Eli Schwartz

<start of post>

Most B2B teams are about to waste a year chasing 'AI visibility'.

Not because AI search doesn’t matter.

But because they’re using the same old playbook with a new label.

They’re paying for vague “GEO/AEO” retainers, publishing mass content, and celebrating dashboards that can’t prove anything moved.

The pitch looks clean. The metrics look busy. The outcome is usually a shrug.

And once the budget is gone, you’re left explaining to leadership why you can’t connect spend to pipeline.

A vendor changes 20 things at once (titles, internal links, schema, rewrites) and calls it a “system”

Reporting that shows impressions and “model visibility” but can’t tie to qualified traffic or revenue

Case studies that only work because the brand was already a category leader

Teams tracking prompts like rankings, even as personalization makes the same prompt mean 1000 different experiences

You can’t optimize what you can’t isolate.

And you can’t isolate what you refuse to measure.

This doesn’t mean AI optimization is fake.

It means most programs are built to look impressive, not to be testable.

Start with a measurement spine you trust, then layer experiments on top of it.

If you want a practical way to do this, use a three-layer approach.

Layer 1: Keep one channel “clean”

Pick a set of pages you do not touch for 30 days. No rewrites. No schema changes. No new internal links. This becomes your control group, so you have something real to compare against.

Layer 2: Run one-variable experiments

One change. One hypothesis. One page group. If you change the page AND the template AND the internal links, you didn’t run an experiment. you created a story you can’t verify.

Layer 3: Measure outcomes that map to money

Traffic is not the outcome. Neither is “being cited.”

Pick 2–3 actions that represent intent for your business: demo requests, pricing page views from qualified sources, sales conversations, email signups from the right pages. Then track those against the experiments.

A lot of teams skip this because it’s slower.

But slow is still faster than spending 6 months on an initiative you can’t defend.

Most companies are optimizing for what they think the model reads, not what the customer needs at the moment they ask.

That’s why customer journey mapping keeps showing up in every strategy that actually lasts.

When you understand the questions people ask before they buy, you stop guessing at “AI keywords” and start building the pages that naturally get pulled into answers.

And yes, shockingly (or not) it performs in traditional search too.

How will we prove this worked?

If the answer is vague, pause.

DM me if you want me to sanity-check the proposal and point out the measurement gaps before you lock in a long retainer.

<end of post>