications are on the page in plain text, not only in the PDF. Its engineers co-authored an application article in a trade publication, and the company appears in an industry association's supplier directory under the right application keywords.
When the engineer asks the AI, Supplier B is far more likely to appear — not because its camera is better, but because its knowledge is visible, specific and confirmed by others. The engineer then verifies, exactly as the engineering survey data suggests: they read Supplier B's application page, download the datasheet, check the trade article and ask for a quote. Supplier A doesn't lose the deal in a competitive evaluation. It loses it before the evaluation exists.
The reverse case is just as instructive. Imagine Supplier A appears in the answer, but the AI describes a product generation it discontinued three years ago, because that is what most of the content on the web still talks about. The salesperson who receives the eventual inquiry — if one arrives — starts the conversation by correcting a misconception they don't even know exists.
What OEM suppliers should change in their communication
None of this requires abandoning what already works. Trade shows, application engineering, technical publications and relationships remain central in OEM markets. What changes is how deliberately you shape the knowledge that exists about you.
- Write for problems, not just products. Build application pages around the questions your buyers actually ask, including their constraints. One deep, honest application page is worth more than ten product announcements.
- Make technical facts readable. Put the specifications that matter in plain text on your pages, not only in PDFs. Keep naming consistent across products, generations and documents.
- Be precise and evidenced. Use numbers, test conditions, measured results and references. Avoid marketing superlatives that an AI can't use and an engineer won't believe.
- Earn independent mentions. Application articles in trade publications, conference papers, standards and association work, distributor and integrator content all reinforce your association with a problem. Your own website is necessary but not sufficient.
- Keep the record current. Retire or clearly label outdated products and documents. What is on the web today is what AI will repeat tomorrow.
- Check your technical access. Decide consciously whether AI crawlers can read your site, and make sure key pages don't depend on logins or scripts to show their content.
What the sales team should do differently
Sales leaders have a role here too, and it starts with information the sales team already has access to.
- Ask how the buyer found you — and what they were told. "What did your research tell you about the options?" is now a qualification question. The answer reveals the frame you're being compared in and any misconceptions you need to address.
- Run an AI visibility check every quarter. Write the ten or twenty questions your best customers would have asked at the start of their projects, run them through the main AI assistants, and record who gets mentioned, how you are described and what is wrong. It takes an afternoon and is more revealing than most market reports.
- Feed what you learn back to marketing. Every recurring question from prospects, every misconception, every lost deal where the buyer "never considered" you is input for the next application page.
- Track visibility as a leading indicator. Pipeline and win rate are lagging indicators. Your share of mentions in AI answers for the problems you solve is an early warning that shows up months before the pipeline does.
The honest conclusion
Generative AI won't decide which supplier wins an OEM design-in. Engineers verify, test and qualify; relationships, application support and supply reliability still decide the outcome. But AI is rapidly becoming the place where the list of candidates is formed — and in deep-tech OEM markets, not being on the list is the most expensive way to lose a deal, because you never see it happen.
The suppliers who adapt early won't be the ones with the biggest marketing budgets. They will be the ones who take the technical knowledge they already have — the application expertise that wins deals once a conversation starts — and make it visible, specific and verifiable before the conversation starts.
FAQ: AI search and OEM supplier visibility
Are engineers really using ChatGPT and other AI assistants to find suppliers?
Increasingly, yes, though more cautiously than other buyers. The 2026 State of Marketing to Engineers report found that 69% of technical buyers use generative AI during the purchasing process, while trust in AI answers remains low at 4.7 out of 10. Engineers use AI for discovery and then verify on vendor websites and trusted publications.
Is this different from traditional SEO?
It builds on the same foundations — useful, well-structured, credible content — but the emphasis shifts from ranking individual pages to being the source an AI uses when it explains a problem and names the suppliers who solve it. Specificity, evidence and independent confirmation matter more than keywords.
How can a supplier check whether AI assistants mention it?
Write the questions your target customers ask at the start of a project, run them regularly through the main assistants, and record which suppliers are mentioned and how they are described. Repeat quarterly and compare the results over time.
What is the first thing an OEM supplier should change?
Take the two or three applications that generate most of your revenue and publish one open, detailed application page for each — written around the customer's problem and constraints, with the key technical facts in plain text.