Website score

Score a B2B SaaS website out of 100 against first-principles product marketing questions, then tell the user exactly what to fix first. Every score rests on words actually on the page, never on what the site probably says.

When to use it

• Before a website rewrite, to find the gaps worth fixing.
• To size up a prospect's or competitor's site quickly.
• After a relaunch, to check the new copy did its job.

Don't use it to write the new copy (run /website-copy after this), or for technical SEO and page speed. For head tags and social previews, use /metadata-lint.

Inputs

Required: the website URL. If it's missing, ask for it.

Helps if you have it:

• context/positioning/ and context/messaging/: lets you judge whether the site says what the company actually decided to say.
• context/icp/: lets you check the copy against the buyer's real language and pains.
• context/competitors/: competitor URLs for a side-by-side comparison.

Agency users: swap context/ for clients/<client>/context/. If a context file is missing, score from the site alone and say in the report which comparison you couldn't make.

Steps

1. Fetch the pages. Homepage first, then pricing, product or features, customers and about if they exist. If the main call to action leads to a demo or contact form, fetch that page too. Use web fetch, or Firecrawl or Exa if you have them. Only score pages you actually fetched. Produces: the page text, with the hero copied word for word.
2. Run the 5-second test. From the hero alone, can a stranger say what it is, who it's for and why they'd care? Record pass or fail, and note whether the hero names a tool the buyer already knows or a task they already do. Produces: a pass/fail line with the reason.
3. Score the nine categories with the rubric below. Each score gets one verbatim quote as evidence, or "not found on the page". Produces: a score table.
4. Total and interpret. Add the nine scores and apply the bands below. Produces: one number and one label.
5. Prioritise fixes in three tiers: fix this week (copy-only changes), this month (new sections or proof), this quarter (structural changes). At least one per tier. Produces: a ranked fix list.
6. Write before/after rewrites for the hero and the two weakest sections. Quote the "before" exactly; write real copy in the "after", not instructions. Produces: side-by-side blocks.
7. Compare to competitors if you have their URLs: one row each for hero claim, problem named, differentiator and main CTA. Produces: a short comparison table.

Scoring rubric (100 points)

Bands: 85 to 100 excellent, 70 to 84 good, 55 to 69 needs work, 40 to 54 weak, under 40 critical.

What good looks like

• Every score has a quote from a fetched page beside it.
• The fixes are specific enough to act on today: "replace 'all-in-one platform' with what it does".
• The after copy is written, not described.
• Numbers on the site are reported as they are. Anything not quantified on the page is marked "not quantified", never estimated.

Mistakes to avoid

• Scoring pages you didn't fetch, or assuming what the pricing page says.
• Rewarding clever copy that doesn't answer the five-second questions.
• Marking down a site for not being your style. Score against the questions, not taste.
• Long fix lists with no order. The reader needs to know what to do on Monday.
• Inventing metrics for the "after" copy. Use a placeholder like [X%] and say the number needs sourcing.

Output

Save to outputs/website/MMYY-website-score-<company>.md:

# Website score: <Company>
Score: <N>/100 (<band>)    5-second test: pass/fail, <reason>

| Category | Score | Evidence (verbatim) |

## Fix this week / this month / this quarter
## Before and after
## Competitor comparison (if run)

Next step: hand the fix list to /website-copy for the rewrite, or to /ab-testing if the fix is a bet you want to test first.