Product-data AI readability
across 110 manufacturer websites.
An AI cannot accurately read
a website built for people.
A 1.0-inch sensor and 20x optical zoom. A professional model that records in 4K and combines durability with mobility, for news, events and production in a wide range of shooting conditions.
| Image sensor | 1.0-type CMOS sensor (approx. 13.4 million effective pixels) |
|---|---|
| Lens | 20x optical zoom, f=8.8–176 mm (25.4–508 mm in 35 mm terms) |
| Recording format | MOV / MP4, 4K 3840×2160 59.94p / 50p *1Value, unit and note in one string |
| Recording media | SDXC memory card ×2 (UHS-II) |
| Terminals | 12G-SDI, HDMI, XLR 3-pin ×2, and more |
| Power | DC 12 V (AC adapter) / battery 7.2 V (BP-U70) |
| Power consumption | Approx. 18 W (while recording) |
| Weight | Approx. 2.3 kg (body only), approx. 2.9 kg (shooting) |
| Dimensions | Approx. 186 (W) × 218 (H) × 447 (D) mm, excluding protrusions |
| Operating temperature | −5 to 40 °C *2 |
| Dust / drip resistance | Equivalent to IP5X |
| Successor / similar | ZC-4820S |
- 1Model number isn't an identifier
- 2The full specs are in the PDF
- 3Value, unit and note in one string
What the page holds is one continuous string. Where the attribute name ends, where the value begins, and what the unit is are not defined.
* The product, model numbers and values are dummies for illustration.
The same page, held as attribute, value and unit. The blue values can be extracted as numbers, units and identifiers, so an AI can compare them directly with a condition such as “4K 60p” or “under 2.5 kg”.
In the US, nearly 40% of B2B buyers already use AI agents in purchasing. What an AI answers from is product data: whether the specifications are written in a machine-readable form, and whether there is a compatibility table.
* Deloitte, “Agentic commerce: the future of B2B commerce” (2026), a survey of more than 1,000 US suppliers and buyers.
The specifications are written in the tables and body text of product detail pages. What is undefined is where the attribute name ends, where the value begins, and what the unit is.
We measured whether product data on the product detail pages of 110 websites, mostly Japanese manufacturers, is in a form an AI can read.
- 015.5% have Product structured data. At 3.6%, the model number reads as an identifier.
- 02Yet the specifications are written in the HTML of 87% of sites.
- 03Search keywords fell by a median of 66.6% over 18 months; mentions in AI answers rose by a median of 81.2%.
- 04The median product-data AI readability score is 42 out of 100.
The information is there.
An AI cannot read it.
One product detail page was checked per site; the figures are the share of sites that met the condition (n=110).
Specifications are written on nearly 90% of sites, yet the model number is an identifier on only 3.6%. When the model number is mixed into running text, the specifications cannot be tied to the product.
Search is falling,
mentions in AI answers are rising.
Fell at 99% of companies, and at 84% at least halved.
Rose at 96% of companies, and at 39% more than doubled. In total, mentions per month rose 61%, from 28,019 to 45,003.
The median score is
42 out of 100.
Whether product data is organised in a form an AI can compare and recommend. The scoring is based on the major open standards: schema.org Product, the Agentic Product Protocol, the UCP Catalog and others. For B2B, price and stock are out of scope.
n=110 · 4.5% score below 30. Most of the points come from specification fields; fewer than 1 in 10 meet the structured-data and identifier checks.
- ScopeSites in our audits, mostly Japanese manufacturers (184 audits). Product data: the 110 sites where a product detail page could be identified, one page each (September 2026).
- Product dataStructured data, identifiers and specification fields in the HTML. Pages gathered from our audits, sitemaps and search, checked by machine and by eye.
- Search keywordsCommercial-database estimates, February 2025 vs July 2026 (sites under 50 keywords at the start excluded, n=289).
- AI mentionsChatGPT, Perplexity, Gemini and SearchGPT. Monthly averages for September 2025–February 2026 vs March–July 2026 (under 3 mentions at the start excluded, n=233).
- Left outMonths counted only in part, and the presence of llms.txt and /.well-known/ucp, whose check was flawed.
- UpdateRe-measured on 2026-09-29. The earlier version also measured category pages, so scores came out low (median 30 → 42).
- LimitsEstimates are not access logs, and the measurement is a point in time. A site without an identified product detail page does not mean it has none.
The same method,
applied to your site.
Send us a URL and we show you an instant report on the spot, scoring what can be verified from public information. No access needs to be granted.
Submitting sends the domain you entered, and the name that site publishes for itself, to our measurement vendor to count mentions in AI answers. Nothing about you is sent. By submitting, you confirm you may have this site assessed and that the choice of site is yours (terms, terms-v1.2 (2026-10-04)).
