OUTLINE OF CHAPTERS
Ask Perplexity, Gemini, Claude or ChatGPT to name a UK manufacturer of one of our client’s core products and they come up in every single answer: 91 out of 91 across August and September. That’s the first half of AI search for manufacturers, and it’s the half most firms stop at. Our tracking shows the second half, the specification, process and comparison questions engineers ask before they know who to call, is where the next shortlists are being decided.
Here’s what the data shows, and the page types we build to win each stage.
The client and how we measure
Our client is a UK precision component manufacturer supplying engineering customers across several sectors.
We track their visibility two ways:
- A weekly tracker of 18 commercial questions across Perplexity, Gemini, Claude and ChatGPT, run several times each because AI answers change from one run to the next.
- A full 40-question baseline (August 2026) covering the whole buyer journey: specification, problem, comparison, supplier and compliance questions, written in buyer language with no brand names. Every question was asked three times on Perplexity, ChatGPT and Google AI Overviews. (Gemini was planned too, but its runs failed on a usage limit, so it’s excluded.)
The two measure slightly different things. The tracker records whether an answer names the client. The baseline counts an answer if it names them or cites their website, because early-stage answers often use a page as a source without naming who wrote it. Either way, we check every mention is actually about them.
Stage one: the supplier questions are won
On the product lines that matter most to the business, AI search now names our client almost every time.
| Question (Aug to Sep 2026) | Answers naming the client |
|---|---|
| UK manufacturer of their main product line | 91 of 91 (100%) |
| UK manufacturer of their second product line | 70 of 70 (100%) |
| All six priority product lines | 372 of 442 (84%) |
Gemini named them in 164 of 164 answers on those questions. Perplexity named them in 133 of 144. On the broadest question we’ve tracked since May, which asks for the best UK companies in their field, their share of answers has risen from under half to over half.
This is the stage most manufacturers check. You type “[your product] manufacturer UK” into ChatGPT, see your name, and conclude AI search is sorted. For our client it genuinely is at this stage. But it isn’t where most buying journeys start.
Stage two: the questions engineers ask first
Engineers rarely open with “who makes this?”. They start with the problem: a part that distorts in production, a prototype that needs three revisions during testing, a tolerance their current process can’t hold. The answer frames which process gets specified, and the process decides which suppliers make the list.
Our August baseline measured that stage:
| Question type | Answers naming or citing the client |
|---|---|
| Supplier | 48 of 63 (76%) |
| Comparison | 11 of 72 (15%) |
| Specification | 10 of 90 (11%) |
| Problem and application | 9 of 90 (10%) |
| Compliance | 0 of 45 (0%) |
The same topic can score completely differently depending on how the question is asked:
- Asking which UK suppliers offer their manufacturing process with in-house finishing named the client in 9 of 9 answers. Asking what finishing options are available for that same process named them in 0 of 9.
- Asking which companies make one of their key components in the UK scored 9 of 9. Asking which process is best for making that component scored 0 of 9.
This isn’t a weakness in the client’s website. It’s two different jobs. Supplier questions reward being a credible, well-described manufacturer, and our client has that covered. Early-stage questions reward something else: pages written specifically to answer an engineer’s problem better than anyone else’s. In our experience, few manufacturers have built that layer yet, which is exactly why it’s the opportunity. It’s what we’re building for this client now.
Why this matters
Buyers are already doing this. In a July 2026 Semrush study of 622 US B2B professionals, 92% said AI had shaped their vendor shortlist, and 45% said it had done so significantly. A UK survey by PR agency Magenta Associates reported 66% of B2B decision-makers using AI tools to support supplier research.
If AI is shaping shortlists, the earlier questions matter as much as the final one. By the time an engineer asks “who makes this in the UK?”, they’ve often already chosen the process. The manufacturer that helped them choose it has a head start.
When a manufacturer tells me they already show up in ChatGPT, my first question is: for which question? Almost always it’s “who makes this”, and that’s the last thing an engineer asks.
Who wins the early questions today
When no manufacturer has written the answer, the engines don’t leave a gap. They fill it with someone else:
- “How is it made?” questions go to academia. Manufacturing-method answers cited ScienceDirect, Nature, PubMed Central and Royal Society of Chemistry papers. Useful to a researcher; no route to a supplier.
- “Which process?” and “what options?” questions go to whichever manufacturers have published explainer pages, several of them outside the UK.
- Compliance questions go to Wikipedia, certification bodies and whoever wrote it down. Answers cited Wikipedia, ISO and certification body NQA, quality-software vendors, and the few precision manufacturers that had published their own certifications and inspection processes.
That second point is the opportunity. A good comparison page gets to explain the trade-offs, in its own words, to an engineer who hasn’t picked a supplier yet.
Three things we’ve learned from tracking
Each engine behaves differently. Gemini named our client in every priority product answer we tracked. In the August baseline, ChatGPT was the hardest of the three engines to appear in at every stage. In the weekly tracker, ChatGPT and Claude both named the client in roughly half to two thirds of priority product answers. Checking one tool gives you a false picture, good or bad.
The word “UK” does a lot of the work. Asking for a UK manufacturer of their main product named our client in every answer. The same question with no location named them far less often. Local supplier questions favour UK firms; global ones are a bigger field.
Being a trusted source is the step before being named. On a broad question about their manufacturing process, Perplexity used one of our client’s pages as a source in 11 of 11 answers. The engine already trusts the content. The next step is making sure the company name travels with it.
Which page wins which question
Each question type needs a different kind of page. This is the map we work from.
| Question type | Page that wins it | What goes on it |
|---|---|---|
| Supplier | Capability page with a location | “[Part] manufacturer UK”, one per product line, with consistent company details everywhere you’re listed. The layer our client has already won, and the foundation for everything else. |
| Specification | Numbers in plain text | Tolerances, thickness ranges, minimum feature sizes and materials in a visible table or FAQ, not only in a PDF datasheet. Engines can only quote numbers they can find. |
| Problem and application | A page that starts with the engineer’s problem | “How do I prototype a part through testing without paying for new tooling every revision?”, not “our capabilities”. The problem, the options, and where your process fits. |
| Comparison | An honest process comparison | Including when your process is the wrong choice. Where our client had one, Perplexity cited their site in every run of three comparison questions. We’re now extending that across the other engines. |
| Compliance | The page nobody writes | Certifications, First Article Inspection Reports, traceability, RoHS and REACH. Asked late by quality and procurement teams; in our experience few manufacturers answer them directly online. |
Across all of them, put your name in the sentence that gets quoted. “At [your company] we can hold…” survives being lifted into an answer. “This process can hold…” makes you an anonymous source.
Run a quick version on your own business
You can do a rough check in an afternoon. Write ten questions an engineer would ask before they know your name, across the five types above. Ask each one three times in ChatGPT, Perplexity and Google. Count how often you’re named, and note who is cited when you’re not. If you want to know whether AI can read your site at all, our free AI website audit is a good first step.
If the supplier questions look healthy and the rest are quiet, you’re where our client was: the foundation is in place and the next layer is there to be won. Measuring it properly, question by question and engine by engine, is how every one of our AEO programmes starts.
If you’d like us to run the full baseline for you, get in touch.