AI-Generated EVPs: Why Skipping the Research Is a Really Bad Idea

Apparently, developing a new EVP has become fairly straightforward. Upload some employee surveys, ask AI what people think about the company, generate a few pillars and tidy up the wording. Right? What once required research, analysis and some difficult decisions can now be completed with a subscription and a spare afternoon. Super efficient... provided nobody asks too many questions about the answer.

SPONSORED

The best employer brands don’t manufacture stories. They uncover what’s already true. Sell The Truth: Leadership Storytelling For Cultures That Win by Bryan Adams shows leaders how to turn truth into stories that build belief, trust and competitive advantage.

Order now

Of course, I understand the attraction. Budgets are tight, people are busy, agencies cost money and everyone wants things finished sooner. But after 12 years of developing EVPs, having had a hand in more than I can count, I think we’re confusing the ability to produce a document with having done the actual work behind it. AI is exceptionally useful for the former. It doesn't excuse skipping the latter.

A proper EVP shapes what a business says about itself, what candidates expect and where recruitment marketing money goes. Those consequences last considerably longer than the afternoon spent generating a faux EVP Guidebook. Getting it right is a genuine big deal and should be done properly and with rigour.

This is a company’s reputation as an employer we're talking about. It deserves a better foundation than something that sounds about right.

Would marketing accept this?

Imagine proposing a consumer brand repositioning without properly researching the customers. Would some customer service feedback, a few internal opinions and a chatbot’s interpretation of the market now suffice? Would any sane business go off and spend the next three years advertising the result? I think not.

A competent marketing leader would definitely have questions. Who does the evidence represent? What do those people want? Why would they choose this brand? What do competitors offer? How will we measure whether the positioning works? Employer branding should expect exactly the same scrutiny. We’re trying to influence a decision that affects someone’s career, income and daily life. Understanding the people making that decision is pretty darn fundamental.

Yet the quick AI (yes, slop) approach starts with whatever information happens to be on hand and asks a machine to make something of it. The missing information stays missing. The gaping holes just become less obvious once everything has been polished into a nice confident presentation. AI is remarkably good at making things sound correct, even when they're not. When Pew Research Center had leading AI models stand in for real survey respondents in September 2026, they missed actual public opinion by 11 to 13 percentage points on average. Confident, tidy and wrong.

Don't kid yourself. The chatbot hasn’t surveyed the people you need to hire

Ask AI how an employer is perceived and it may return useful material. Reviews, news coverage and public discussion can reveal issues worth investigating. But an account of what is available online simply cannot establish what a particular talent segment thinks about you. Trust me, I've looked into this in depth and it doesn't even come close.

It doesn't take long to unpick. Are those views AI confidently cites from people in the right country, occupation and career stage? Do they concern the current business or an operation someone left five years ago? What about qualified people who have never applied, never posted a review or barely heard of the company? If the recruitment challenge involves experienced engineers in Singapore, let's say, then a general internet summary is a weak substitute for research among relevant engineers in Singapore.

Proper primary research recruits people against defined criteria and asks questions designed to inform decisions. It also gives you the ability to compare and contrast talent segments (so long as the sample supports the drill downs), enabling you to create very sharp messaging. It can establish not only what a perceived strength is, but whether it matters, whether a supposed weakness exists and whether candidates associate the business with anything worthwhile at all. Asking a chatbot to be more specific does not yield the same results as drilling into fresh proprietary data. Just like asking four chatbots doesn't recruit four samples either.

Existing employee surveys are valuable, but they answer their own questions. People who joined can explain why they joined. Leavers can explain why they left. Neither group fully explains why the people the business wants next are choosing somebody else. Feeding their responses into AI doesn't close that gap.

Imagine employees consistently praise autonomy, so the EVP becomes all about freedom. External research might reveal that candidates already expect independence and actually worry about being left without support. The business could then spend a year enthusiastically promoting the very thing that puts people off. That is an expensive mistake to make because the research seemed... expensive.

FROM THE EBN NETWORK

Employer branding now has a measurement standard. The Talent Gravity Standard is a six-driver framework for quantifying employer attractiveness and the gap between brand promise and employee experience.

Read the Standard

Research earns its keep in the difficult meetings

The value of research also becomes obvious when the proposition reaches leadership. I’ve worked on plenty of EVPs where the CEO enters the room with a preferred narrative already in mind. Sometimes it is supported by the evidence. Sometimes it's wishful thinking. Without data, pushing back becomes a contest between the practitioner’s opinion and the CEO’s conviction. No real data to fall back on? Good luck with that.

With proper evidence, that discussion has somewhere useful to go. It goes from being an unwinnable battle of will with a company CEO, to a professional, data-driven conversation, which most CEOs I've worked with tend to respect. Only when we bring the receipts can we stand up and explain what employees experience, what relevant talent values and where the proposed promise falls short. Leadership still makes the final decision, but it now has to confront the gap between the story it wants and the evidence available. That gap matters well beyond the meeting room. In a Gartner survey published in September 2024, only 33% of employees said their organisation consistently delivered on its EVP promises.

This is where logic and reason tend to prevail.

The same applies when leaders disagree. I’ve seen conflicting feedback turn an approval process into endless loops of revisions, with each person wanting their particular idea included. Without a shared basis for choosing, the eventual EVP becomes a little of everything, carefully diluted until nobody objects. In these scenarios, only credible research can protect the proposition from that process. AI can keep rewriting the compromise indefinitely; the data helps establish what deserves to survive.

In-house can be cheaper. It isn’t automatically quicker.

Avoiding agency fees is a legitimate saving. It is also easy to underestimate the work a good agency should bring ready to use: a developed methodology, research expertise, statistical knowledge and a process for getting from findings to decisions.

In my experience, an agency is almost always quicker at doing the hard work. I’ve been brought in by employers who had already spent eight months trying to design an internal survey and getting nowhere... then completed the entire EVP four months after starting. No shortcuts, just knowing how to collect, analyse, and work through both the data and the creative process that follows it. The internal team absolutely understood the business. What they lacked was an established way to do this particular job and the years of domain knowhow that's hard to hand over in a call.

Developing a sound research methodology is a substantial project in itself. A questionnaire generated by AI does not establish that the questions measure the right things, that the sample is appropriate or that the conclusions are justified. These are research skills. Enthusiasm and a good prompt cannot be relied upon to supply them. Errors in things like sample size can be very costly.

Luckily, there's now more help available for in-house teams. Methodologies like the Talent Gravity Standard give employers access to a published, best-practice framework for developing or measuring their own EVP. And crucially, without having to pay an agency for the privilege. Any employer can use the core framework under a free licence, and in-house practitioners in EB Club get the full toolkit, including the survey instrument and scoring rubric, at no cost. The choice is no longer between a Blue Peter-style DIY effort and an expensive agency engagement. Employers can have the best of both worlds, saving money, time and a fair amount of head scratching along the way.

FROM THE EBN NETWORK

Your EVP is a guess until the data says otherwise. Fathom builds EVPs from evidence, measuring the gap between what you promise and what people actually experience.

See how it works

And how will anyone know whether it worked?

Skipping external research also creates a problem that appears after the launch. The careers site has been rewritten, campaigns are running and leadership wants to know whether perceptions have improved (this is an exercise in brand building, after all). Well, what were those perceptions before the work started?

Without a proper and empirical baseline, measuring improvements in brand becomes impossible to do properly. The team may have clicks, traffic and application numbers but little basis for assessing changes in brand equity or brand associations. Those are different measures. A busy campaign does not establish a strong employer brand, it doesn't show that more of the right talent find the company a more attractive place to work than they did before.

Research done properly before activation provides a starting point for comparable measurement later. There is still some work to do to tie brand to improved business outcomes (like reduced cost per hire, increased quality of talent etc), but it gives the organisation something credible to track and measure. Leaving it out and hoping to demonstrate brand improvement afterwards is shockingly poor planning. An industry report or a chatbot’s reconstruction of what people probably thought will not replace the missing measurement.

Use AI to do the work better

There are plenty of sensible jobs for AI in EVP development. Organising material, assisting with thematic analysis, exploring interpretations and drafting language from validated findings can all save time. Someone competent still needs to check the analysis, and confidential employee feedback needs an approved environment and proper handling (i.e. not your 20 bucks a month ChatGPT subscription).

A smaller budget can still support a smaller, focused research programme. Prioritise the talent audiences that matter most, reuse suitable evidence and investigate the important gaps. Be honest about what remains untested. An in-house team can still do this well, and an agency can certainly do it badly. The crux is whether the decisions have real evidence behind them.

What I strongly warn against is removing essential research and presenting the resulting speed as an improvement. We haven't discovered a more efficient way to understand the audience if all we did was stop asking it questions.

Instead of building a shiny new ship, the best you can hope for is a rickety little raft that won't make it through even the lightest of storms.

The EVP is usually the cheap part anyway. The business will spend far more activating it through advertising, content, careers technology and years of recruitment communication. At least it should. If building an EVP costs more than the years of branding and recruitment marketing that's done as a result of it, serious questions need asking.

If employer branding wants to be taken seriously, it should be prepared to meet the standards expected of serious brand work. Use AI to save time where it can. But before spending years asking people to believe a promise, do the research that establishes whether the promise deserves to be made.

Takeaways

SOURCES