AI Commerce Intelligence

A graded, honest view of how AI is changing commerce.

AI commerce is full of demos, announcements, and confident predictions. Very little of it is graded. This is where Squiggle tracks what's actually happening — separating what's been observed from what's been claimed, and what's claimed from what's merely hoped.

Why this exists

Because most of what's written about AI shopping isn't evidence.

A demo is not a product. An announcement is not adoption. A vendor's usage claim is not independent measurement. Most coverage of AI commerce blurs these together, because urgency reads better than accuracy.

Squiggle already reads real product catalogues to understand how AI shopping assistants decide what to recommend — that's the Product Understanding work in our Research Library. AI Commerce Intelligence is a separate, sibling effort: instead of studying individual product listings, it watches the wider market — AI shopping assistants, agentic commerce, merchant platforms, and how buyers actually behave — and holds every claim to the same evidentiary standard.

What we monitor

Four standing areas of observation.

1

AI shopping assistants and platforms

What's actually launched, changed, or discontinued across the AI systems buyers use to shop — and how that capability is shifting.

2

Agentic commerce

The gap between what's been announced, what's demonstrable, and what's actually available to merchants and used by buyers today.

3

Global markets

How AI commerce is developing across regions — reported with an honest note on how much we can actually verify in each one.

4

Merchant AI-readiness

Whether real merchant catalogues are becoming easier for AI to understand over time — the one dataset Squiggle can measure directly, from its own audits.

How evidence is graded

Every claim carries a grade. No exceptions — including our own.

Four grades, always shown, always in plain English:

Strong

Independently corroborated, not resting on a single interested party's word.

Moderate

Credible and reasonably sourced, but not yet independently confirmed.

Weak

An early signal or a single interested party's claim — flagged as such, never treated as settled.

Contested

Credible sources disagree, or the evidence cuts both ways.

Every graded item also states whether it describes self-reported behaviour, independently observed behaviour, or something still aspirational — a demo, a pilot, an intention. Blurring those three is how most AI-commerce coverage misleads without ever stating an untruth. We keep them separate on purpose.

Evidence, not hype

A quiet month is reported as a quiet month.

We do not manufacture significance. An announcement is covered as an announcement, not as proof of adoption. A survey of stated intentions is not evidence of behaviour. A percentage without a stated base isn't a number we'll repeat as one.

When we're wrong, we say so — publicly, and promptly. We're just as willing to publish a finding that complicates Squiggle's own story as one that supports it. A publication that never corrects itself isn't a careful one; it's one that hasn't been checked.

How we cite sources

No "research shows." Every external claim names what it's standing on.

When a Key Development rests on something outside Squiggle's own audit data — an announcement, a study, a report, a public statement — we cite it, and every citation carries the same fields:

What the source is

Title, and the publisher or author where one is publicly known.

When it's from

Publication date where available, plus the date we accessed it — AI-commerce pages change quietly.

Where to check it

A direct link, so anyone can verify the source themselves rather than take our word for it.

What it's cited for

The specific claim in our own text that source supports — never a bare link with an implied but unstated connection.

A finding drawn entirely from Squiggle's own audit data — our one dataset we can measure directly — is labelled as ours rather than forced through an external citation it doesn't need. Everything else gets one. We don't publish a market claim, a statistic, or a third-party finding without a source a reader can independently check.

Publication status

No issue has published yet — deliberately.

The Monthly Intelligence Review begins as an internal-only practice cycle. We publish only once a run of internal issues has shown the cadence can be kept without missing a month — a stale intelligence page would undermine the exact authority this section exists to build. There is no public weekly feed; the weekly work is an internal log that feeds each month's issue.

The first public Monthly Intelligence Review appears here once that practice period is complete and reviewed. Check back, or see the archive for its current status.

Want the product-level version of this discipline?

Squiggle's free Store Audit applies the same evidence standard to your own catalogue — what's actually missing, not a guess.

Get my free Store Audit →