We score 100 UK employer brands every month on exactly this question for The SetpointHQ Index, probing ChatGPT, Claude, Gemini and Perplexity with real candidate questions. The pattern is consistent: on the 1 September 2026 run the median brand scored 27.5 out of 100 on citation, the measure of whether assistants actually name you. Here is what separates the brands that get named from the brands that get skipped. It is the same question the pillar guide to employer brand AI visibility opens with, answered as a method rather than a diagnosis.

The short answer: 7 steps

1. Add Organization and JobPosting structured data. Machine-readable markup is the single strongest predictor of citation in our data.
2. Let AI crawlers in. Check your robots.txt and firewall rules for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. The careers site access and schema guide covers how to check both of these against your live site.
3. Fix your knowledge graph entry. A correct Wikidata entity with your official website attached anchors everything else.
4. Publish hard numbers on your careers pages. Salaries, headcounts, progression stats. Assistants quote specifics, not adjectives.
5. Answer the questions candidates actually ask. Pages shaped as questions and answers get lifted into answers far more often than brochure copy.
6. Use 1 consistent brand name everywhere. If your careers site, LinkedIn and Wikidata disagree about what you’re called, the machines hedge by omitting you.
7. Measure it monthly. Ask the 4 assistants the questions your candidates ask, and track whether you appear. What gets measured gets cited.

Why famous brands are invisible

Fame does not transfer. The machines know these companies in detail and still skip them, because knowing who you are and having something to say about you are different problems. Across the September cohort the median knowledge graph score is 89.4, so recognition is close to universal, while the median schema score is 35.0 and 21% of brands score zero on it outright.

The mechanism matters. When an assistant assembles an answer about employers, it retrieves and synthesises sources it can parse and trust. A careers site with structured job data, quotable facts and a clean entity record is usable raw material. A beautiful brand campaign in a JavaScript app the crawlers cannot render is, to the machine, silence.

It varies more by sector than most talent teams expect, and not in the direction brand budget would predict.

Consumer tech and healthtech sit at the bottom, which surprises people, because these are companies that build software for a living and could fix their own markup in an afternoon. Professional services and legal sit at the top, largely because their sites are text-heavy, server-rendered and full of named people with verifiable credentials. The sectors hiring hardest in volume, healthcare at 53.8 and logistics at 50.6, both sit below the all-sector median, which is a problem when those are precisely the sectors where a candidate is most likely to ask an assistant who is worth working for.

The UK retail run in July 2026 showed the same split inside a single sector: 6 of the 10 biggest retail employer brands scored 97 or higher on knowledge graph presence while the same cohort’s median citation score was 27.5, and 3 of the 10 scored 25 or below on schema. The full breakdown of that cohort, including the 2 brands at either end of it, sits in the GEO guide.

The 7 steps in practice

1. Structured data first

Organization markup on your corporate root, JobPosting markup on every live role. This is the layer employer brands fail hardest on, and it is also the cheapest of the 7 steps. Nearly a quarter of the cohort has none at all.

2. Crawl access

We regularly find employer brands whose security rules block the exact crawlers they need. If GPTBot cannot fetch your careers page, nothing downstream can save you. Check the 4 by name rather than assuming a blanket rule is friendly.

3. Knowledge graph hygiene

Assistants disambiguate brands through knowledge graph entities. A missing or wrong Wikidata record means your citations can leak to a similarly named company. This is the one dimension most brands already pass, so a failure here is unusual and worth fixing immediately.

4. Quotable evidence

Assistants prefer sources with specifics. “Competitive salary” is invisible. “Starting salary £26,500, median time to first promotion 18 months” is quotable, checkable and attributable to you rather than to the sector.

5. Question-shaped pages

Candidates ask assistants questions. Pages with clear headings, direct answers and numbered lists match the retrieval pattern. This page is built that way deliberately.

6. Entity consistency

Same name, same domain, same description across your site, LinkedIn, Glassdoor and Wikidata. Inconsistency reads as uncertainty, and uncertain sources get dropped rather than hedged.

7. Measurement

AI visibility moves as models and indexes update, so a one-off fix decays. The brands climbing our Index treat it like SEO circa 2010: a discipline, not a project. We rescore the full cohort monthly for exactly that reason.

Where to start

If you want the full picture of how visible your brand is today, that is what our Recruitment Marketing Audit establishes: your scores on all 5 dimensions and the single finding costing you the most applications. What the AI visibility half measures is set out in full, scorecard included, and if you are comparing us against someone else, how to choose an AI visibility audit gives you 5 criteria and 4 red flags to hold us to. Agencies and in-house recruiters who want the tooling should look at Studio. For the wider methodology, see our guides to generative engine optimisation and answer engine optimisation for recruitment.

Source: The SetpointHQ Index, 100 UK employer brands, scored monthly across ChatGPT, Claude, Gemini and Perplexity. All-sector and sector figures as at the 1 September 2026 run, rubric 2026.06.5. Retail cohort figures as at the July 2026 run.