Agentic Landmark · Entry instrument
Agent Catalog Read
What an agent can extract from your product pages, product by product. A structural read for agentic commerce.
The Agent Catalog Read shows what an agent can and cannot extract from your product catalog, one product at a time. It produces no score. It runs before the Agent Readiness Index and answers the question a score cannot: where, exactly, the chain breaks. The Read is free.
The layer it inspects
An agent reads the page it holds.
An agent does not see your brand. It sees the document in front of it at the moment it reads, and it builds its answer from the fields in that document. The Read inspects that layer and nothing else:
- whether each product carries a populated offer, with price, availability and currency;
- whether the content that decides a comparison is present before scripts run;
- whether schema arrives in the page the server sends, or only after scripts have run;
- whether a product carries one identity across its canonical URL, its schema and its feed.
First of all, it checks whether a browser asked for a product address stays on that product page.
The gap is common. Adobe's Q2 2026 AI Traffic Report scored product detail pages at 63.5 on its AI Citation Readability Score. That is among the top 20 percent of companies by AI visit share, and it is still the second lowest of any commercial page type in that group, after search results pages at 59.7.
Can an agent get in, 2 checks
- Crawler permissions
- Sitemap and catalog scale
Can it read the product, 3 checks
- Product schema populated
- Content without JavaScript
- Schema injection timing
Can it match the product elsewhere, 3 checks
- Canonical and identity consistency
- Feed present and valid
- Attribute completeness
Recorded, not counted, 1 check
- llms.txt
Where it sits
One floor under three channels.
- Search engine optimization (SEO) decides whether a page ranks.
- Generative engine optimization (GEO) decides whether an AI engine cites it.
- Answer engine optimization (AEO) decides whether an answer engine surfaces it.
The outcomes do not transfer between channels. Boston Consulting Group found only 8 to 12 percent overlap between traditional search results and AI-generated answers.
The inputs are shared. Crawler access, sitemaps, structured data, content present before scripts run and a consistent product identity feed all three. They also feed a fourth use that agentic commerce adds: an agent putting your product into a comparison.
The Read measures that shared floor. It measures none of the outcomes above it.
What it is not
Not a citation count, not an SEO audit, not a score.
The Read asks no answer engine anything. It counts no mentions and benchmarks no share of voice.
When a buyer asks a category question, an engine can do one of two things. It can quote a source, which it assembles from prose. Or it can build a comparison, which it assembles from fields, and it can only fill a cell it can parse. A brand can win the quote and be missing from the comparison.
Citation tools measure the citation surface. This Read measures the comparison surface.
The Read is not an SEO audit either. It does not look at rankings, keywords, backlinks or page speed.
Most readiness scanners return a grade. A grade needs weights, and weighting one finding against another is a judgment about your business that cannot be made from your public site alone. So the Read counts, and every count carries its denominator.
Method
Nothing is reported absent from a single look.
The Read is a census, not a sample. Every product URL the traversal reaches in the locale read is read, and the locale is named. Each product address is requested twice: once as a plain request, once by a browser. Where the two disagree, the affected checks are recorded as undetermined, with the reason, rather than failed. Anything the Read could not determine is stated above the findings it limits, never left out.
Every fetch is stored with a content hash, so a second run is a page-against-page comparison, not a fresh opinion.
Two exercises, in order
The Read finds the break. The ARI measures what it costs.
A composite averages across dimensions. An average is a claim about the whole; a break is a claim about one link. Strength further down the chain cannot make up for a break at the top of it.
The ordered read:
- Held by a browser
- Read as data
- Compared on its attributes
- Chosen from a shortlist
- Bought
This Read measures the first three links. It does not measure the last two.
| Agent Catalog Read | Agent Readiness Index Assessment | |
|---|---|---|
| Answers | Can an agent get to and read your products, page by page? | How exposed is your revenue, and where first? |
| Produces | Counts with denominators; no score | Composite, dimension scores, risk band |
| Reads from | Your public site | Your business: channel mix, marketplace share, category exposure |
| Acted on by | Your engineering team | Leadership and budget holders |
| Cost | Free | Scoped engagement |
The Read's findings are evidence the ARI weighs, mostly under Structured Data Maturity. The ARI adds what a public site cannot show. It is worth commissioning once the chain holds; scored before then, it measures dimensions on a chain that stops several steps upstream.
What you receive
A document your engineering team can act on.
You get one document, organized product by product. Leadership reads the findings in order. Engineering opens the technical record beneath them, where every observation can be reproduced from a URL you can check yourself.
A fix step appears only where a finding was determined, and it says what to check on which page. Where the Read could not determine something, it says so and recommends nothing.
The cost panel is left blank. The numbers that would price the findings are agent-referred sessions, conversion rate and order value, and they are yours. The Read will not estimate them from public signals.
The Read covers one locale of a catalog published in a sitemap. It covers structure only. Where a brand surfaces in answer engines is a separate measurement and not part of it.