---
title: "Week of 14 September 2026: one Index release, two question shapes"
description: "The Machine Relations Index release of 18 September 2026 split one category into a worth question and a comparison question. I read those as different founder decisions."
canonical: "https://jaxonparrott.com/brief/week-of-2026-09-14"
last-updated: "2026-11-04"
---

# Week of 14 September 2026: one Index release, two question shapes

This brief covers the Machine Relations Index release dated 18 September 2026, identifier mri_score_v2.0+2026-09-18+8fa38e54dd0a, window 10 May 2026 through 18 September 2026. The dataset this edition is based on is the Machine Relations Index deposit at doi:10.5281/zenodo.23652423. I am not republishing the tables. I am saying what I do with two counts from the AI Visibility and GEO category.

The worth segment in that release had 137 observed answer runs and 467 cited sources. The comparison segment had 132 observed runs and 417 cited sources. reddit.com was cited in 38 of the 137 worth runs. Those figures are the release's, and they are the only figures this brief uses.

What I take from the split is that a founder cannot brief one program and call it AI visibility. A worth question and a comparison question are different retrievals. A source that leads one does not, on this release, tell you the other. I would not let a team average the two segments into a single score for the week. Averaging hides which buyer question is unanswered.

The operating consequence is narrow. If the buyer's question is whether a category of tool is worth paying for, the worth segment is the one I open, and I look at the class of the sources, not at a vendor's claim about its own product. If the buyer's question is a head-to-head, I open the comparison segment and I do not import the worth ranking into it. The Index page for the worth question is the table. This brief is the decision not to collapse the table.

I will replace this edition when a later Tuesday release is in the observed file. Until then, the 18 September release is the one I am willing to cite, and only with its identifier next to the counts.

Based on the [Machine Relations Index dataset](https://doi.org/10.5281/zenodo.23652423), release `mri_score_v2.0+2026-09-18+8fa38e54dd0a`.