The AI Compression Problem Your EVP Isn't Ready For

Last week I wrote about narrative compression - how AI squashes everything it knows about your company into a handful of sentences when a candidate asks what it's like to work for you. If you haven't read that piece yet, start there. This one goes deeper.

What's ringing around my ears is that when AI compresses your employer brand, it doesn't give candidates your five EVP pillars. It doesn't walk them through your carefully segmented messaging. It gives them two or three descriptors. Maybe four if you're lucky.

The question that should be scratching away at the back of your mind is, "are those descriptors yours, or are they the same generic filler AI says about everyone?"

Compression vs. dispersion

If AI visibility auditing becomes standard practice - and I think it will (especially for larger organisations) - I think we'll see two distinct patterns emerge.

The companies that win will compress cleanly. Ask six different AI tools what it's like to work there, and you'll get broadly the same answer. The themes will be consistent. The language distinctive. The brand will have a clear shape in AI's understanding - not identical across every tool, but recognisably the same company.

The companies that struggle will fragment. ChatGPT will say they're known for innovation. Gemini will talk about work-life balance. Perplexity will focus on prestige. Claude will mention their graduate programme. None of them wrong, exactly, but none of them telling the same story. The brand will have no shape. It disperses.

That dispersion is the coherence gap. And it won't be random. It'll be a symptom of something structural.

Why some brands will fragment

If your employer brand compresses into mush, it's almost never because you haven't invested in employer branding. Most companies with a coherence gap have an EVP. They've done the research. They've got the pillars and the messaging framework sitting in a PDF somewhere.

The problem is what's happening downstream. And it usually comes down to one of three things.

Your high-frequency signals are contradicting your strategy. Your EVP says "collaborative and entrepreneurial." Your job descriptions and adverts say, "must thrive in a fast-paced, high-pressure environment." Your Glassdoor reviews say "long hours and siloed teams." AI doesn't read your EVP document. It reads the signals that appear at volume, and those signals are telling three different stories. I've been banging on about the need to improve jobs ads for too long. Now, more than ever, this is something people need to take really seriously. Not only to improve the compatibility of the talent you attract, but to also ensure the LLM aren't going rogue in how they represent you.

Your operating model is working against you. This is the one that doesn't get talked about enough. Large, decentralised organisations - where different countries or business units are doing their own thing because "they know what's best" locally - are structurally more likely to have a coherence gap. Not because decentralisation is wrong, but because nobody is looking at the aggregate signal. Each region might be telling a perfectly fine story in isolation.

But when AI absorbs all of those stories at once, the result is noise, not narrative.

You're optimising for the wrong surfaces. The employer brand team is pouring energy into the annual campaign, the brand film, the big LinkedIn moment. Meanwhile, there are 500 job descriptions live right now, each one written by a different hiring manager, each one using different language, and those are the signals AI sees most often. Charlotte Marshall nailed this distinction recently - the gap between low-frequency, high-investment content and high-frequency, low-investment signals. If you're only managing the first category, you're managing a tiny fraction of your actual brand footprint.

How to test this yourself

You don't need an enterprise-scale audit to get a rough sense of where you stand. Here's a simple exercise.

Pick three AI tools - ChatGPT, Gemini, and one more of your choice. Ask each of them the same branded question: "What is it like to work at [your company]?" Then ask the same question about two of your closest competitors.

Now compare the answers across all three tools. For your company:

• Are the core themes consistent, or does each AI tell a different story?
• Write down the top three descriptors each tool uses. How much overlap is there? Now do the same for your competitors.
• Which company compresses more clearly - you or them?

If your competitors compress into distinctive, repeatable themes and you fragment into generic descriptors, that's your coherence gap made visible.

This is measurable, not theoretical

What I described above is a quick-and-dirty version. The structured approach - which I'm seeing applied at enterprise level right now - goes much further. Multiple AI tools, controlled prompt design across branded and non-branded questions, repeated runs to test stability, variation across candidate personas and regions.

The output isn't subjective. You can measure how concentrated or dispersed your descriptors are across AI outputs. You can map which themes are stable across different tools and which ones only appear in one. You can compare your compression pattern directly against your competitors and see, with evidence, who has a clearer signal.

What makes this genuinely different from traditional perception research is that traditional methods tell you what people think when asked. This tells you what AI will tell people before they ever get to your careers pages. It's not replacing candidate surveys and focus groups. It's revealing an entirely new layer of perception that most employer brand teams aren't even aware exists.

The part to make you wince

If AI is compressing your employer brand into a few generic descriptors, throwing more money at campaigns won't fix it. Even if you do win an award at an industry ceremony where you were one of 137 others. The compression happens upstream, at the signal level. The brand films and launch events might generate a momentary spike, but AI's understanding of you is built from the steady drumbeat of everyday content - the signals that repeat at volume- week after week. Month after month.

Fixing a coherence gap means working on signal consistency across your high-frequency surfaces. Job descriptions, job adverts, career site copy, recruiter talk tracks, candidate FAQs, the language your employees use on LinkedIn. It's less glamorous than a big campaign. It's also significantly more effective at shaping how AI understands and represents you.

Because the next generation of search doesn't rank you by how loud you are. It ranks you by how clear you are. By how consistent you are. And right now, consistent clarity is rarer than you'd think.