AI visibility platform or SEO suite: which one measures what you need?

We get this question in one specific form — Profound or Ahrefs? — and the honest answer is that it is the wrong comparison, but there is a real question underneath it that deserves a definitive answer.

What this page will and will not do

We publish an index in this category, so we are an interested party and it would be silly to pretend otherwise. We also do not publish competitors' prices or feature lists: this market changes monthly, and a stale comparison table presented as current is exactly the kind of confidently-wrong artefact we spend our time arguing against.

What is durable is the method. Two tools can both be correct and disagree, because they asked different questions of different assistants a different number of times and divided by a different set of competitors. That is not a defect in either one. It is the thing you are actually choosing between.

The two approaches

Tools that sample assistants directly

Profound and most AI-visibility natives work this way: a question set is run against the assistants on a schedule and what comes back is recorded. Our own index is built the same way, so this is the approach we know from the inside.

What it gets you

• You control the exam. You decide which buyer questions count, which means you can measure the questions that actually precede a purchase in your category rather than the ones that happen to have search volume.
• It is a direct observation. You are reading what an assistant said, not inferring it from a proxy.
• It covers assistants that have no search-demand signal at all. Conversations inside ChatGPT or Claude do not show up in anybody's keyword tool.
• It can be versioned. A frozen question set is what makes a trend line mean something over years.

What it costs you

• You can invent your own corpus. This is the real failure mode: write questions your buyers would never phrase that way, and you will produce a confident number about an imaginary market. It fails silently, because the measurement still works — it is just measuring the wrong thing.
• It is expensive to do at a sample size that means anything. Answers are non-deterministic, so a rate needs repetition. Our own brand-level readings run roughly 300 sampled answers across four providers.
• Small samples produce wide intervals that most vendors do not show you. In one of our own readings a 10.7% mention rate carried a true range of 7.5–15.1% across 252 sampled answers. A dashboard reporting "10.7%" and then "9.8%" next month has reported nothing at all.

Tools built on observed search demand

Ahrefs' Brand Radar is the clearest example: prompts are derived from real search behaviour rather than written by you.

What it gets you

• The questions are demonstrably asked. You are not guessing at intent; you are starting from evidence that people type this.
• Scale and cost. Riding an existing crawl and keyword corpus is far cheaper per question than paying an assistant to answer one 300 times.
• Continuity with what you already track. If your reporting is built on search data, the vocabulary matches.

What it costs you

• Less control over the exact set, which matters when your category's buying question is not a phrase anyone searches.
• Search-derived prompts are not shaped like conversational ones. People type three words into a search box and three sentences into an assistant. A prompt reconstructed from the former is a reasonable proxy, not the same event.
• The proxy can drift from the thing. You are measuring a model of demand, and models of demand go stale in ways that are hard to see from inside the tool.

So which one

The choice is not better-versus-worse, it is which error you would rather make.

• If your worry is "we are optimising for questions nobody asks" — take the demand-derived route.
• If your worry is "we need a fixed, versioned exam we can trend for years" — take the controlled question set.

Then ask the question that separates a serious vendor from a dashboard of vibes, whichever route you took: how do you version the question set, and what happens to my history when it changes? If the set changes silently and the number moves, that is the tool moving, not your brand — and every trend line before the change is fiction. The fuller list of buying questions includes the ones that are uncomfortable for us to answer.

What changes when it is shopping visibility

Most of this argument holds for brand-level measurement. Shopping visibility moves the goalposts, because "were we named" is no longer one event — it is three that fail independently: your brand is named, a specific product of yours is surfaced, and the route offered to buy goes to you rather than to a marketplace.

That has two consequences for this choice.

Product-level rates are lower, so the intervals are wider, not narrower. If a brand-level 10.7% needs 252 answers to land within 7.5–15.1%, a product-level rate of a couple of percent needs considerably more sampling to say anything at all. Whichever approach you buy, treat a bare product-visibility percentage with no range attached as a midpoint someone has chosen not to show you the spread of.

Neither approach can tell you the ranking function. Platforms describe their product picks as unsponsored and ranked on relevance, product-data quality, clarity, price competitiveness and merchant trust. That is the stated position and we have no way to verify it from outside. Any vendor claiming to know how the picks are made is guessing with confidence.

What both approaches can do is tell you whether you are surfacing at all, and at catalogue scale the lever that moves it is your own product data's legibility rather than anything you buy.

Where we are the wrong choice

Plainly, because a comparison that concludes "buy ours" is not a comparison:

• You want daily alerts and an operational dashboard. We publish periodic readings, not a live monitoring feed. A prompt monitor fits better.
• You need SKU-shaped product monitoring. We measure at brand and category level and publish it as an index. We do not run product-level shopping monitoring, so both families above fit that job better than we do.
• You want hundreds of prompts across dozens of markets. Our corpora are deliberately small, curated and versioned — a different trade-off, not a better one.

Where we do fit: you want an external, published, methodologically explicit number you can point at, and you want to know how you stand against a stated field rather than only against your own past.