The Index

Issue 1, 5 October 2026

Every two weeks we publish what moved. The underlying data updates weekly on the category pages; this page is the editorial read on it. First issue covers rounds 7 and 8 of Ecommerce Email & Retention — the first two rounds measured the same way.

Movers

All figures measured.

# Product ChatGPTPerplexity Change, cross-engine
1 Omnisend 61% 59% 0.0 pt
2 Klaviyo 45%† 56% -2.5 pt
3 Brevo 35% 37% +4.4 pt
4 Mailchimp 29% 34% +3.4 pt
5 MailerLite 27% 29% -1.0 pt
6 ActiveCampaign 23% 22% +3.9 pt
7 Shopify Email 18% 13%† -1.0 pt
8 Sender 8% 20% +2.0 pt
9 Shopify Messaging — 21%† +4.9 pt
10 GetResponse — 19% +2.5 pt
11 Drip 14% 4% -5.4 pt
12 Kit 11% 6% +0.5 pt

— not recommended in any sample

† Corrected 7 October 2026: extraction. The stored answers were not changed; the judgment of whether a product was recommended in them was re-made, by the extractor or by a person reading the answers, and the figures marked † were recomputed. How corrections work

† Corrected 5 October 2026: linking. Product names in this round's answers were re-linked to the products they refer to, and the figures marked † were recomputed. No question was re-asked. How corrections work

Change is week over week, round 7 to round 8, both measured with extractor version 5. Ranked by cross-engine Pick Rate; each engine column is that engine’s own figure. All 53 products

Scanned 5 October 2026. 204 samples across ChatGPT and Perplexity, round 8. Change against round 7, scanned 29 September 2026. Methodology v2026-09-29.1.

Where the engines disagree

We never average the two engines. A product can be a Perplexity favourite and invisible on ChatGPT.

All figures measured.

Product ChatGPTPerplexity Gap
Shopify Messaging — 21%† 20.6 pt
GetResponse — 19% 18.6 pt
Sender 8% 20% 11.8 pt
Klaviyo 45%† 56% 10.8 pt

— not recommended in any sample

Products whose Pick Rate differs by 10 points or more between the two engines, largest gap first. The gap is computed from unrounded rates.

Scanned 5 October 2026. 204 samples across ChatGPT and Perplexity, round 8. Methodology v2026-09-29.1.

Sources the assistants cited

#SourceCitations
1 omnisend.com 161
2 apps.shopify.com 119
3 shopify.com 106
4 techradar.com 105
5 brevo.com 84

omnisend.com is both the most-recommended product and the most-cited source this week. We do not yet know which way that runs.

Scanned 5 October 2026. 204 samples across ChatGPT and Perplexity, round 8. Methodology v2026-09-29.1.

Consensus

How often an assistant agrees with itself — the overlap between its three answers to the same question. High consensus means the category has settled; low means it is still open.

All figures measured.

Engine Round 8 Round 7 Change
ChatGPT 84% † 89% -4.7 pt
Perplexity 81% 82% -1.3 pt

† Corrected 7 October 2026: extraction. The stored answers were not changed; the judgment of whether a product was recommended in them was re-made, by the extractor or by a person reading the answers, and the figures marked † were recomputed. How corrections work

Scanned 5 October 2026. 204 samples across ChatGPT and Perplexity, round 8. Round 7 figures from the 29 September 2026 scan. Methodology v2026-09-29.1.

Corrections and incidents

  • 1 October 2026. Extraction correction: 8 products’ round 7 figures were recomputed and marked † on their pages.
  • 5 October 2026. The scheduled scan did not fire. It was run manually three hours later with identical settings, and the schedule was moved off the top of the hour.
  • 5 October 2026. Linking correction: 5 products’ round 8 figures were recomputed after product names were re-linked; they are marked † in the tables above.
  • 5 October, late evening — a delayed automatic run re-read every answer. The scheduled scan that failed to start at 02:23 UTC fired eight and a half hours late. It was correctly blocked from scanning again, but the block did not stop the later steps, so all 204 stored answers were re-read by the extractor. Reading the same text twice, it reached a different judgment on three answers, and three products’ figures changed on the site without a correction marker: Klaviyo (ChatGPT 44% → 46%), Judge.me (4% → 3%) and Yotpo (2% → 1%). On 7 October we read those answers ourselves and ruled on each: Klaviyo is 45%, Judge.me 3%, Yotpo 2%. All three now carry †. A blocked run can no longer touch anything, and a snapshot that changes a published figure without a marker is now refused before it reaches the site.

How we handle corrections is set out in the methodology.

Reading this issue

One week is one data point. We do not yet publish a noise threshold for week-over-week change; read the direction, not the decimal.