Does the way candidates ask AI about employers change which employers are named?

Having spent much of my career leading Talent Acquisition teams, I've seen how much effort goes into understanding what candidates want from employers.

We've researched career motivations, tested employer messaging and spent considerable time making sure our EVPs address the things candidates care about.

But AI introduces another challenge. Candidates aren't necessarily researching employers using the language we've chosen to describe our organisations.

A candidate might ask:

"Which companies offer good career progression for someone working in finance?"

Another might ask:

"Where could I work as an engineer without having to relocate?"

Someone else might simply ask:

"Is this company actually a good place to work?"

These are quite different questions, even though they're all about choosing an employer.

And the way they're phrased can influence which organisations appear in the answers.

That creates a challenge for anyone trying to measure employer brand visibility across AI.

We wanted to understand whether the way we phrased questions was affecting the results we measured through BrandScore.

So we ran a pilot using more natural, conversational prompts rather than relying solely on our existing structured assessment questions.

The pilot covered 2,640 AI-generated answers.

The findings were interesting.

Overall employer presence decreased by around two percentage points compared with the structured prompt approach.

But that relatively small overall movement concealed much larger differences between engines.

• ChatGPT employer presence increased by 13 percentage points.
• Gemini employer presence decreased by 11 percentage points.
• The more natural questions produced approximately 1.3 additional employer names per answer on average.

That last finding matters too.

Even when an employer's visibility doesn't change dramatically, the number of other organisations appearing alongside it can change.

Those employers are part of the competitive picture candidates may encounter.

To be clear, this was a controlled prompt-design pilot, not research tracking the exact questions real candidates typed into AI.

We were testing how differences in question wording affected generated responses.

But the results demonstrate that the design of the questions matters.

AI systems don't necessarily interpret and answer questions in the same way.

Some may provide longer lists of employers. Others may offer shorter lists of employers or spend more of the response explaining their reasoning.

They may also differ in how they interpret a question about career development, flexibility or workplace culture.

That means a change in question wording can affect the employers being named, even when the underlying candidate priority remains broadly similar.

This is particularly important when comparing employer visibility across several AI experiences.

A change in measured visibility might reflect a real difference in the organisations being named.

But it could also reflect differences in how the question was constructed, how each engine responded or how many employers appeared in the answer.

Without consistent measurement, those factors are easy to confuse.

I think it does.

Traditional employer brand research has always paid attention to how questions are asked.

Anyone who has designed a candidate survey knows that changing the wording of a question can influence the response.

Ask someone whether career development is important and you may get one answer.

Ask them to choose between career development, flexibility and salary, and the result may be different.

We understand why question design matters when measuring people.

We should apply the same discipline when measuring AI-generated employer answers.

The challenge is that AI introduces multiple variables.

You're not just dealing with different questions. You're dealing with different engines, response formats, market contexts and sometimes substantial variation between repeated answers.

And unlike a conventional survey, the system producing the response can change without the employer doing anything differently.

This makes the methodology behind an AI employer brand assessment particularly important.

No. I think that would create a different problem.

Candidates can research employers in several ways.

Some ask broad discovery questions without naming an employer.

Others compare two organisations directly, investigate particular working conditions or ask whether an employer has a good reputation.

Some already know where they want to apply and are looking for reassurance.

Others are trying to discover employers they haven't previously considered.

A meaningful assessment needs to account for these different types of questions.

It also needs to distinguish between being named and receiving an explicit endorsement.

An organisation appearing in a long list of potential employers is not necessarily being endorsed for a candidate's circumstances. AI Answer Presence measures naming, not endorsement.

The number of employers named, the wording of any explicit endorsement and the context of the question all matter. They should not be treated as interchangeable measures.

For me, the most important lesson isn't that conversational prompts are better than structured ones.

It's that the questions used to measure AI visibility can materially influence the results.

If an employer is told it appears in 40% of assessed AI answers, that figure needs context.

What questions were asked?

Were they relevant to the people the organisation wants to recruit?

Were they discovery questions, direct employer questions or comparisons?

Which AI engines were measured?

And were the same questions and measurement rules used when comparing results over time?

Without that information, a percentage tells you relatively little.

We've spent years trying to understand how candidates make decisions about employers.

AI makes that work more complicated, but the underlying principle hasn't changed.

Before trusting the answers, we need to understand the questions.

BrandScore's natural-language prompt pilot examined 2,640 AI-generated answers and compared results with a structured prompt approach. The observed changes included a two-percentage-point overall reduction in measured employer presence, a 13-percentage-point increase for ChatGPT and an 11-percentage-point decrease for Gemini.

These are differences within the pilot, not universal estimates of how those AI engines respond. The pilot did not establish how frequently real candidates use particular question formats or whether the resulting answers affect application decisions.

• Question wording matters. Different ways of asking about the same career priority can produce different employer appearances.
• Candidate intent matters. Discovery, comparison and employer-specific research should be assessed separately.
• AI engines behave differently. Results from one engine cannot automatically be applied to another.
• Visibility needs context. A percentage is only meaningful when you understand the questions, markets and measurement criteria behind it.
• Consistency matters over time. Changes in question design should not be confused with genuine movement in employer visibility.