Skip to content
Meet Scar · A schema agent

Scar found 15 problems on our own About page.

Scar writes and checks the structured data behind our website: the code search engines read to learn who we are, what each page is, and how it all connects. It publishes only what a page supports, and a job isn't done until the live site serves it.

PublishedScar · run replay · 40×
04:58
dispatch to verified report
Prepare
Describe
Write
Verify
Report
policyNo approval required for this page
writeStored in the CMS and read back
verifyLive page, try 1: not served yet
verifyLive page, try 2: served, exact match
preparePage 2 of 2: this page
describePage described from its own content
checkChecks passed, write staged and signed
writeStored in the CMS and read back
verifyLive page, try 1: not served yet
verifyLive page, try 2: served, exact match
report2 of 2 verified. Published.
marshal.ing/ai-agent-use-cases/schema
Live
10
Pages with structured data Scar wrote and verified on our live site
Since July 28, including this one
Speed
~4min
From assignment to a verified live page
Two pages in one run: 4 min 58 s
Cost
$0.56
AI model cost for one page, capture to live check
Measured on one run in August
Hands
0
People touching the work between assignment and verified page
We read the report after
What Scar is made of

Three parts think. The other 21 make it safe to publish.

Scar is a publishing system with a model inside it. The model decides what a page means and how it connects to the rest of the site. Everything else is ordinary code, built so that decision is checked before it ships and proven after.

3/24parts that are the model
ModelCode, checks, tools
Instructions + authority
01Input
Assignment: pages, and deploy or audit
02Context
Instructions and our identity reference
03Access
Only granted tools resolve
04Control
Stop control and approval policy, checked before every write

Prepare

Code gathers the evidence
05Check
Address, redirects, and duplicate work
06Tool
Capture the live page
07Script
Extract the page's visible facts
08Script
Bind to exactly one CMS record, or skip

Describe

Model decides the meaning
09Model
Describe what the page is and what it connects to
10Script
Add our five identity records, every time
11Check
20+ structure and evidence rules
12Model
One correction turn, never a loop

Write

One governed write
13Script
Exact bytes and a fingerprint
14Tool
Stage: keep other SEO fields, recheck the revision, sign
15Decision
Publish now, or wait for approval
16Tool
Apply, locked to the revision it checked

Verify

Stored, then served
17Check
Read back what the CMS stored
18Read
Fetch the live page fresh, never from cache
19Check
Exactly one copy on the page, and it matches
20Decision
Verified, or partial and resumable
Execution + stored evidence
21Runtime
Sandbox with no network or credentials
22Store
Run folder: captures, graph, check results
23Model
Summary of judgment calls for the report
24Output
Typed report, one result per page
Audit mode reuses the evidence parts with no write tools granted.View the engineering schematic
One real run

What we hand Scar, and what comes back.

We gave Scar one address. It did everything between that and the report, and nobody touched the work in between.

01We write this
02Scar writes this
03You judge this

marshal.ing/customers

Mode
Deploy. Write the page's structured data, then prove it's live.
Before
A 384-character placeholder from our site code.
Rules
Describe only what the page shows. No invented ratings, dates, or claims.
Approval
Not required for this page. One policy setting turns it on.
PublishedVerified live
Deploy run · one page · July 28, 2026
Description authored43 parts
Checks20 of 20, twice
Stored copy read backIdentical
Live page200, one block
Page-only markers3 of 3 found
Impossible stringNot found
Old placeholderGone
Report delivered21 min 27 s
Flagged for you: first write refused. Scar shortened the change preview, and the gate requires it word for word. It staged the change again in full, then applied it.

What the page now says, and what Scar refused to say.

Added
  • Our five identity records
  • A list of 7 case studies, each tied to its company
  • 8 customer companies
  • 9 people, with their titles
  • 9 testimonials
Refused
  • Star ratings. The page shows none.
  • A breadcrumb. The page doesn't display one.
  • A company name for the anonymous case study.
  • Anything for eight logos with no story behind them.
  • FAQ markup. Off by default in our rules.
See the live page
Why Scar is over-engineered

Every check is a scar.

A working agent is mostly a record of what went wrong and what we changed. Four entries from Scar's casebook, all found on our own website.

Verification

The cache that said nothing changed.

What brokeAfter Scar updated a page, the tool it used to look at that page served a copy saved before the update. It happened on two runs in a row.
The costTrusted, that copy would have turned a working update into a reported failure.
Scar nowFetches the live page fresh and compares it with the exact data it signed. It also looks for a string that can't exist. If that ever matches, the check is broken, and Scar says so.
Audit

The audit that almost came back clean.

What brokeA page capture was saved in a wrapper our extractors didn't expect. Instead of refusing, they returned nothing: no structured data, no findings.
The costA clean bill of health for a pricing page that was serving a malformed, contradictory description. Scar caught it only because it could see the raw page.
Scar nowUnwraps every capture in code and stops with an error on anything it can't read. An empty result has to be earned.
Audit

The founder who became two people.

What brokeScar's first audit of our About page found our founder described under an ID left over from our old domain. That ID redirected to our homepage.
The costTo a machine, Kurt was two different people, and one of them pointed at a homepage. Our consistency checks missed it because they only compared records they recognized.
What we didThe audit changed nothing. It sent the fix to our site code, where the problem lived, and we cleared the old IDs site-wide. That audit found 15 problems on one page.
Assembly

The page that forgot who we are.

What brokeScar published a description containing the page and nothing else: no company, no founder, no website. Our code added those only when a page referred to them, and this page didn't.
The costSearch engines were told what the page was, but not whose it was.
Scar nowEvery page gets our five identity records, every time. The final check fails if one is missing.

We found and fixed each of these ourselves. The business an agent works for should never have to.

Look under the hood

This is what Scar wrote for this page.

Scar deployed this page's structured data on September 23, 2026, and confirmed the live site was serving it. The viewer reads it straight from this page's code.

10 nodes · 11 links · read live from this page
Who we are
This page
What it’s about
[
  {
    "@context": "https://schema.org",
    "@graph": [
      {
        "@id": "https://machinelike.ai/ai-agent-use-cases/schema#webpage",
        "@type": "WebPage",
        "url": "https://machinelike.ai/ai-agent-use-cases/schema",
        "name": "AI Schema Agent: How Scar Builds Structured Data | Marshal",
        "description": "Meet Scar, the AI agent that builds, audits, and verifies JSON-LD structured data on our website. See how Marshal governs its work and checks live results.",
        "isPartOf": {
          "@id": "https://machinelike.ai/#website"
        },
        "publisher": {
          "@id": "https://machinelike.ai/#org"
        },
        "inLanguage": "en",
        "mainEntity": {
          "@id": "https://machinelike.ai/ai-agent-use-cases/schema#scar"
        },
        "about": [
          {
            "@id": "https://machinelike.ai/ai-agent-use-cases/schema#scar"
          },
          {
            "@id": "https://machinelike.ai/ai-agent-use-cases/schema#thing-structured-data"
          },
          {
            "@id": "https://machinelike.ai/ai-agent-use-cases/schema#thing-schema-verification"
          }
        ],
        "mentions": [
          {
            "@id": "https://machinelike.ai/ai-agent-use-cases/schema#thing-read-only-audit"
          }
        ]
      },
      {
        "@id": "https://machinelike.ai/ai-agent-use-cases/schema#scar",
        "@type": "Thing",
        "name": "Scar",
        "description": "Marshal's AI schema agent. Given a page assignment, it reads the published content and any structured data already present, locates the matching record in the website's content system, prepares a preview of the description it intends to update, applies the change under the permissions in place, and then checks both the stored data and the public page before calling the update finished."
      },
      {
        "@id": "https://machinelike.ai/ai-agent-use-cases/schema#thing-structured-data",
        "@type": "Thing",
        "name": "Structured data",
        "description": "An explicit, machine-readable description of what a page contains, written in a format such as JSON-LD and placed in the page's code so search engines can see the connections a visitor reads: this is a business, this person is its founder, this article belongs to this publisher."
      },
      {
        "@id": "https://machinelike.ai/ai-agent-use-cases/schema#thing-schema-verification",
        "@type": "Thing",
        "name": "Schema verification",
        "description": "A two-step check after an update is applied. The agent first reads back the stored data to confirm the intended change survived intact, then visits the public page for evidence that the new data is being served. When an update is saved but not yet served, the report states that difference rather than treating the write as the finished job."
      },
      {
        "@id": "https://machinelike.ai/ai-agent-use-cases/schema#thing-read-only-audit",
        "@type": "Thing",
        "name": "Read-only audit",
        "description": "A separate inspection workflow that makes no edits. It compares what the live website serves with what the content system stores and what the shared business reference says should be consistent, then reports mismatches, outdated information and structural problems; correcting a finding needs a separate update task with its own authorization."
      },
      {
        "@id": "https://machinelike.ai/#org",
        "@type": "Organization",
        "name": "Machine-like",
        "legalName": "Growth Marshal, LLC",
        "url": "https://machinelike.ai/",
        "description": "Machine-like is the Managed Agent Operations company that designs, deploys, and operates AI agents as a service for small businesses.",
        "foundingDate": "2024-08-21",
        "founder": {
          "@id": "https://machinelike.ai/#founder"
        },
        "logo": {
          "@id": "https://machinelike.ai/#logo"
        },
        "sameAs": [
          "https://www.linkedin.com/company/marshal-agent",
          "https://www.crunchbase.com/organization/marshal-agent",
          "https://opencorporates.com/companies/us_ny/7402713",
          "https://marshal.ing"
        ],
        "address": {
          "@type": "PostalAddress",
          "addressCountry": "US",
          "addressRegion": "NY"
        },
        "areaServed": "Worldwide",
        "naics": "541512",
        "leiCode": "254900O2PF4PDTG4J395",
        "taxID": "99-4584195",
        "duns": "133330638",
        "identifier": {
          "@type": "PropertyValue",
          "propertyID": "NY Department of State DOS ID",
          "value": "7402713",
          "url": "https://data.ny.gov/resource/n9v6-gdp6.json?dos_id=7402713"
        },
        "alternateName": [
          "machinelike",
          "Marshal"
        ]
      },
      {
        "@id": "https://machinelike.ai/#founder",
        "@type": "Person",
        "name": "Kurt Fischman",
        "jobTitle": "CEO & Founder",
        "url": "https://machinelike.ai/team/kurt-fischman",
        "image": {
          "@id": "https://machinelike.ai/#founder-image"
        },
        "worksFor": {
          "@id": "https://machinelike.ai/#org"
        },
        "sameAs": "https://www.linkedin.com/in/kurtfischman"
      },
      {
        "@id": "https://machinelike.ai/#logo",
        "@type": "ImageObject",
        "url": "https://machinelike.ai/brand/marshal.png",
        "contentUrl": "https://machinelike.ai/brand/marshal.png",
        "caption": "Marshal logo"
      },
      {
        "@id": "https://machinelike.ai/#founder-image",
        "@type": "ImageObject",
        "url": "https://cdn.sanity.io/images/yslppoaf/production/e66fa03f826dc004f750c3e95731f2d1290da79a-7952x5304.jpg?rect=2502,334,2785,2785&w=560&h=560&q=85&fit=crop&auto=format",
        "contentUrl": "https://cdn.sanity.io/images/yslppoaf/production/e66fa03f826dc004f750c3e95731f2d1290da79a-7952x5304.jpg?rect=2502,334,2785,2785&w=560&h=560&q=85&fit=crop&auto=format",
        "caption": "Kurt Fischman"
      },
      {
        "@id": "https://machinelike.ai/#website",
        "@type": "WebSite",
        "url": "https://machinelike.ai/",
        "name": "Machine-like",
        "publisher": {
          "@id": "https://machinelike.ai/#org"
        },
        "inLanguage": "en"
      }
    ]
  }
]
When this page changes, Scar rewrites the description, and this viewer follows.Check it yourself
Why not a plugin

Plugins write it. Scar maintains it.

SEO plugins and schema generators produce structured data from templates and settings. That's the easy half. The hard half is keeping it true as your site changes.

The jobTypical plugin or generatorScar
Describes your business the same way on every pagePluginIf every template and setting agreesScarChecked on every write
Leaves out what the page can't supportPluginFills whatever the template asks forScarRefuses ratings, dates, and claims without visible evidence
Confirms the live site serves itPluginNot its jobScarA fresh check of the live page after every write
Catches drift after a redesign or domain movePluginWhen someone happens to lookScarRead-only audits, on a schedule or on request
Fixes it when it breaksPluginYouScarWe do
If an agent ran your version

Write it. Then prove it landed.

Scar's job has a shape most small businesses already have: information that has to appear in more than one place and stay correct after it does. We build the agent for your stack, run it, and fix it when something changes.

Business details

Name, address, hours, and phone kept identical across your site, your Google Business Profile, and directories.

Listings and prices

A new listing or price change pushed to your site and your marketplace, then confirmed live on both.

People pages

Attorney or agent bios kept in sync between your site and bar or brokerage directories.

Product feeds

Feed updates confirmed as approved in Google Merchant Center, with rejected items flagged for you.

01
Prepare
02
Describe
03
Check
04
Write
05
Verify
06
Report
Where you come in: after the report. You see what changed and tell us what to adjust.

It's a setting, not a rebuild. We ran Scar both ways on our own site: with approval on our Security page and without it on our Customers page. Same agent, same tools. One policy setting changed.

Free · Read-only

Get a free schema audit of your site.

Send us your site. We'll audit your key pages the way Scar audits ours: read-only, page by page. Nothing on your site changes.

Get a free schema audit

No credit card. We change nothing on your site.

Your auditRead-only
What comes back, page by page
What search engines see on each key page
Where your business is described differently across pages
Claims your pages make that nothing on them supports
What you think is published that the live page doesn't serve
What to fix first, and where each fix lives
Before you hand over your structured data

Questions owners ask

Will structured data get me into ChatGPT or Google's AI answers?

No one can promise that. Structured data makes your site easier for machines to read correctly: who you are, what each page is, and how it all connects. Whether a search engine or an AI tool shows you is its decision. Scar's job is narrower: an accurate description, applied with permission, and confirmed live.

My SEO plugin already adds schema. What does this add?

A plugin writes markup from templates and settings. Scar owns it afterward: one consistent description of your business on every page, nothing the page can't support, and a check that the live site actually serves it. When a redesign or a domain move breaks something, finding and fixing it is our job, not yours.

Does it work with my website?

Scar is built for our CMS. Yours would be built for your stack, such as WordPress, Webflow, Shopify, or HubSpot, through access you authorize. The agent never sees your password.

Why won't Scar add FAQ markup?

Google stopped showing FAQ rich results on May 7, 2026. Markup that earns nothing and repeats what the page already says is noise, so Scar leaves it out unless we deliberately decide otherwise. These questions don't carry it either.

What does it cost?

Plans start at $249 a month per workload, with no setup or integration fees, and you can start with a free trial. The AI is a small part of that: one page cost Scar about $0.56 in model usage. You're paying for the build and the ongoing operation around it.

Let's get back tothe work of humans⁠1

Machine-like is the services layer that bridges frontier AI and business operations.