Why we built steek honestly: confidence intervals over vanity metrics

The problem: the number is lying to you

Open most AI-visibility tools and you get a big, calm number: your visibility is 42. It feels authoritative. It is also, quietly, a fiction — because the thing it measures doesn't hold still. Ask an AI engine the same question twice and it can pull a different set of sources each time. Profound's tracking put numbers on this: 40–90% of the domains an engine cites can change when the same question is re-asked over time. A metric computed on top of that much movement, reported as a single clean figure, is telling you a certainty it doesn't have.

The damage isn't abstract. A single number invites you to react to every wiggle: visibility dropped from 42 to 39, so you rewrite a page, change a supplier, blame a campaign — when the “drop” was sampling noise that would have reverted on its own. You end up optimizing against randomness, which is worse than doing nothing because it costs time and convinces you that you understand a system you don't.

The decision: put the interval on everything

So the first real product decision wasn't a feature — it was a refusal. We refused to ship the calm single number. Every metric in steek is reported with a confidence interval: not “your visibility is 42” but “your visibility is 42, likely between 36 and 48.” That's a little more to read. It's also the difference between a measurement and a guess dressed as a measurement. When the interval is wide, you know to sample more or wait; when two weeks' intervals don't overlap, you know the move is real and worth acting on.

This is genuinely unusual in the category. Plenty of good tools report AI visibility as a point estimate; across the ones we track in our roundup of AI-visibility tools, the honest limitation we keep noting is exactly this — a single number with no stated uncertainty. We're not claiming the others are dishonest; we're claiming that quantifying your own uncertainty is a real choice, and most tools haven't made it. It's the one thing we most want to be known for.

The rule that comes with it: no fabricated data

An interval is worthless if the inputs are invented, so the second half of the decision is a hard rule: we don't fabricate data. In practice that means three concrete things. We don't invent customer counts or testimonials. We don't publish ratings or reviews we didn't earn — including in our structured data, where inventing an AggregateRating would be both a brand violation and against Google's own policy. And we don't render a competitor's demo-gated price as a confident headline figure — where a rival hides its pricing, we mark our numbers as reported or estimated and stamp when we checked, rather than laundering a guess into a fact. If we're unsure, we hedge it or leave it out. That rule governs this blog too: every external statistic here is attributed to a named source you can check.

Why honesty is a feature, not a tax

It would be easy to frame all this as a cost — the responsible-but-boring choice that makes the product harder to skim. We think that's backwards. The interval is the feature. It's the thing that lets a small business owner look at their AI visibility and make a good decision instead of an anxious one. A dashboard that only ever tells you a confident story is optimizing for how it feels to look at, not for the quality of what you do next. We'd rather build the tool you can trust at 2am when the number moved and you have to decide whether it matters.

There's a business argument too, and we'll own it: honesty is our wedge. We're a small, EU-hosted tool going up against better-funded incumbents. We can't out-spend them, but we can be the one that refuses to flatter you with a vanity metric — and in a young, hype-filled field, being the boring honest one is a position worth holding. The GEO and AEO space has enough confident guesswork; we'd rather be right and hedged than loud and wrong.

The facts we're willing to state plainly

Honesty cuts both ways — it also means being concrete about what steek is. It tracks how often ChatGPT, Perplexity, Gemini and Google AI Overviews mention, cite and recommend your brand across a prompt set you choose. It shows real monthly search volume behind each prompt rather than a 1–5 bar. For stores, it ties per-SKU AI Shopping visibility to real revenue in GA4. It's EU-hosted. Pricing is flat: from €19/mo billed annually (€29.90 monthly), every major engine included, no per-engine upcharge, with a 10-day free trial and no credit card. And where we're honestly not the strongest — day-one historical backfill, enterprise-scale prompt volumes — we say so on the comparison pages rather than pretend otherwise.

See it for yourself

The most honest thing we can do is let you check. Run the free AI-visibility checker on the homepage — no signup — and you'll see the interval on the very first read. If you want the fuller picture, compare us honestly against the field in the best AI-visibility tools and our side-by-side comparisons, or see how this plays out for a small business specifically.