Imagine checking your AI-driven ranking report twice within an hour and getting two completely different results. That’s exactly what recent research highlights: AI visibility scores vary significantly between consecutive runs, making a single measurement highly misleading. For small and medium businesses on Cyprus or across the EU — especially those relying on local SEO to reach Greek-, Russian-, or English-speaking clients — this instability can lead to poor decisions and wasted budgets.

What the Data Shows

A new academic paper, published by researchers studying large language model output consistency, demonstrates that over 40% of observed ranking shifts are statistically indistinguishable from noise. The authors propose a ‘stopping rule’ — a method to determine when enough repeated measurements have been taken so that the average ranking becomes reliable. In practice, this means you need at least 5–7 separate runs before trusting a position change.

Practical Implications for Cyprus-Based Businesses

  • GDPR & localisation: Google’s AI overviews and third-party tools like Semrush, Ahrefs, or Mangools now generate visibility scores for multilingual queries (en/ru/el). A single snapshot might show your Limassol law firm ranking #1 for ‘corporate lawyer Cyprus’ in one run and #8 in the next — both within minutes.
  • Budget waste: Acting on a one-time drop could trigger unnecessary SEO retainer increases (typically €500–€2,000/month in Cyprus). The study recommends ignoring any change unless it persists across three independent checks.
  • E-commerce CRM alert: If your store’s product feed relies on AI-generated visibility metrics for Google Shopping optimisation, applying a single run’s data may skew inventory priorities — especially for seasonal items like olive oil or tourist services.

How to Respond as a Business Owner

First, demand from your agency or internal team that they run every AI visibility report at least five times at random intervals. If they can’t, consider switching tools. Second, look for trends over weeks, not hours. A Cypriot restaurant chain that saw a sudden AI visibility drop in mid-September 2024 but then recovered within three days was simply experiencing noise — no action needed. Third, where possible, prioritise traditional ranking data (Google Search Console, manual SERP checks) over AI-generated scores until the industry adopts standardised stability metrics.

The bottom line: treat AI visibility numbers like a weather forecast — useful as a trend, dangerous as a fact. For a studio like 62px, that means building your site strategy around consistent user intent signals rather than volatile machine outputs.

Source: Search Engine Journal, @MattGSouthern, reporting on new academic research. Original post: “AI Visibility Rankings Aren’t Stable – New Research Shows It’s Mostly Statistical Noise.”