How we check which test an employer uses.

What we checked

The question is narrow: for each UK employer with a structured early-careers intake, which publisher runs each automated test step in the 2026/27 cycle? That means applications opening in 2026 for roles starting in 2027. An automated test step is an online test, a recorded video interview, a game-based test or a coding test.

We built the list of employers from graduate-employer rankings, student careers sites and each firm’s own UK careers pages. It covers 312 employers in 11 sectors, from investment banks, trading firms and asset managers to insurers, accountancy firms, consultancies, regulators and fintech firms. Each programme (spring week, internship, placement, graduate scheme, apprenticeship) is recorded separately, because one employer can use different tests for different programmes.

How each employer is checked

The research was carried out by AI research agents, following a fixed written method, between 10 and 12 October 2026. The sources they used are recorded with their links, and our employer tables link the newest one, so you can check it yourself. Where an employer’s pages blocked automated access, our founder, Nicholas Saunders, checked them by hand in a browser.

Every employer goes through steps 1 to 5, in the same order. Step 6 re-checks random samples after the fixes.

1. 01

   A first research pass.

   One research pass follows a fixed search routine: the employer’s own careers pages, FAQs and candidate guides; the test publisher’s customer stories; university careers services; dated candidate accounts; and prep sites last. For every automated test step it records the publisher, a short quote, the link, and each source’s date, or “undated” where the page shows none.

2. 02

   A second pass, done blind.

   A second, separate pass researches the same employer from scratch. It never sees the first pass’s answer, so it cannot simply agree with it.

3. 03

   A third check re-opens every source.

   A third check opens every link both passes cite. It confirms that each quote is really on its page and notes the date of the evidence. A quote it cannot find is struck out, along with anything resting on it.

4. 04

   Settling the differences.

   Where the two passes disagree, each difference is settled from the evidence that survived the third check, re-opening the deciding pages again where needed. The mark is the strongest one that evidence allows, and no stronger.

5. 05

   Quality check of every record.

   Every record was checked again, re-opening its sources. Errors were counted before they were fixed: before the fixes, 2.4% to 2.7% of the claims checked had a material error.

6. 06

   Fresh samples after the fixes.

   A new random sample of 15% of the second-round records was then checked once more: 5 of its 726 claims had a material error, 0.7%. A smaller re-check of 12 first-round records found 3 in 680, 0.4%.

A material error means a publisher that is wrong or has no support, or a mark that is too strong or too weak. Counting smaller slips too, such as a quote’s punctuation, a page date or a source filed under the wrong type, the rate in the same sample was 8.7%. Those smaller slips do not change the publisher or the mark a page shows.

What the marks mean

We label each employer fact with one of three marks. The mark is about the publisher’s name, not about whether the test step exists.

How sure we are: C Confirmed by the employer or the test publisher R Reported by other sources NP Not published anywhere we could find How we check

Older evidence is never shown as current. A step shows as Not published, with “previously reported” beside it, when the only evidence naming a publisher is from before the 2025/26 cycle (which began around August 2025), is undated, or is contradicted by newer pages. The year, or “earlier cycle”, appears beside it. PwC’s game-based test is an example: a June 2025 page names Arctic Shores, but three 2026 sources say PwC UK has moved to SHL. That tells you what an employer used before without claiming it still does.

Here is how the marks look in an employer table. These five rows are examples; each guide carries the employers it covers.

What counts as evidence

Sources are ranked. The rank of the best surviving source sets the mark.

• The employer’s own material: its careers pages, FAQs and candidate guides. Or the test publisher’s own customer story naming the employer. Only these can make a step Confirmed.
• Independent reports: university careers services and dated candidate accounts on forums and review sites. These can make a step Reported. A forum post with no date does not count.
• Prep sites come last. They sell practice tests, so they count as reports, never as confirmation. Two prep sites that copy each other count as one source, and several prep sites agreeing does not lift a step above “one source” unless they cite a dated candidate account.
• Job adverts copied onto jobs sites count as reports, not as the employer’s own page.
• An invitation email’s sender domain, or the test’s login address, names the publisher. When a source describes one, it counts at the strength of that source: the employer’s own guide makes it Confirmed, a candidate’s account makes it Reported.
• A page we cannot open is not used, however often it appears in search results.

A practice link is not proof. PwC’s assessment and selection page tells candidates to “have a go at taking the SHL practice tests”. That page sits within PwC’s student careers pages; it blocks automated access, so our founder read it by hand on 12 Oct 2026, and its exact address was not recorded. Several prep sites also name SHL. We still show PwC’s online test as SHL, Reported, one source. Suggesting a publisher’s practice tests is not the same as saying that publisher runs your test.

What we do not do

• We do not guess. If no source meets the rules, the answer is Not published. An honest gap beats a confident wrong answer.
• We do not name a publisher because a test “looks like” one. A test’s style, or a prep site calling it “SHL-style”, is not evidence.
• We do not show old evidence as current. It appears only as “previously reported”, with its year.
• We do not pick a side when sources conflict. Both names appear, marked “sources conflict”.
• We do not answer from memory. Every employer fact traces to a source opened during the research.

Where the evidence runs out

Across the 1,480 programmes we recorded there are 1,715 automated test steps. 1,323 of them have no published provider: about three in four. For each of those, no source met our rules after the full search, so we show Not published rather than fill the gap.

Looked at by employer, we can name a publisher for at least one test step at 94 of the 312 employers. Within one employer, programmes can differ: a spring week may use a different test from the graduate scheme. Our tables say so where that happens. An employer can also change publisher between years, which is why each employer table shows the date of its newest evidence, or says undated when that source has none.

For your own application, check the invitation itself.

When you know the publisher, the free practice tests page has one free test for each of SHL, Aon and Cappfinity. The test providers compared page sets out how the publishers’ tests differ.

When we check again

Every record is due for a re-check by 1 July 2027. Each employer table shows the date of the newest evidence for each employer, or says undated, so you can see how fresh it is. If you find an error on any page, email nicholas@acumenlogic.co.uk and we will look at it.

How we build and check the practice tests themselves is a separate process, set out on how we work.