AI recommendation movers: how sticky is the top?

A single reading of where AI ranks your brand feels authoritative. But a rank is a point in time, and the only way to know whether it means anything is to take a second reading and Compare.

So we did. We snapshotted AI brand rankings on Nov 30, 2025 and again on Apr 29, 2026, then compared them on a fixed panel of 2,029 brands that were ranked inside the top 3,000 in both windows. That panel is the basis of this report: every mover below is a brand that was clearly visible at both ends, so its movement is real and not a side effect of the index seeing more of the web over time.

Start at the very top, where ranks are supposed to be most stable. Of the brands ranked in AI's top 100 in November, only 58.2% were still in the top 100 by April. The other 41.8% churned out. The leaderboard AI shows buyers is sticky, but it is not frozen, and nearly half of it turned over in a single season.

Look at the wider tiers and stability improves, though not by much. Top-500 retention is 65.9% and top-1,000 retention is 66.0%, versus 58.2% for the top 100. The sharpest competition is right at the top: the smaller and more prestigious the slot, the more contested it is.

On the full visible panel the Nov-to-Apr rank correlation is 0.64, the kind of number that looks orderly until you read the spread underneath it. The median brand moved 367 ranks, and 38.9% of brands moved more than 500 ranks. Most brands did not hold their place; they drifted, often a lot.

A small set of brands sat almost perfectly still. Amazon held at #3, YouTube slipped one to #6, HubSpot from #10 to #11, FanDuel stayed exactly at #44. Add Zendesk, eBay, Pipedrive, DraftKings, and Notion (which nudged up to #26) and you have AI's blue-chip set: the brands the model won't forget.

At the other extreme, some brands climbed from deep in the panel into prominence. TikTok Shop went from #2,861 to #373, Autodesk from #2,481 to #75, Freshworks from #2,207 to #42. Cash App, VMware, Fanatics, and SourceForge all made similar jumps. These are not new brands the index just discovered; they were visible in November and moved far up the list.

The clearest category-wide movement in the data is in AI infrastructure itself. The LLM-ops names, the tools teams use to run and monitor AI models, rose together: vLLM climbed from #2,986 to #787, alongside Braintrust and Arize Phoenix. When a whole adjacent category moves in the same direction, it is a sign AI is rewriting which tools it considers standard, not just reordering individual brands.

Movement runs both ways. Vrbo fell from #38 to #2,953, one of the steepest drops on the panel. LendingTree, Clover, FindLaw, Xbox, and MasterClass all slid from prominent slots deep into the tail. A high rank in November was no guarantee of a high rank in April.

One pattern in the fallers is hard to miss. The reference authorities are losing ground: Stack Exchange, Consumer Reports (#110 to #2,178), Healthline (#53 to #2,033), the FDA (#273 to #2,375), and Cleveland Clinic all fell sharply. When AI recommends a brand, it is tilting away from neutral reference institutions and toward commercial brands that can actually be bought.

The flip side of the churn is who arrived. Several brands moved from outside the top 300 straight into the top 100: Atlassian #495 to #20, Oracle #319 to #43, NVIDIA #565 to #46, Tesla #409 to #69, plus SAP, Intuit, and ChatGPT itself (#809 to #76). The pattern is enterprise software and AI consolidating at the very top of what the model recommends.

Over this window the number of brands the index tracked grew from roughly 204,000 to 551,000. That growth is the reason we did not simply rank every brand by raw rank change. A brand that appears to move from #480,000 to #400 mostly tells you the index started seeing it, not that AI changed its mind.

So genuine movement is measured only on the 2,029 brands ranked inside the top 3,000 in both snapshots. Every riser and faller above cleared that bar. Brands that only appear to move because the index now covers more of the web are excluded by design, which is what makes the changes trustworthy.

The headline is uncomfortable for anyone who treats an AI-visibility score as a verdict. The top of AI's rankings moves enough that a single reading is closer to noise than signal: four in ten of today's leaders won't be leaders next season.

What holds up is the trajectory. The blue chips earned their place by showing up reading after reading; the risers are visible weeks before their gains are obvious; the fallers slide before anyone notices. What matters is the trend across snapshots, not the rank in any one of them, which is exactly why this is a reading we re-run.

We took two snapshots of daily AI brand rankings, one on Nov 30, 2025 and one on Apr 29, 2026, drawn from an observed sample of answers across ChatGPT and Google AI Overviews. For each snapshot every tracked brand has a rank by how prominently AI surfaces it. We then restricted the comparison to the 2,029 brands ranked inside the top 3,000 in both windows, so that a brand's rank change reflects genuine movement rather than the index covering more of the web over time.

Retention is the share of brands in a given tier in November still in that tier in April. Rank correlation, median absolute rank change, and the share moving more than 500 ranks are computed on the full visible panel. This is the legacy ChatGPT and Google AI Overviews era; figures describe the two engines in scope, not Perplexity, Gemini, or Copilot, and movement is restricted to brands visible in both windows. It is a reading we plan to re-run each quarter.