GTMRun

How scoring works

Last updated: 2026-11-16

A GTMRun check is deterministic: the same site in the same state gives the same result. Every row is one of three things, and nothing else:

RowMeaning
Passwe measured it and it is in place
Fixwe measured it and it is missing or broken; the guide says how to fix it
Not measuredwe could not tell; the row earns no points and is not counted as a failure

There is no AI assessment and no opinion score. Everything a founder is asked to fix was observed in DNS, in an HTTP response or in the rendered page.

The scored checks (32 points) #

GroupPointsGuides
Email authentication10SPF 3 · DKIM 3 · DMARC 2 · MX 2
Site basics8HTTPS 2 · Redirect 2 · HSTS 2 · Viewport 1 · Page weight 1
Crawl and content8robots.txt 2 · Sitemap 2 · Blog 2 · Structured data 1 · Title and description 1
Conversion6Pricing 2 · Call to action 2 · Analytics 1 · Email capture 1

The result says "23 of the measurable 25 points", not "23 of 32": points behind rows we could not measure are taken out of the denominator, so a site is never marked down for what we could not see.

Informational checks (0 points) #

Shown with a guide, never scored: soft 404s, indexing and snippets, JavaScript rendering, readable structure and media, and the rows below that other tools measure.

Rows measured by other tools #

Three rows come from portfolio tools and are labelled with their source on the result page:

RowToolWhat it adds
Canonical URL, one H1, image alt textserp page auditrenders the page in a browser, so it sees what JavaScript builds
Images inside the mobile budgetimage-optimizer auditevery image the homepage serves, its bytes and Slow-4G cost
Domain in the public link graphserp authoritythe Common Crawl domain graph; third-party, not a Google signal

When a tool does not answer within its time limit, its rows show not measured and the result page says which tool did not answer. The deterministic re-check (the button on a result page, or the gtm_recheck MCP tool) skips these tools on purpose: it is the fast, dependency-free way to confirm a fix.

Evidence tags #

Every guide says how strong the case is:

  • <Evidence kind="google_requirement" /> Google documents it as required, for Gmail delivery or for being crawled, indexed or shown in Search.
  • <Evidence kind="google_recommendation" /> Google documents it as good practice.
  • <Evidence kind="gtmrun_heuristic" /> Our go-to-market readiness heuristic. We score it because it affects whether visitors become customers, and we say plainly that it is not a Google ranking factor.

Pass, fix, not measured #

"Not measured" usually has one of four causes:

  1. JavaScript-rendered pages. If your HTML is an app shell, our raw-HTML checks cannot see content that JavaScript builds, but Google can. See JavaScript rendering. The serp rows above are measured on the rendered page precisely to cover this case.
  2. Catch-all hosts. If your site returns 200 for pages that do not exist, we cannot tell whether /pricing is real. See Soft 404s.
  3. Lookup failures. If a DNS lookup fails, the email rows show not measured instead of zero.
  4. A tool did not answer. serp or image-optimizer timed out or was rate-limited.

Why there is no AI score #

The legacy GTMRun scan added a 68-point "AI assessment". Two scans of the same site one minute apart scored it 8 and 16 with nothing changed. A number that moves on its own cannot tell a founder whether a fix worked, so the rebuilt check has none: measured or not measured only. What an agent is good at — drafting the fix, the post, the reply — is the product that sits behind "Get the fixes drafted", and every draft is approved by a person before it ships.

Re-checking after a fix #

Checks run every time; a fix shows up on the next run. DNS changes can take minutes to propagate. Some things only Google can report (indexing, clicks, AI-feature impressions): those live in Google Search Console, the measurement of record.

Sources #