ATOM
Login Start Trial

Last verified 26 August 2026
Reviewed quarterly

What we found auditing our own portfolio

Five properties. Five failures. Every dashboard green.

Every site we own turned out to be invisible to AI agents — for a different reason each time. We only found it because we fetched our own pages as a crawler and counted the words.

What you want is simple: when someone asks the question your business answers, you are the source the machine cites. Everything below is what stands between you and that.

Run the same check on your site.

Five readers, one URL, thirty seconds. Free, and you keep the findings and the fixes.

or see an example result

Nothing measured yet. Enter your address to run a live check, or see an example result.

The week that produced this page

We were the ones who should have caught it.

Five properties. Every one built deliberately, on a modern stack, by people who knew what they were doing. Two of them are the front door to a capital business. One is a clinical site where being wrong has consequences.

Over five days, every single one turned out to be unreadable by AI agents. Not one shared a cause with another. And every one of them returned a healthy status, rendered perfectly in a browser, and reported green on every dashboard we owned.

Property 1

No render layer had ever been wired. Served nine words to every crawler since launch.

Since launch
Property 2

Blocked all seven major AI crawlers in its own robots.txt while serving them 5,796 perfect words.

Months
Property 3

A "do not index" instruction correct in source and never delivered to 646 pages.

Months
Property 4

Sixteen AI agents receiving a blank shell because a user-agent list was out of date.

Unknown
Property 5

A credential rotation silently returned every crawler an empty page for a full day.

Found by accident

If it took us five days and a dozen corrections to our own instruments to find this on sites we built — the question isn't whether it's happening to yours.

Why one fix doesn't finish it

There are four ways to be invisible. Only one depends on how your site was built.

Most conversations collapse these into a single problem. They're independent, they fail independently, and fixing one does nothing for the other three.

01

Delivery

Can AI agents read the page at all, or do they receive an empty document?

Stack-dependent

02

Depth

Is there anything on the site substantial enough to be worth citing?

Stack-independent

03

Freshness

Is what you published still true, and would you know if it stopped being true?

Stack-independent

04

Wiring

Does any of it connect to a customer, or does the trail end at a pageview?

Stack-independent

On a traditional CMS you pass the first one by accident — and fail the other three exactly like everyone else.

Server-rendered by default is a real advantage, and it solves the cheapest problem on the list. A brochure renders perfectly and gets cited never.

Free, and complete

Layer 0 in full. Every check, every test, and the fix.

This is the layer the checker above measures, and we're publishing all of it — the tests and the remediation. You can close this page having fixed something real. That's deliberate.

Layer 0

Delivery

Whether an agent can read the page. Everything above this is irrelevant if it fails, and it fails silently.

Free · Fixable this week
Common practiceWhat's requiredHow you'd know
"It loads in my browser"Rendered HTML delivered to every agent, including ones that don't exist yetFetch your homepage as GPTBot and as a random unrecognized agent. Count the words. Under 50 means invisible.
Assume Googlebot behaves like every crawlerGooglebot executes JavaScript. Most AI crawlers do not.Compare word counts across Googlebot, ClaudeBot, and a nonsense agent string.
A render service is "installed"Render service verified running, with a fail-open path when it errorsCheck for the render header on a live crawler fetch. Absent means it never ran.
robots.txt written once, never rereadrobots.txt read line by line — and understoodDoes any AI agent have Disallow: /? Does every named group carry the full disallow set?
One block of rules for everyoneAwareness of the specificity trap: a named agent reads only its own groupIf GPTBot has its own group with no Disallow lines, it can reach everything — including /admin.
—Status-code capability: can the site emit 404, 410, 301?Request a path that cannot exist. A 200 means you can't signal "gone" to anyone.
Head tags set by JavaScriptTitle, description and directives in the initial HTMLView source, not the inspector. If your title only appears after JS runs, no agent sees it.
"We deployed it"Deploy confirmed serving. Built is not served.Publish a unique phrase. Fetch production. Look for it.
An uptime monitorDelivery monitored continuously, not checked onceAn uptime check returns a healthy 200 for a page serving nine words.
Why we give this one awayMost of these fixes take an afternoon. Finding them took five days and twelve corrections to our own instruments. Detection is the hard part — not the repair. If this page saves you a month of invisibility, it has done its job whether or not you ever speak to us.

The other eight

Delivery is the loudest failure. It isn't the expensive one.

Layer 0 can be measured from outside, which is why it's free. The remaining eight require reading your corpus, your structured data, your analytics and your publishing history. Here's what each one asks — and the question that tells you where you stand.

Layer 1

Depth

Whether there's anything substantial enough to be worth citing.

What on your site could a competitor not reproduce in an afternoon?

Months
Layer 2

Entity

Whether machines know who you are, consistently, everywhere.

Search your brand. Is the database result about you correct?

Days
Layer 3

Structure

Whether your content is machine-liftable and your instructions arrive.

Can you link to one specific term in your glossary?

Days to weeks
Layer 4

Freshness

Whether it's still true — the fastest-growing liability, and almost nobody measures it.

How many of your articles contain a statistic older than 24 months?

Ongoing
Layer 5

Distribution

Whether you appear where the answer is formed, not where traffic used to start.

How many independent sources reference you by name?

Ongoing
Layer 6

Capture

Whether interest becomes an identity — from a visitor who arrived already informed.

AI referrals land deep, not on your homepage. Does your deepest page convert?

Days
Layer 7

Attribution

Eight links from question to revenue. Most businesses have none instrumented.

Which specific question produced your most recent customer?

Weeks
Layer 8

Trust

Whether the machine has reason to believe you specifically. Decisive in health, finance and law.

What on your site proves you're qualified rather than asserting it?

Months

Fixing Layer 0 while Layer 4 rots gains you nothing. They compound in one direction only.

A perfectly delivered page of two-year-old claims is worse than no page — an engine that cites you and turns out to be wrong stops citing you. And every one of these breaks again. A single credential rotation took two of our properties offline to crawlers in a day.

What waiting costs

Invisibility isn't a pause. It compounds.

Losing a month of visibility doesn't cost you a month. Crawlers reduce frequency against hosts that return errors or empty documents, and they don't resume at the old rate the moment you fix it.

On one of our properties a render failure took AI crawl volume from roughly twelve thousand requests a day to near zero within twenty-four hours. It stayed there two weeks while the site returned a healthy 200 to every human who looked.

The offsetting fact, and it's the more useful one: almost nobody is ahead of you. These failures are the norm. There is no established leader in most categories — only a field nobody has measured.

The recovery curve

Day 0Delivery breaks. Nothing errors. Everything reports healthy.
Days 1–7Crawlers receive empty or failed responses and reduce frequency.
Weeks 2–8Citations go stale. New content is never seen. Competitors fill the answer.
FixThe fault is corrected in an afternoon, once someone finally looks.
AfterCrawl frequency rebuilds gradually. Citation position does not return automatically.

What we do about it

Nine systems. You run one of them.

The reason most businesses lose this isn't capability. It's that nine continuous systems is not a job anyone has time to hold, and the failures are invisible without instruments built to find them.

For reference: a technical SEO audit runs $8,000–15,000 and covers roughly Layer 0 and Layer 3, once. It tells you what was true the week it was written.

Step one · free

The Delivery Check

What you just ran. Five readers, your real word counts, robots findings.

  • Every Layer 0 fault named
  • Copy-paste fixes included
  • Yours to keep, no call required

Free, permanently

Step two

The Nine-Layer Audit

Everything the free check can't see from outside. Your corpus, schema, entity graph, freshness, attribution.

  • Fifty-one checks, per layer
  • Findings ranked by cost, not severity
  • The order to fix them in
  • Delivered inside a week

Priced to qualify, not to profit

Step three

GrowthOS

The nine systems, run continuously, with production and measurement attached.

  • Daily delivery and directive monitoring
  • Content produced against your question space
  • Freshness and decay managed
  • Query-to-member attribution

By fit

What we'll say plainly

If the audit finds nothing material, we'll tell you and refund it. That has happened once. We'd rather tell you your house is in order than sell you a system you don't need — and if it isn't, you'll have the findings whether or not you go further.

Start with the free check

What this actually means

The site was the deliverable in 2015.

An agency built it, it looked good, it ranked, and the path from question to customer ran through a results page you could open and see.

Now the site is one link in a chain of nine, and the links on either side of it are the ones that decide outcomes. Owning one link means operating blind about the other eight.

The question was never whether you have a good website. It's whether the systems now answering your buyers' questions can read you, trust you, and cite you — and whether you'd know if they stopped.