# What you measure when position one is worth nothing > Rankings are losing clicks to AI answers. How to measure AI visibility instead — three metrics that are enough for a small business. For twenty years, position one on Google was the metric everything hung on. It is losing its meaning right now — not because rankings stopped mattering, but because an answer has moved in between the ranking and the customer, and the customer never clicks it. ## Why position is losing its meaning Position measures where you stand in a list — not whether anyone still looks at the list. And that is exactly what shifted: in the US in 2024, around **58.5 per cent** of all searches ended without a single click into the open web; by early 2026 it was **68 per cent** (SparkToro with Datos, and Similarweb). For Europe the 2024 figure was 59.7 per cent. On top of that comes the direct effect of AI summaries: where an AI overview sits above the results, the organic click-through rate fell from 1.76 to **0.61 per cent** across an analysis of 25.1 million impressions — a drop of 61 per cent (Seer Interactive). Your position can stay exactly where it was. It's just that nobody clicks any more. ## What takes the place of position What replaces position is the **mention**: how often an AI names your business in its answer when someone asks a question you would be the answer to. The trade term is Share of Model Voice, SOMV for short — the share you hold in a machine's answers. The difference is fundamental. A position describes a place in a list that may no longer be opened at all. A mention describes whether you appear in the answer the customer actually reads. ## How do you measure a mention? You measure it by asking machines exactly the questions real customers ask — regularly, in the same words, over months — and recording whether and how your own business shows up. Three things need to be kept apart: - **Is the business named?** Yes or no, per question and per model. - **How is it named?** Mentioned in passing, actively recommended, or given as the first choice? - **Who gets named instead?** The answer to that is often more instructive than your own mention. ## The one distinction that explains everything What matters is **whether the model looks things up when asked, or answers from memory.** These are entirely different disciplines, and mixing them leads to wrong conclusions. Our own measurement from 10 August 2026 shows the difference in a single line. Ten customer questions, several models, 109 measurement points: - With **live search**: 9 of 30 probes name us. - From **model knowledge**: 0 of 79. Which means: whoever searches today finds us. Whoever answers from memory does not know us. That is not a defect but a timeline. The first figure responds to content within days. The second moves over months, because it depends on how often and how unambiguously a brand appears across the web before a model is trained. Anyone disappointed after four weeks was looking at the wrong metric. ## Three metrics instead of one For a small business, three numbers are enough, and none of them is average position: 1. **Mentions in AI answers** — against a fixed, short list of real customer questions. Kept separate for live search and model knowledge. 2. **Clicks on the pages that sell** — not impressions, not total clicks. Only the pages with an offer behind them. 3. **Enquiries** — the only number that counts in the end. Everything before it is an interim reading. If you want to see how that works in practice, [this self-check](https://btlabs.dev/en/posts/does-ai-really-cite-me-self-check) shows the concrete route — this article answers *what* you measure, that one *how* you verify it. ## The honest limit None of these metrics tells you **why** an AI names someone. They show that it happens, and from when it changes. The cause behind it — what you offer, how unambiguously it is described, what others say about you — remains a question you answer rather than measure. And a figure drawn from ten questions to a handful of models is a sample, not market research. It is good for spotting movement. It is not good for celebrating percentage points. If you want to know which questions would be the right ones for your business, [talk to us](https://btlabs.dev/en/contact) — that's half an hour of work and the rest runs by itself. ## Sources - SparkToro with Datos (2024) and Similarweb (early 2026): share of searches without a click into the open web, US 58.5 % → 68 %, EU 59.7 % (2024) - Seer Interactive: organic click-through rate with an AI overview displayed, 1.76 % → 0.61 % across 25.1 million impressions - Bain/Dynata (December 2024, n = 1,117): around 60 % of searches end without a click through - Own measurement, btlabs.dev: Search Console data 13/07–09/08/2026 (955 impressions, 14 clicks); citation monitor of 10/08/2026 (10 questions, 109 measurement points) --- Source: https://btlabs.dev/en/posts/what-you-measure-when-position-one-is-worth-nothing Last-Modified: 2026-09-06T07:00:00.858Z Languages: [de](https://btlabs.dev/llms/de/posts/was-du-misst-wenn-platz-1-nichts-mehr-zaehlt) · [it](https://btlabs.dev/llms/it/posts/cosa-misuri-quando-il-primo-posto-non-vale-piu) See also: [llms.txt](https://btlabs.dev/llms.txt) · [ai.txt (Policy)](https://btlabs.dev/ai.txt) · [identity.json](https://btlabs.dev/identity.json)