What sources does AI actually use when recommending employers?

For most of my career in Talent Acquisition, we've spent considerable time thinking about what candidates see when they visit our careers websites.

We've invested in EVP development, employee stories, careers content, recruitment campaigns and increasingly sophisticated ways of measuring engagement.

But candidates don't always start their research on an employer's website anymore.

They can ask ChatGPT, Gemini or Perplexity about potential employers, working conditions, career development and company culture before visiting a single careers page.

And the information they receive may come from sources the employer has never considered part of its employer branding strategy.

That creates a different challenge.

It's no longer enough to know what your organisation is publishing. You also need to understand what information AI is referencing when it describes you.

Across our BrandScore research, we recorded 35,529 genuine citations spanning 3,001 domains.

Only approximately 4.3% of those citations pointed to employers' careers websites.

That surprised me.

Careers websites are where organisations typically publish their most carefully developed employer brand information. They're often the primary destination for EVP messaging, employee stories, benefits and career opportunities.

Yet within the AI responses we examined, careers sites accounted for a relatively small proportion of the cited sources.

This doesn't mean AI ignores careers websites, or that they contribute only 4.3% of the information used to generate responses.

A citation tells us which source an AI-generated response visibly referenced. It doesn't reveal every source that may have influenced the answer.

But the finding does suggest that employers need to think beyond their own published content when considering how AI represents them.

Not entirely. But control has always been limited.

Even before AI, an organisation couldn't determine what employees said on Glassdoor, how journalists described it or what candidates heard from their professional networks.

Employer branding has always involved the difference between the story an organisation tells and the reputation it earns.

AI introduces another layer.

It can combine information from multiple sources into a single answer, often presenting that answer as a convenient summary for someone considering their career options.

The candidate may never see the original sources, particularly when an answer doesn't provide citations.

This makes the accuracy and consistency of publicly available employer information increasingly important.

It also means the strongest employer brand message isn't necessarily the one candidates will encounter.

I don't think volume is the answer.

If an organisation is already publishing employee stories, EVP content and careers information, creating another twenty articles won't necessarily change how AI describes it.

The starting point should be understanding the existing picture.

Which sources are being cited? Are those sources relevant and current? What do they say about the organisation? Where does the AI-generated description differ from the EVP?

There may be straightforward opportunities to improve information that's incomplete, difficult to access or out of date.

There may also be uncomfortable findings that require attention beyond marketing.

For example, if external employee feedback consistently contradicts an organisation's claims about leadership or career development, publishing more positive employer brand content doesn't resolve the underlying issue.

That is where employer brand measurement and employee experience need to connect.

This is an important limitation.

AI systems don't consistently expose the information behind their answers.

Some provide links to sources. Others generate responses without showing citations. The same AI experience can behave differently depending on the question and the way it retrieves information.

So citation analysis is useful, but it isn't a complete map of everything AI knows or has used.

I would be cautious of anyone claiming they can definitively identify every source responsible for an AI-generated employer recommendation.

What we can do is examine the evidence that is visible, compare responses across engines, and observe whether descriptions and recommendations change over time.

That gives us a more defensible basis for deciding where employer brand teams should focus their attention.

I think it is.

The traditional employer brand toolkit remains relevant. Careers websites, employee stories, recruitment marketing and EVP research still matter.

But the boundaries are becoming less clear.

Public employee feedback, third-party recognition, company information, professional communities and other sources can all contribute to the wider picture candidates encounter.

For Talent Acquisition and Employer Brand leaders, the question isn't simply whether we're communicating our EVP effectively.

It's whether the wider evidence about our organisation supports the reputation we're trying to build.

And whether AI is reflecting that evidence accurately.

That's a much more useful question than how many new pages we should publish.

The figures cited come from BrandScore's analysis of 35,529 recorded citations across 3,001 domains in AI-generated employer responses. Careers website citations accounted for approximately 4.3% of the recorded total. The findings describe visible citations within the measured sample, not every information source used by AI systems or the relative influence of those sources.

• Look beyond your careers website. Understand which external sources AI visibly references when describing your organisation.
• Check the accuracy of the wider evidence. Identify outdated information, missing context and contradictions with your EVP.
• Don't confuse citations with influence. A cited source is observable evidence, not proof of everything an AI system used.
• Connect employer branding with employee experience. Where public feedback contradicts your messaging, investigate the underlying issue.
• Measure repeatedly. Compare descriptions, recommendations and cited sources across relevant AI experiences over time.