Demo 08 · AI-Ready Product Data Audit · free · no sign-up

Your buyers’ AI is already shopping. Can it read your catalog?

AI buying agents compare suppliers on structured attributes and explicit pricing. Specs in a PDF, “call for price” and a search that needs JavaScript make you invisible to them. Watch one try to buy an end mill from a sample distributor, see the catalog scored on our 100-point rubric, then apply the fixes and run it again.

A metalworking buyer types into their AI assistant

“3/8 4-flute AlTiN carbide end mill, 1-1/8 LOC, for aluminum. Price and stock, need it by Friday.”

An MRO buyer types

“Nitrile gloves, large, price per case of 10.”

See how it does
Watch the AI buyer shop Score a page of yours

Sample data · fictional distributor Simulated agent · no live AI call · nothing sent

Agent console · the buyer’s AI shops Meridian Sample Supply (fictional)

“3/8 4-flute AlTiN carbide end mill, 1-1/8 LOC, for aluminum. Price and stock, need it by Friday.”

Score22Agent-invisible 11 failed · 3 unsure

Showing meridian-sample.example/p/HCY-375-4-AT

meridian-sample.example/p/HCY-375-4-AT

What a person sees

Halcyra Carbide

Halcyra 3/8" 4FL End Mill

PDF Spec sheet

Call for priceCheck availability

What the agent’s fetcher receives no JavaScript run

  1. GET /p/HCY-375-4-AT → 200 OK
  2. <title>HCY-375-4-AT | Meridian Sample Supply</title>
  3. no Product markup on this page 4.1
  4. <h1>Halcyra 3/8" 4FL End Mill</h1>
  5. <p class="brand">Halcyra Carbide</p>
  6. no manufacturer part number, no GTIN 3.3
  7. <div id="specs"></div> (filled in by specs.js) 2.1
  8. <a href="/pdf/HCY-375-4-AT.pdf">Spec sheet (PDF)</a> 5.4
  9. <button onclick="quote()">Call for price</button> 7.1
  10. <div id="stock">Check availability</div> (log in + stock.js) 7.2
  11. no workpiece-material or coating guidance 8.3

  1. 01 Searches its index for Meridian end mills

    Failed: Nothing. Meridian’s robots.txt keeps OAI-SearchBot, Claude-SearchBot and PerplexityBot out of /products/. Meridian is only in this run because the buyer typed its name.

    1.1 Proven

  2. 02 Opens Meridian’s site search for “3/8 4 flute end mill”

    Failed: The page says “Loading…”. The results are drawn by JavaScript, and this agent does not run JavaScript.

    2.3 Emerging

  3. 03 Checks the sitemap for product pages

    Unsure: It lists 3 of every 5 products, all dated the same day, so nothing says which prices changed.

    1.2 Proven

  4. 04 Cross-checks the shopping feed

    Failed: No product feed, so the shopping surfaces only know what their crawlers happened to read.

    10.2 Proven

  5. 05 Reads the Halcyra Carbide HCY-375-4-AT page

    Failed: The name and photo are there. The spec table, price and stock are filled in by script after the page loads, so the agent got empty boxes.

    2.1 Proven + emerging

  6. 06 Looks for the flute count

    Failed: Couldn’t read the flute count: it’s inside a PDF spec sheet.

    5.4 Emerging

  7. 07 Looks for Product structured data

    Unsure: Two of three pages have none, and the third contradicts its own page.

    4.1 Proven

  8. 08 Reads the Trevarn Endmill page

    Failed: The title is “3/8 X 1-1/8 X 3 4FL EM ZRN”. Is 1-1/8 the length of cut or the shank? Is 3 in inches? No labels, no units, so it can’t be matched to “1-1/8 LOC”.

    5.2 Proven

  9. 09 Checks the part identity

    Failed: No manufacturer part number and no GTIN, so it can’t confirm this is the tool the buyer’s system already knows.

    3.3 Proven

  10. 10 Reads the Kestova Carbide page

    Unsure: The page says “In stock”; its data says OutOfStock and lists the diameter as 0.000. A careful agent trusts neither.

    5.5 Emerging

  11. 11 Looks for a price

    Failed: “Call for price.” An agent can’t call. It needs a number, or a rule it can apply.

    7.1 Proven

  12. 12 Looks for stock

    Failed: “Check availability” needs a login and JavaScript. Stock unknown.

    7.2 Proven

  13. 13 Works out whether it arrives by Friday

    Failed: No lead time anywhere as data. It can’t promise Friday.

    7.3 Proven + emerging

  14. 14 Checks the tool suits aluminum

    Failed: Nothing says which coating suits which material, so it can’t warn the buyer that AlTiN is a poor choice for aluminum.

    8.3 Emerging
Agent’s rankingNot shortlisted

Score 22 / 100 · 11 failed, 3 unsure

It recommended another supplier.

The agent couldn’t confirm that Meridian sells a 4-flute 3/8 end mill with a 1-1/8 length of cut, let alone its price or a Friday ship date. It answered with a supplier whose data it could read.

And Meridian was only considered because the buyer typed its name. A buyer who didn’t would never see it.

Second buyer · MRO · same agent, same fixes

“Nitrile gloves, large, price per case of 10.”

0 of 6 read
  1. Searches its index for Meridian nitrile gloves Failed: Nothing: the search crawlers are kept out of /products/, so the gloves are not in its index. 1.1
  2. Opens Meridian’s site search for “nitrile gloves large” Failed: The search page says “Loading…” until JavaScript runs. 2.3
  3. Looks for the large size Unsure: One page for all four sizes. Large has no SKU or URL of its own, only a dropdown option. 3.1
  4. Works out what “a case of 10” is Failed: The page says “BX”. Nothing says how many gloves are in a box or how many boxes are in a case. 7.4
  5. Looks for the case price Failed: “Call for price.” 7.1
  6. Looks for stock Failed: “Check availability” needs a login and JavaScript. 7.2

Not shortlisted It can’t price “a case of 10” from this catalog, so it quotes another supplier.

The fixes

Apply the fixes. Watch the data change, then run the agent again.

Owner’s call Publishing price and stock is a business decision, not an engineering default. In an engagement each one becomes a Decision Card for the person with the authority.

The rubric · 10 categories · 40 tests · 100 points

Scored the way the platforms document it.

Each test is tagged by how well its importance is established. Proven means the platform or standard owner documents it (Google, OpenAI, Anthropic, schema.org, GS1, ISO). Emerging means vendor studies or our own inference, so those tests are capped low: a client is never sold a fad. Proven tests carry 57 of the 100 points, emerging ones 31, and 12 are split.

  1. 90–100 Agent-preferred
  2. 75–89 Agent-ready
  3. 60–74 Agent-visible
  4. 40–59 Agent-fragile
  5. < 40 Agent-invisible
01 AI access & discoverability 5.5 / 10 9.5
AI access & discoverability: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
1.1 Answer-engine bot accessAn agent cannot cite or buy what its fetcher is refused. Proven 4 2 Googlebot and Bingbot get in. robots.txt shuts OAI-SearchBot, Claude-SearchBot and PerplexityBot out of /products/.
1.2 Sitemap coverage and freshnessFreshness signals decide how fast a price or stock change reaches the engines. Proven 3 1.5 The sitemap lists 3,112 of 5,240 SKUs, and every lastmod carries the same date.
1.3 One canonical host everywhereSplit hosts split citations and can send agents to staging URLs. Proven 2 2 One host; the other redirects. Already right.
1.4 Change propagationBing and Copilot pick up changes in hours instead of weeks. Proven + emergingIndexNow proven (Bing) · llms.txt emerging, worth at most 0.5 1 0 No IndexNow; no llms.txt.
02 Machine-readable rendering 1 / 8 6.5
Machine-readable rendering: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
2.1 Product facts in server HTMLThe major AI crawlers do not run JavaScript. Proven + emergingProven for Google · emerging-strong for other AI crawlers 3 0 Specs, price and stock are filled in by script after the page loads.
2.2 SpeedSlow pages are crawled less. Proven 2 1 First byte is fast; phone LCP is 3.9 s.
2.3 Search and filters work without JSAgents answer “find me a 3/8 4-flute…” by fetching a search URL. “Loading…” means they guess elsewhere. EmergingEmerging (strong) 2 0 Site search answers “Loading…” until JavaScript runs.
2.4 Answer-shaped contentAnswer engines quote passages that answer the question directly. Proven 1 0 No selection guides or answers in HTML.
03 Product identity & identifiers 4.5 / 10 10
Product identity & identifiers: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
3.1 One stable URL per orderable SKUAgents key on the part number. An unreachable variant is invisible. Proven 3 1.5 Coated variants share one URL behind a dropdown.
3.2 Brand and MPNBrand plus MPN is what Google requires when there is no GTIN. Proven 2 1 Brand on every page; manufacturer part number on none of the three.
3.3 GTINWithout a GTIN, Google gives “limited visibility”. Marketplaces join records on it. Proven 3 0 No GTINs anywhere.
3.4 Entity consistencyContradictions lower an engine’s confidence and invite hallucination. Proven 2 2 Company, lines and contacts agree everywhere. Already right.
04 Structured data (schema.org) 2.5 / 12 12
Structured data (schema.org): each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
4.1 Product JSON-LD on every SKU page, validIt is the most reliable machine-readable layer. Proven 3 1.5 Product markup on 1 of 3 sampled pages, and it contradicts the page.
4.2 Merchant-listing completenessRequired for Google merchant listings and shopping surfaces. Proven 3 0 No price, image or availability in the markup.
4.3 Variants modeled correctlyWrong modeling makes engines quote the wrong price against the wrong part. Proven 2 0 Variants modeled as one Product with several SKUs.
4.4 Specs as typed PropertyValuesAn agent can compare 0.375 INH with 9.525 MMT. It cannot reliably compare the string 0.375″. Proven + emergingVocabulary proven · use by AI emerging 2 0 Specs, where present, are strings like “3/8"”.
4.5 Semantic correctnessMisused properties put false facts into an agent’s answer. Proven 2 1 Visible breadcrumbs, no BreadcrumbList.
05 Attribute data quality 3 / 14 14
Attribute data quality: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
5.1 Complete spec set per product family“3/8 4-flute 1-1/8 LOC” is answerable only if every one of those is a filled field. EmergingEmerging (strong) · House, Industrial Distribution, Sep 2026 4 2 About 60% of end-mill attributes filled.
5.2 Numeric, unit-explicit, dual unitsAgents convert and compare units. Mixed formats break filters. Proven 3 0 Fractions with no units or labels: “3/8 X 1-1/8 X 3”.
5.3 Standard attribute vocabularyCAM systems, PIMs and agents map fields without guesswork. Proven 3 0 In-house field names only.
5.4 Specs not locked in PDFsPDF-only specs are skipped. Disagreement lowers trust. Emerging 2 0 Flute count, helix and LOC live only in PDF spec sheets.
5.5 Data integrityOne absurd value can get a whole catalog treated as low-trust. Emerging 2 1 A “0.000” diameter, and a page that says In stock while its data says OutOfStock.
06 Classification & taxonomy 1 / 6 6
Classification & taxonomy: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
6.1 Hierarchical breadcrumbsIt gives agents the “is-a” path for each SKU. Proven 2 1 Visible breadcrumbs, not as data.
6.2 Standard classification codesAriba and Coupa catalogs require UNSPSC. Proven 2 0 No UNSPSC, ETIM or ECLASS codes.
6.3 Application taxonomy as dataBuyers ask by application (“for 6061”), not by geometry. EmergingEmerging (strong) 2 0 No workpiece-material facets.
07 Commercial data 2 / 14 14
Commercial data: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
7.1 Explicit pricingOwner’s call“Call for price” makes a supplier invisible to agents. Merchant listings require a price. Proven 4 1 “Call for price” on 2 of 3 pages; the third shows a price behind a login prompt.
7.2 Truthful, machine-readable availabilityOwner’s callAvailability is a ranking input and a required feed field. Proven 4 0 “Check availability” needs a login and JavaScript.
7.3 Lead time as dataAgents compare suppliers on “when can I have it”. Proven + emergingProven in feeds · emerging in schema 2 0 “Usually ships fast.”
7.4 MOQ, pack quantity, quantity breaksB2B buyers act on totals. Without breaks an agent’s quote is wrong. EmergingVocabulary proven · use emerging 2 0 No quantity breaks or minimums stated.
7.5 Shipping and returns as dataAgents check terms before recommending. Proven 2 1 Return and shipping policy pages exist, not as data.
08 Compatibility & application data 0 / 12 2
Compatibility & application data: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
8.1 Crawlable cross-referenceMost switching starts from “what’s the equivalent of my current tool”. EmergingEmerging (strong) 4 0 No cross-reference.
8.2 Compatibility encodedAgents assemble complete orders (saw + arbor + spacer). Proven + emergingVocabulary proven · use emerging 2 0 No holder or collet links.
8.3 Suitability, including what not to use it forIt stops agents recommending the wrong tool. Emerging 2 0 Nothing says which coating suits which material.
8.4 Speeds and feeds as data“What speed do I run it at” is the most-asked machining question. EmergingEmerging (strong) 4 0 No speeds and feeds.
09 Engineering assets 1.5 / 6 1.5
Engineering assets: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
9.1 CAD per SKUEngineers and their agents spec what they can drop into the model. EmergingEmerging (strong) 3 1.5 CAD for one supplier line, behind a login.
9.2 Tool-data standards and CAMThe CAM ecosystem is where tooling decisions are now made. Proven 3 0 No ISO 13399 or CAM libraries.
10 Agent & channel interfaces 1 / 8 6
Agent & channel interfaces: each test, its evidence and Meridian's state
TestEvidencePtsNowWhat the catalog shows
10.1 Bulk machine-readable catalogAny agent, PIM or distributor can ingest the whole line correctly. Emerging 2 0 No bulk catalog file.
10.2 Shopping feedsFeeds are the documented intake for AI shopping surfaces. Proven 2 0 No shopping feed.
10.3 Query API / MCPLets procurement agents ask instead of scrape. A hedge on 2027. Emerging 2 0 No query API.
10.4 Distributor & e-procurement readinessMost B2B agents transact through a catalog someone else hosts, and most distributor attributes arrive in a supplier’s feed. Proven + emergingUNSPSC for Ariba proven · the rest emerging 2 1 Item kits go to two distributors by spreadsheet.
Total22 / 10081.5

What the ten fixes don’t do

81.5 is agent-ready, not perfect.

The last 18.5 points are real work, and we would rather show them than hide them.

  • 6pts
    Cross-reference and compatibility 8.1 · 8.2

    Mapping each competitor and legacy part number to yours, with the matching basis, is a project of its own. It is also the first question a switching buyer asks.

  • 4pts
    Speeds and feeds as data 8.4

    Starting parameters by material come from each supplier line. Worth it where the lines publish them.

  • 4.5pts
    CAD and CAM libraries 9.1 · 9.2

    STEP files and ISO 13399 packages come from the supplier lines. Most distributors link them rather than make them.

  • 1pts
    Phone speed 2.2

    A web-performance job, not a data job.

  • 3pts
    The emerging bets 10.3 · 1.4 · 2.4

    A query API, llms.txt and a full set of answer-shaped guides. Scored low on purpose until the platforms make them matter.

Where industrial catalogs are today

Most industrial tool sites we checked publish no product-and-price structured data at all.

That is the gap this demo is about. The engines document what they read: crawlable pages, product data with a price and an availability, feeds. A catalog that supplies it gives the agent something to cite. One that doesn’t leaves the agent to guess, or to quote a competitor.

Why now · Gartner: 45% of B2B buyers used generative AI in a recent purchase (Gartner press releases, 2026). 6sense: 94% of buyers rank a shortlist before contacting sellers.

How this works

A simulation of documented behaviour. Not a live AI call.

The AI buyer on this page is a deterministic script, not a model. It behaves the way the platforms document their fetchers: it honours robots.txt, it does not run JavaScript, it reads the HTML the server sends and the structured data in it, it cannot read a table out of a PDF reliably, and it cannot phone for a price. Run it twice and you get the same result, step for step.

Every failure names the rubric test it comes from (the chip beside each step). Every fix changes specific tests, and the score is the rubric’s own arithmetic, half points and all. The distributor, its supplier lines, part numbers, GTINs and prices are fictional; the GTINs use a prefix reserved for internal use, so they can never match a real product.

In a real audit we also run live probes: realistic buyer questions put to ChatGPT search, Perplexity and Google AI Mode, scored only from the citation URLs in each answer, never from what the model says about itself.

The basis: Bryan House, “Product Data Is Now a Supply Chain Problem”, Industrial Distribution, Sep 16 2026. AI buying agents read structured attributes and explicit pricing logic; specs in PDFs, exceptions in reps’ heads and “call for price” make a supplier invisible to them.

Check a page of yours · paste only

Paste one product page. Get its partial score.

Open one of your product pages, view its source (Ctrl+U, or ⌘+Option+U on a Mac), copy all of it and paste it here. This page can’t fetch a URL: the browser blocks it from sending or loading anything, so it reads only what you paste, as text, in this tab.

Your pasted page never leaves your computer

0bytes sent Requests since you pasted 0 Policy connect-src 'none'

Waiting for a page

A single page can show 26 of the 40 tests. The other 14 need your site, your feeds or your stock data, so they are listed with the reason rather than guessed.

Supplier feeds · fix 10 · the same rubric

Your feed to distributors, scored on the same rubric.

Distributors rarely write their own product data. It arrives as feeds: ETIM-classified item-setup spreadsheets from each line, and content from data pools and content services such as IDEA, DDS and Trade Service. So the rubric that scores a web page also scores a feed. Manufacturers: this is how your feed reads to a distributor’s item setup and to the AI agents downstream. Distributors: this is fix 10, every supplier line’s content graded, so a weak feed becomes a conversation with that line.

Supplier content scorecard · Meridian Sample Supply (fictional)

Each line scored on the 13 rubric tests a feed can show (34 of the 100 points). The rest of the rubric lives on the website.

Kestova Carbide End mills · 5 rows

In-house column names, specs left in the PDF, a placeholder, an inch value in a millimeter column, a bad check digit.

44%15 of 34 feed points

Kestova Carbide item-setup feed, illustrative layout
ItemMfrPart #UPCDescClassUOMPriceSpec sheetDia [mm]LOC [mm]FlutesFinish
KSV-250-3-UCKestova·400000250137KESTOVA 1/4 3FL EM UNCEnd millsEA31.75pdf/ksv-250.pdf6.3515.883Uncoated
KSV-375-3-UCKestovaKS3-375-U400000375236KESTOVA 3/8 3FL EM UNCEnd millsEA44.10pdf/ksv-375.pdf0.00025.43Uncoated
KSV-500-3-UCKestovaKS3-500-U400000500132KESTOVA 1/2 3FL EM UNCEnd millsEACallpdf/ksv-500.pdf0.5·3·
KSV-500-3-ATKestovaKS3-500-A400000500232KESTOVA 1/2 3FL EM ALTINEnd millsEA71.40pdf/ksv-500a.pdf····
KSV-625-3-UCKestovaKS3-625-U·KESTOVA 5/8 3FL EM UNCEnd mills·96.80pdf/ksv-625.pdf15.88·3Uncoated

Illustrative layout · class and feature codes are placeholders in ETIM’s format, not real ETIM codes · fictional rows

  • 5.3 Standard attribute vocabulary (fail) 0 / 3 0 of 4 attribute columns carry an ETIM feature code.
  • 6.2 Standard classification codes (fail) 0 / 2 0 of 5 rows with an ETIM class code, 0 with UNSPSC.
  • 7.3 Lead time as data (fail) 0 / 2 No lead-time column.
  • 7.4 MOQ, pack quantity, quantity breaks (fail) 0 / 2 No pack-quantity column: a distributor cannot price “per case”.
  • 9.1 CAD per SKU (fail) 0 / 3 No CAD column.
  • 3.2 Brand and MPN (partial) 1.5 / 2 4 of 5 rows carry both brand and manufacturer part number.
  • 3.3 GTIN (partial) 2 / 3 3 of 5 valid GTINs; 1 fails the check digit; 1 missing.
  • 5.1 Complete spec set per product family (partial) 2.5 / 4 60% of the required attributes filled.
  • 5.4 Specs not locked in PDFs (partial) 1.5 / 2 1 row with a spec-sheet link and almost no attributes.
  • 5.5 Data integrity (partial) 1 / 2 2 rows with a placeholder or a value that disagrees with its own description.
  • 7.1 Explicit pricing (partial) 3 / 4 4 of 5 rows carry a numeric list price.
  • 10.4 Distributor & e-procurement readiness (partial) 0.5 / 2 Item-kit essentials per row: a classification code, a valid GTIN, attributes, an image and pack data.
  • 5.2 Numeric, unit-explicit, dual units (pass) 3 / 3 9 of 9 measured values are plain numbers with a stated unit.

Row by row · what goes back to Kestova Carbide

  • KSV-500-3-UC 3.3 GTIN 400000500132 fails its check digit
  • KSV-500-3-AT 5.4 specs only in the spec-sheet PDF
  • KSV-375-3-UC 5.5 placeholder “0.000” in Dia
  • KSV-500-3-UC 5.5 diameter 0.5 sits in a mm column but reads as inches: the description says 0.5 in (12.70 mm). Flagged for a person, never auto-converted
  • KSV-500-3-UC 7.1 price “Call” is not a number

Score a feed of yours · CSV · stays in this tab

Drop your item-setup feed.

Save the sheet as CSV (comma, tab or semicolon) with a header row. Columns are matched by name (item number, brand, manufacturer part, GTIN, description, ETIM class, UNSPSC, order unit, pack quantity, price, lead time, image, spec sheet, CAD), and ETIM feature columns by their code, like EF009901 Cutting diameter [mm]. Not sure of the layout? Download a sample and compare.

or drop it here Your file never leaves your computer
0bytes sent Requests since you dropped a file 0 Policy connect-src 'none'

Waiting for a feed

Your file is read as text in this tab, scored on the same rubric tests, and dropped. Units are checked too: a diameter that reads as inches in a millimeter column is flagged for a person, never converted for you.

Part of the Data Watchdog desk

Your catalog, scored every month. Only the new decisions sent.

Data Watchdog runs this audit on your real site, feeds and item data: the automated pass on every product family, the live answer-engine probes, and a prioritized fix list where every fix shows its points and its effort. Anything that publishes a price, stock, a cross-reference or terms comes to you as a decision, never as an engineering default. Then it re-scores monthly, so a regression shows up as one new card, not a lost quarter.

Monthly desk

Data Watchdog

Starting at $1,950/mo

Re-scans your item, price and customer data monthly, includes a monthly AI-readiness score for your catalog, and sends only the new decisions.

Talk to Mike about an audit See Data Watchdog

More demos: all demos · Margin X-Ray · PO & Quote Intake