Forty-seven of forty-nine UK engineering firms are visible only through their own pages

Every number below was counted from fifty individual records, one per firm. Where a record did not say enough to count something, it was left out and the exclusion is stated. Nothing here is estimated to fill a gap.

What was actually done, and what it does not show?

Fifty UK firms in industrial engineering, power electronics, renewables, energy infrastructure and precision manufacturing were each given three questions a buyer might type, written in UK phrasing and aimed at capability rather than category.

Each question was run once through a UK search index on 10 September 2026.

For each one the firms named in the results were recorded, along with whether the target firm appeared, which rival appeared most often, and whether the sources carrying the firm were its own pages or somebody else’s.

Two limits matter more than any finding, so they come first.

These were search-surface samples, not assistant answers. No ChatGPT or Perplexity screenshots were taken. What was recorded is what a UK search index returned, which is the material those assistants draw on rather than the answer they produce.

The published method screenshots three assistants; this sample did not. Anyone repeating one of these questions in ChatGPT will see something related but not identical.

Three questions per firm, not the five the programme uses. Three is enough to show whether a firm is reachable at all.

It is not enough to describe its share of an answer set, and no percentage in this article should be read as a share of voice. Each question ran once, on one date. Search results move with location, session and day.

Two further caveats. The prompts were written per firm by one person, using the same rules as the client battery, but nobody agreed them in advance and no firm was given a chance to challenge them.

And this began as a qualification pass to decide who to approach, not as a study designed in advance, which is why the questions vary in shape from firm to firm.

What this dataset does not claim

• What any assistant said. Only what a search index returned on one day.
• What any individual buyer was shown.
• That the fifty firms represent UK engineering as a whole. They were chosen because they fit one agency’s ideal customer profile.
• That a firm which appeared is well served, or that a firm which did not is badly run. Both statements would need more than three questions.
• Anything about a firm beyond what its public pages and the public results say.

Why is no firm named?

None of the fifty asked to be measured, and none has seen its record. The sample was taken to decide who to approach.

A public list of named firms described as invisible would be a claim about somebody’s business published without their reply, and that is not a trade this site is willing to make for a headline.

So every example below is described instead: a Nottinghamshire motor rewind workshop, a Bolton standby power firm, a national distributor. The firms themselves are in the private records, and any firm that asks can have its own one back.

How many firms appeared for their own buyers’ questions?

One of the fifty records is excluded from every count in this article. That firm is a Hampshire instrument distributor which was screened out of scope before any search ran: its three questions were drafted and never used, so no evidence exists for it.

Counting a blank as a zero would have made the headline number look worse than the evidence supports. That leaves 49 firms and 147 question checks.

Of those 147 checks, 85 returned the firm and 62 did not. Spread across firms:

Forty of the forty-nine were missing from at least one question a buyer in their own market would ask.

Six were missing from all three, including a Hampshire water-industry systems integrator trading since 1988, a Leeds grid connection contractor, a Greater Manchester switchgear maintenance firm, a Cheltenham battery storage developer, a Liverpool grid connections consultancy and a Wrexham control panel manufacturer.

The Wrexham firm is the sharpest single case in the set. It was absent from all three of its questions, including one that named its own county and town.

Who is carrying the firms that do appear?

This is the finding that holds up best, because every record carries an explicit line on it and the answer is nearly unanimous.

Of the 49 firms with evidence, 47 were visible through their own pages alone. Every result carrying them was something they had written about themselves: a service page, a product page, their own article, in one case their own social profile.

No directory entry, no trade article, no supplier register, no forum answer was recorded naming them.

Two had third-party corroboration. One is a Tyneside switchgear and transformer supplier, near the top of all three of its questions, where one of the results was a trade page written about the firm rather than by it.

The other is a Northamptonshire generator supplier, where an independent directory listing described the business. That is the whole list. Two out of forty-nine.

It is worth being precise about why this matters rather than overstating it. Self-description is how most of these results are built for everyone, rivals included. The point is not that these firms did something wrong.

It is that when an answer rests only on what a firm says about itself, there is nothing holding the firm’s place if a better-written page appears, and nothing to make an assistant treat the claim as established rather than asserted.

The firms in the sample that led their questions led them with their own pages too, which means their lead is made of the same material.

How often did a question return no UK specialist at all?

This one needs its counting rule stated, because it is easy to inflate. A question was counted here only where the record says in terms that no UK specialist, independent or provider ranked at all.

Questions where non-specialist sources merely dominated, with a UK firm still present somewhere on the page, were not counted.

On that rule, 5 of the 147 questions returned no UK specialist:

• A variable frequency drive repair and service exchange question with a warranty qualifier. Page one was almost entirely United States repair firms, plus a social post and a manufacturer’s warranty PDF.
• “Who can repair an obsolete industrial PCB with no schematics.” Page one was Reddit, YouTube, Facebook, overseas explainer blogs and two United States repair firms. The record calls it the clearest content gap of that firm’s three.
• A detuned harmonic filter capacitor bank question. Page one was overseas manufacturers and product directories, with no UK installer on it.
• A BS EN 50160 compliance and IEEE 519 assessment question. Page one held standards documents and databases, a manufacturer FAQ and one software vendor, and no UK service provider at all.
• An industrial control system cyber security and safety instrumented systems question. Page one was a job listing, regulator guidance, a university, a certification body, a journal database and two standards explainers.

A sixth is a genuine conflict inside its own record and is therefore excluded from the count rather than quietly folded in.

On a switchgear protection testing and preventative maintenance question, one part of the record states that no UK specialist ranked at all, and another part of the same record names two UK firms lower down the page dominated by United States contractors writing to American standards.

Both statements cannot be right. Until the run is repeated, that question is uncounted.

Five of 147 is a small number and should be read as one: these questions exist, and they are not everywhere. What they have in common is worth more than the count.

Each one is a question phrased the way an engineer would phrase it, with a standard number, a fault condition or a warranty term in it, and each one is answered by material written for a different audience entirely.

What kinds of source fill these answers?

Three source types could be counted reliably, because the records name them explicitly wherever they appeared. Counts are of records, so a source type appearing on two of a firm’s questions counts once.

A fourth group, the directories, buyers’ guides, trade registers and standards documents, appears repeatedly across the records and is described in too many different ways to count without guessing, so no figure is given for it.

The pattern is visible without a number: accreditation registers own the grid connection questions, an accreditation directory owns the headline calibration question, and an automation directory was the only source to recur across two questions for one control systems firm.

The recruitment finding is small and strange enough to be worth naming. On five firms’ questions, job adverts were sitting on page one of a question about who performs a service.

On one component-level electronics repair question, three job boards were padding the page. A buyer asking who repairs a thing is being shown people advertising to hire someone who repairs that thing.

Which capabilities were covered, and which were empty?

Firms were grouped by the capability line in their record. The buckets are small, some very small, so these are the sample’s shape rather than a verdict on a sector. Read the two-firm and three-firm rows with that in mind.

A thirteenth capability, test and measurement, held only the excluded distributor, so it has no row.

The motor rewind row is the one worth stopping on, because all three firms behaved identically. Each appeared on exactly one of its three questions, and in every case that one question was the one carrying its county or town.

On the national questions, “who can rewind a large industrial electric motor in the UK” and “electric motor rewind specialists UK”, all three were absent, and the names that came back were a different set from the local ones.

The grid connections row has a different cause. Six firms, eight appearances, and on almost every question the results were led by accreditation registers, distribution network operator pages and one or two large contractors.

One of those contractors was named on all three questions of three separate firms in the sample. Where a register ranks above every firm it lists, the register decides the shortlist.

Does adding a place name change the answer?

Ten of the 147 questions carried a town, county or region. Eight of those ten returned the firm. Of the remaining 137 questions, 77 returned the firm, which is 56%.

The more useful version of that number is who it applies to.

Four firms in the sample appeared on their place-named question and on neither of the other two: the three motor rewind workshops already mentioned, and a Bolton standby power firm which was absent from a data centre generator maintenance question and from a hospital and telecoms backup power question, but held positions one and three when Bolton was in the query.

Two firms were absent even from their place-named question: the Wrexham control panel manufacturer, and a north west England utilities contractor asked about network design and substation construction in its own region.

So a place name helps, and it is also a ceiling. A firm that only exists in the answer when a buyer names its town is invisible to every buyer who does not, and buyers with a national supplier list rarely do.

Is anyone actually winning these questions?

Less often than the firms themselves assume. In 17 of the 49 record sets, no rival firm appeared on more than one of that firm’s three questions. The space is not being taken by a dominant competitor; it is not being taken at all.

In the other 32, a rival recurred, and in several cases the recurring name was not a competitor in any commercial sense: an equipment manufacturer’s own service pages, an accreditation register, or in one drives case the manufacturer whose products the firm services, present on all three questions and outranking every independent.

An MD who believes a specific rival is beating them online is, on this evidence, more often wrong than right.

What does this mean for a firm reading it?

Three things follow, in the order they are worth acting on.

First, the odds are that you are missing from at least one question your buyers ask. Forty of forty-nine were, and none of those firms knew it before the check. This is cheap to find out and expensive to assume.

Second, if you do appear, look at what is carrying you. On this sample it is almost certainly a page you wrote. That is a position resting on one leg.

The firms that will be hard to displace are the ones with a page of their own and something independent naming them, and almost nobody in this sample had the second thing.

Third, look at the question you would most like to win and ask what the current answer is made of.

If it is built from manufacturers, distributors, standards documents, overseas firms or job adverts, the answer is not being contested by a UK specialist, and a straightforward page written by an engineer can take it.

Two ways to see your own version of this. The free five-prompt snapshot is the same check run properly on your firm: five questions, three assistants, dated screenshots, back within three working days, no call required.

The fixed-fee audit is the longer version, with the competitor teardown, the technical findings and a roadmap. The snapshot is enough to tell you whether the audit is worth buying, which is what it is for.

How we would like this cited

KS Media, “Forty-seven of forty-nine UK engineering firms are visible only through their own pages”, 10 September 2026, https://ksmedia.marketing/insights/uk-engineering-ai-visibility-2026-09.

Quote the numbers with the sample date and the method attached: a search-surface sample of three questions per firm, run once through a UK search index on 10 September 2026, not a census of assistant answers.

Attribution and a link is all that is asked. If a figure here is later corrected, this page is corrected and the change is dated on it.