I asked four AI tools the same question. I got four different answers.
“What are the best companies for industrial CNC lathes?”
ChatGPT led with DMG MORI. Claude, Gemini, and Perplexity all led with Mazak. Every tool agreed on four names: Mazak, Okuma, DMG MORI, and Haas. After that, the lists diverged completely.
Makino, a legitimate top-tier precision manufacturer, appeared in exactly one out of four. Hardinge showed up in two. Tsugami, Tornos, and Index Traub each surfaced once. These are real companies making real industrial equipment, and whether they appear in a buyer’s AI conversation comes down to which tool that buyer happened to open.
Run the same query later today and you may get different results. Reword it slightly and the list changes. Use a different account and the order shifts again. I know this because I run these queries for a living, and the one constant is that no single AI answer is stable enough to mean anything on its own.
This isn’t a glitch. It’s how the technology works. And it breaks the entire measurement model that marketing learned from search.
Rankings assume a fixed list. AI doesn’t produce one.
Search trained a generation of marketers to track position. Where do we rank for this keyword? Are we above the fold? Did we move up two spots this month?
That model worked because search returns a stable, ordered list. Position one is position one. Position five is position five. You can track it, graph it, and build a reporting cadence around it.
AI doesn’t return a list. It returns a synthesized answer. It pulls from training data, cross-references live sources, and constructs a response in real time. The output depends on how the question is phrased, what the model already knows, and what it can verify. Change one word and the recommendation changes.
Any single screenshot of an AI answer tells you almost nothing. You showed up this time. You didn’t show up next time. That’s not a signal. That’s noise.
The tracking industry is selling dashboards, not insight
Most AI visibility tools right now are repackaged rank trackers. They run a prompt, screenshot the result, and show you “where you ranked.” Then they run it again tomorrow and show you the delta. Monthly reports. Trend lines. The illusion of precision.
Rand Fishkin and SparkToro tested this rigorously. They ran 2,961 prompts across ChatGPT, Claude, and Google AI Overview. The finding: if you ask an AI tool for brand recommendations a hundred times, you have less than a 1 in 100 chance of seeing the same list twice. The order is random. The number of recommendations is random. The list itself is random.
Fishkin’s assessment was blunt: most of this industry is selling executive anxiety in dashboard form. The data looks precise. The charts look authoritative. But they’re tracking a metric (position in a single AI response) that isn’t stable enough to mean anything.
This is the SEO rank tracking business model, repositioned for AI. Same monthly retainer. Same dashboards. Same fundamental misunderstanding of what’s being measured.
What actually works: inclusion rate
The SparkToro research found something useful underneath the noise. When you run the same category of question dozens of times across multiple prompt variations, a pattern emerges. The percentage of times a brand appears, its inclusion rate, is surprisingly stable.
Inclusion rate measures whether you appear at all, not where you appear or in what position. A company with 70% inclusion rate across 50 query variations is in a fundamentally different position than one with 15%, regardless of where either one “ranks” in any individual answer.
This is the metric that matters for Coverage. Not “position three in ChatGPT.” That’s a meaningless number. Inclusion rate tells you whether AI systems can find and represent your company when buyers ask about your category.
But here’s the part most coverage tools miss: inclusion rate is a trailing indicator. It tells you what’s happening. It doesn’t tell you why.
The companies that show up consistently share something upstream
Go back to the CNC lathe test. Four AI tools, four different answers, but four names appeared every time: Mazak, Okuma, DMG MORI, and Haas. They didn’t agree on the order. They didn’t agree on who’s fifth. But those four showed up regardless of which tool you asked.
What do they have in common? It’s not ad spend. It’s not content volume. It’s not who has the best website.
Their information is clear, structured, and consistent across the web. Directory listings match. Product descriptions match. Company descriptions match. Press coverage reinforces the same positioning. There’s enough specific, verifiable data about each one that AI systems can find them, understand them, and recommend them with confidence.
The companies that appeared in some tools but not others share a different trait. Their data is ambiguous, thin, or contradictory. The AI can’t verify which version is accurate, so it skips them when it has higher-confidence alternatives.
This isn’t a Coverage problem. It’s a Clarity problem.
What we see in practice
At CKI Labs, we run Coverage diagnostics across ChatGPT, Claude, Gemini, and Perplexity. We see the same inconsistency Fishkin documented. But we also see something his research didn’t test for: most companies aren’t losing to randomness. They’re losing to Clarity.
One company we worked with had been in business for decades. AI tools could not accurately state how long. One page says 50 years. Another says 65. The company’s own website contradicts itself, and AI surfaces both numbers depending on which page it pulls from. That’s not an AI inconsistency problem. That’s a data integrity problem.
Another company’s entire product catalog is rendered in JavaScript that AI systems can’t read. AI can’t recommend products it cannot see. The inconsistency Fishkin measured assumes AI has access to all the candidates and picks randomly. Baker isn’t even in the pool.
Then, another company was well-known to AI tools. They don’t coherently describe what makes their products different or why someone should choose them. The inclusion rate is moderate. The representation is shallow. Known but not recommended.
These companies don’t lose to randomness. They lose to their own data.
The Ceiling Rule
Fishkin proved that rankings in AI responses are noise. He’s right.
But he also proved that inclusion rate is signal. The brands that appear consistently across dozens or hundreds of runs share something: AI systems can read their data, trust their data, and connect their data to relevant queries.
We call that Clarity. It’s the first layer in the C3 Diagnostic, and it caps everything above it. Coverage cannot exceed Clarity because AI cannot recommend data it cannot trust. Conversion cannot exceed Coverage because visitors cannot convert on a site they never reach.
Most companies spending money on AI tracking are trying to measure Coverage while ignoring Clarity. That’s like paying for a weather station while ignoring the hole in your roof. The measurement is fine. The problem is structural.
What to actually do
If you want to improve how AI systems represent your company, start here:
Fix your data first. Pull your company description from your homepage, your about page, your Google Business Profile, your LinkedIn, and any directory listings. Do they say the same thing in the same words? If not, you’re feeding AI systems contradictory information and hoping it sorts it out. It won’t.
Measure inclusion rate, not position. Run 20-50 variations of buyer-intent prompts across multiple tools. Calculate your appearance percentage. Track that over time. That’s your Coverage signal. Ignore position-based rankings entirely.
Check representation accuracy. When you do appear, is what AI says about you correct? If AI recommends you but describes you wrong, that’s not a win. It’s a hallucination problem that erodes trust with every interaction.
Solve Clarity first. Consistent, structured, verifiable data is the upstream lever. When your information is clear, Coverage improves as a downstream effect. Then you can focus on what actually multiplies revenue: whether the visitor AI sends you converts.
The question isn’t “where do we rank in AI?” The question is “can AI reliably describe what we do?” If the answer is no, no dashboard is going to fix that. Fixing your data will.
