The State of AI Visibility in UK Uniform & Workwear

Three months inside ChatGPT, Gemini and Perplexity's answers about who supplies the UK's uniforms, who they name, who they cite, and how often they get it wrong.

Between May and September 2026, SiteSignal tracked how ChatGPT, Gemini, Perplexity and Google AI Overviews answer questions about UK workwear and uniform suppliers. We logged every brand these models named in response to workwear and uniform queries, then checked where their answers came from and how often they got the facts wrong. The headline finding: this is one consolidated market, and a citation economy that runs almost entirely on a brand's own website. Here's what we found.

What this report is, and isn't

This isn't a market-wide sweep commissioned to survey every UK uniform supplier. It's built from SiteSignal's ongoing AI-visibility monitoring in this sector, aggregated across independent monitoring runs and every competitor brand the models named alongside our tracked terms. 285 overlapping brands appeared consistently across separate runs, which is why we're treating this as one shared, corroborated market view rather than a single narrow sample. In line with client confidentiality, we don't identify the specific accounts underlying this monitoring.

We ran every prompt against ChatGPT, Gemini and Perplexity daily, with Google AI Overview sampled where it appeared. May and September are partial months in this dataset (tracking began May 18, and September runs to the 13th at time of writing), so the leaderboard and trend figures below use the three complete months: June, July and August 2026.

Our Visibility Score is our own, disclosed formula, not a reproduction of any other index. For each month, it's an equal blend of appearance share (a brand's mention count relative to the month's most-mentioned brand) and rank quality (how early a brand tends to appear in a model's answer, scaled 0 to 100). A brand needed at least 15 tracked mentions in a month to qualify.

One limitation worth stating plainly: early in the tracking window, some brands were logged by domain (garmentec.co.uk) and later by name (Garmentec). We merged these programmatically for the roughly 25 highest-volume brands and verified the merges by hand. A handful of low-volume, long-tail entries may still be split across both forms.

Read everything below as one industry's case study, not a general AI-visibility rulebook. Every number describes UK uniform and workwear specifically: a small-basket, trust-led B2B purchase with a fragmented supplier base. Citation mix, funnel behaviour and hallucination rates would plausibly look different in a category like consumer electronics or healthcare. Treat our method as reusable and these numbers as local to this sector.

The market is smaller than it looks

Ninety-seven distinct brands cleared a meaningful visibility bar over three months. But the leaderboard barely moves. The same eight to ten names anchor every monthly ranking: Uniform Express, The Uniform Consultants, Workwear Express, Garmentec, Staff Uniforms, Field Grey, Kylemark and Simon Jersey. Beyond that group sits a long tail of 80-plus smaller suppliers. Most of them surface once and don't come back.

Ranked by tracked appearances across all platforms. Visibility Score is in the right-hand column.

The standout move of the quarter isn't a volume story at all. The Uniform Consultants held roughly flat on mentions across the summer, but its average rank position improved from 2.69 in June to 2.18 in August. That's the sharpest quality gain of any brand in the set, appearance count aside. Appearing often gets you noticed. Appearing first is what actually wins a "which supplier should I use" answer.

Biggest gains and drops, June to August

Measured on tracked-appearance volume across the three full months:

Studio 104 fell out of the top ten in August, replaced by Incorporatewear, which grew from 82 to 99 mentions, up 21%.

September: an early signal, read cautiously

September is a 13-day partial month at time of writing, which makes a straight comparison to a full month misleading in a specific way we checked for rather than assumed. Comparing each brand's share of that month's mentions, the same approach we use throughout this report, made almost every brand look like it had gained ground. That turned out to be an artifact of the shorter window: fewer days means fewer brands clear any given mention threshold, which mechanically inflates the share of whichever names remain. So for September specifically, we compare appearances per tracked day instead, since that isn't distorted by month length.

The one finding here that looks like more than noise: Garmentec's daily rate collapsed by more than two-thirds, a sharper single-brand reversal than anything else in three-plus months of tracking, top or bottom of the leaderboard. Meanwhile Dimensions and Murray Uniforms, both mid-table names all summer, are appearing more than twice as often per day in September, and the three brands anchoring the top of the leaderboard since June all cooled 11 to 24% on this measure, even while remaining September's top three in absolute terms. Thirteen days isn't enough to call any of this confirmed. It's worth checking against a full October read before treating it as real.

AI cites your own website, not press coverage

This is what AI actually cites when it answers a workwear question. We classified 44,627 citation events by source type. The result is stark:

Nine citations in ten trace back to a supplier's own website or an independent write-up. Reviews and forums, the sources most marketing teams chase first, barely move the needle in aggregate.

Inside "editorial"

Within independent citations, the largest single slice, 7,200 citations, is still content living on a brand's own site that a model treats as editorial in character: spec pages, buying guides. The next largest, 3,485 citations, is genuine third-party research and reference material. Press coverage and guest-posted content combined account for under 1,300 citations, under 3% of the total.

If you want to be the source an AI model reaches for, the evidence points at one place: your own site's depth, not a PR campaign.

Tone and quality of what gets cited

We scored every citation for sentiment, source authority and how directly reusable the content is for an AI answer. Three findings stand out:

• Neutral sentiment: 53.6%
• Positive sentiment: 46.0%
• Negative sentiment: 0.4%

Under half a percent of all citations carried negative sentiment. In three-plus months of tracking, AI answers in this sector essentially never cite a source to say something unflattering about a supplier. That's either a genuinely calm market, or a sign these models default to reporting what a supplier says about itself rather than surfacing criticism of it. Our data can't distinguish the two.

Content quality skews toward the usable middle: 64% of citations came from medium-authority sources (21% high, 15% low), and a third were rated highly reusable by the model doing the citing, meaning the content was structured clearly enough to lift a fact from directly, without interpretation. That's the practical target for a supplier's own spec and FAQ pages: not maximum polish, but maximum extractability.

Reddit went from nowhere to one of the biggest citation sources in three months

Reddit didn't drop. It's the fastest-growing source in the dataset. Going in, our working assumption was that forum citations, Reddit specifically, had been fading out through the summer. The data says the opposite, and by a wide margin. Total citation volume grew as our monitoring scope expanded (8,297 citations in June to 15,907 in August), so the fair comparison isn't raw counts but each platform's share of all citations that month.

By September (partial month), Reddit's share had climbed further still to 12% of all citations, bigger than reviews, social mentions and user-generated content combined. Facebook is the platform that actually collapsed, losing three-quarters of its relative share from June to August. Instagram softened on the same trend, down to roughly a third of its June share by August. LinkedIn moved the opposite way, climbing steadily and holding its gains rather than spiking and receding.

A caveat worth stating plainly: Reddit's and Wikipedia's jumps both land in August, the same month our monitoring pipeline's citation classifier was extended to catch a broader range of source pages. So part of this rise is almost certainly monitoring catching up to citations that were already happening, not the models suddenly discovering Reddit overnight. Facebook and Instagram's decline shows the opposite pattern: a steady month-over-month fade with no pipeline change to explain it away, which is why we're more confident calling that one a genuine shift rather than a measurement artifact.

AI won't say your name until someone's ready to buy

Brands are invisible at the awareness stage, and dominant at evaluation. We tagged every prompt by funnel stage. Splitting visibility by stage turns up the single sharpest pattern in the whole dataset.

At the awareness stage, category-education questions with no specific supplier in view, a tracked brand was named in zero of 355 prompts. That's expected: these are generic-knowledge questions, not shopping ones. What's not expected is the gap between consideration and evaluation. Consideration-stage prompts ("what should I look for," "what are my options") name a specific supplier only 8% of the time, while evaluation-stage prompts ("which one should I pick") name one 78% of the time, and once named, it's recommended essentially every time. In this market, models appear to hold back specific brand names until a query all but demands a decision, rather than seeding them earlier in the research process.

Practical read: a content strategy built around comparison and "options" pages may be reaching the stage where models are least likely to say a brand's name at all. The bigger opportunity looks like the decision-stage query itself.

Not every platform names names equally

Two things are worth separating here. First, Google AI Overview is a fundamentally weaker channel for this sector. It only rendered at all for 18% of the exact same query set, and even then named a supplier in just 7.6% of all attempts, roughly a fifth of ChatGPT's rate. Second, across every platform, a structured ranked list is the exception, not the rule (under 3% of answers everywhere). Almost all supplier recommendations here arrive as prose, naming one or two brands directly, rather than a numbered top five. A content or PR strategy built around "getting into the AI top 5" is optimising for a format these models mostly don't use, at least not for this kind of query.

Which platform is most likely to get you wrong

ChatGPT commits and is usually right. Gemini hedges. Perplexity commits and is often wrong. Splitting our fact-check results by platform shows the three models fail in genuinely different ways, not just at different rates.

Gemini rarely commits to a specific factual claim at all. Nearly three-quarters of checks came back "insufficient evidence to call," by far the highest of the three, and its mismatch rate (4.3%) is also the lowest, specifically because it so rarely states a checkable fact in the first place.

ChatGPT commits most often and is right most often: a 68.8% match rate against a 23.5% mismatch rate.

Perplexity commits almost as confidently as ChatGPT (only 4.4% "insufficient," the lowest hedge rate of the three) but is wrong or partially wrong on nearly half of what it states, 47% combined. For a supplier who cares which platform is likeliest to misrepresent them with false confidence, this dataset points squarely at Perplexity.

Where AI gets a supplier's facts wrong

SiteSignal fact-checks specific claims models make about individual suppliers against ground truth supplied by each brand. Across 4,903 checks run against two suppliers in this dataset, 47% matched, 17% were flat wrong, 12% were partially wrong, and the rest had too little evidence to call. But the average hides the real story, which is that hallucination isn't evenly spread across fact types. It concentrates almost entirely in one field.

We checked five fact types: business or brand name, headquarters location, industry or domain, founder(s), and privacy policy URL. Business name and headquarters are almost never wrong. A privacy policy link is wrong more often than it's right, at 58% of the time, combined across both suppliers studied.

The founder question is the odd one out. Of the two suppliers checked in depth, one had its founder correctly named 73% of the time. The other had the wrong founder named 63% of the time. That split, not a shared weakness across the board but a sharp divergence between two otherwise-similar businesses, is consistent with a model confusing similarly named companies rather than simply guessing at random. It's a concrete, checkable reason to keep a founder's name and history clearly and uniquely stated on-site.

Seven patterns that held across the quarter

1. Rank matters more than raw mentions. The Uniform Consultants proved this directly: flat mention volume, but the best average rank of any brand in the leaderboard by August (2.18) after starting the quarter mid-pack. Appearing often is necessary. Appearing first is what a "who should I use" answer actually rewards.
2. The market is narrower than it looks. Ninety-seven brands cleared the bar over three months, but the same eight names hold every top-ten position. A long tail exists, but it churns. Most of it appears once and doesn't return.
3. Citations reward depth, not distribution. PR coverage and guest posts together are a rounding error next to a brand's own site and independent research write-ups. Distribution campaigns appear to buy little visibility here. On-site depth appears to buy a great deal.
4. Hallucination is a field problem, not a brand problem. No brand in this dataset was "hallucinated about" broadly. Instead, specific fact types are unreliable everywhere (privacy policy links) while others are reliable everywhere (brand name, HQ), meaning the fix is field-specific, not a general trust problem with AI answers.
5. Where AI looks for opinion is shifting fast. Reddit went from a rounding error to one of the largest single citation sources tracked, inside a single quarter. Whatever caution applies to the measurement, the direction is unmistakable: a supplier with no presence in relevant Reddit threads is increasingly absent from a growing slice of AI-mediated opinion, not just search results.
6. Visibility is a decision-stage phenomenon here. Zero brand mentions at the awareness stage, 8% at consideration, 78% at evaluation. In this sector, a model appears to withhold a specific supplier's name until a query is nearly a direct request for one. Content aimed at earlier-stage "options" questions is competing for a share of voice that barely exists.
7. The three main platforms fail in different directions. ChatGPT states more facts and gets more of them right. Gemini mostly declines to state a checkable fact at all. Perplexity states facts almost as confidently as ChatGPT but is wrong close to half the time. "AI got a fact wrong" means something different depending on which of the three is doing the talking.

Before you apply any of this

This is a UK uniform and workwear study, not a general AI-visibility playbook. Every figure here, the citation mix, the funnel gap, the platform reliability differences, the hallucination rates, describes one sector: a low-frequency, trust-led, small-basket B2B purchase in the UK. Categories with different purchase psychology are likely to show a genuinely different shape, not just different numbers. A high-consideration, high-search-volume category, say consumer insurance or electronics, would plausibly show less concentration at the evaluation stage, since awareness-stage content there routinely names brands. A category with heavier UGC culture might already show Reddit at a mature, stable share rather than mid-surge. And a category with more litigated public facts, openly listed executives, public financials, could plausibly show a very different hallucination profile than this one's founder-name blind spot.

Use the method. Track your own category before assuming these numbers transfer to it. The one finding we'd expect to generalise furthest is the structural one: that a brand's own site, written with enough depth to be lifted directly into an answer, consistently outperforms distribution and PR spend as a path to AI visibility. Everything else here is a measurement of this specific market, at this specific moment, and should be treated that way.