Part 2: The hidden problem: you can’t “SEO” your way into AI answers with a URL and a score

There’s a comforting fantasy doing the rounds in marketing right now: “If we plug our URL into an AI visibility tool, we’ll get a score… and the score will tell us what to do.”

It feels like SEO. Familiar inputs. Familiar outputs. Neat dashboards. A reassuring red/amber/green.

And it’s also the fastest route to false confidence.

Because AI answers don’t work like Google rankings, and they definitely don’t work like a site audit. You can’t “SEO your way in” with a URL and a number - not when the thing you’re trying to influence is a conversation.

Why URL-in / score-out tools mislead you

Most “AI visibility” tools built in a hurry have the same basic pattern:

1. Enter your domain + brand
2. Run a quick set of generic prompts
3. Get a single visibility score
4. Get vague recommendations like:
   • “Improve authority”
   • “Write more thought leadership”
   • “Add case studies”
   • “Optimise your content”

That output looks actionable… until you actually try to act on it.

Because it doesn’t answer the only questions that matter:

• Which buyers? (CIO vs Ops vs RevOps)
• Which moment? (exploring options vs shortlisting vs due diligence)
• Which competitors? (incumbents vs niche upstarts)
• Which claims? (what the model will or won’t believe about you)
• Which sources? (where the model is pulling evidence from)

A single score collapses all of that into one number. It’s tidy. And it’s useless.

The real issue: AI answers aren’t “ranking” your site

AI systems don’t reliably “visit your website and decide you deserve to rank”.

They assemble answers from a mix of:

• their training data,
• retrieval systems,
• public web content,
• and whatever the prompt makes salient.

So when you run a generic set of prompts and get a generic score, you learn almost nothing about the actual buying journey you need to win.

Even worse: you learn the wrong lesson.

You start optimising your site broadly (“more content”, “more keywords”, “more pages”) instead of fixing the specific moments where you’re losing.

The confidence trap: vague to-dos that never change outcomes

Here’s what “false confidence” looks like in practice:

• You see a score improve from 42 → 58.
• You ship three “AI-friendly” blog posts.
• You feel like you’re making progress.

But nothing changes in pipeline.

Because you didn’t identify:

• the prompts buyers actually use,
• the criteria they use to shortlist,
• or the proof they need before they’ll take you seriously.

You improved a score, not a decision.

What actually works: control the prompt library + see the evidence

If you want to influence AI answers, you have to stop thinking like a crawler and start thinking like a buyer.

That means your system needs to do two things:

1) Control the prompt library

Not “a few prompts”. A proper library mapped to:

• ICPs and roles,
• problems and use cases,
• buying stages (explore → compare → validate → decide),
• and the language buyers use when they don’t know your category yet.

2) Show the evidence

Not just “you were mentioned”. You need to see:

• where you were included or excluded,
• who you were compared against,
• what claims the model made about you,
• and what proof it demanded (or invented).

That’s the difference between “we’re visible” and “we’re credible”.

What to measure instead (so you can actually change outcomes)

If you want a measurement framework that leads to clear priorities, focus on three things:

1) Buyer-stage visibility (not one score)

Measure whether you appear in answers at the right stage:

If you’re only visible when the prompt includes your brand name, you’re not visible. You’re being retrieved.

2) Competitor inclusion (who you’re actually up against)

Track:

• which competitors consistently appear,
• how you’re described relative to them,
• and what categories the model places you in.

This is where most teams get surprised - because the AI’s “competitive set” is often not the one in your pitch deck. If you don’t measure competitor co-mentions and comparison framing, you’ll optimise the wrong narrative.

3) Proof gaps (why the model won’t confidently recommend you)

This is the money metric.

For each critical prompt, capture:

• which claims the model won’t make about you,
• what proof it wants (case studies, security, compliance, pricing transparency, integrations, outcomes),
• and what sources it’s using (or failing to find).

Then your “to-do list” stops being vague (“do more thought leadership”) and becomes concrete:

• “We’re excluded at Validate stage because there’s no public evidence of X integration.”
• “We’re described as a ‘tool’ not a ‘platform’ because our narrative lacks Y.”
• “We never win in regulated contexts because proof of governance is missing.”

That is prioritisation you can execute.

The punchline

If your AI visibility tool starts with your URL and ends with a score, it’s measuring the wrong thing.

The game isn’t “optimise my site for AI”.

It’s: win the prompts that matter, at the buyer stages that decide, with the proof that creates belief.