One screenshot per template, not per page
29,613
addresses in one sitemap
22
shapes of path
Before a launch you want to see every kind of page, not every page. This is a pre-launch audit: a sampler you paste your sitemap into, a sheet of six things to look for, and a script.
The sampler
The rule: take the first path segment and the number of segments. /blog/launch-notes becomes blog / depth 2. Pages that share both usually come from one template, so one picture stands for the group.
A group with one address is a page of its own, such as the home page or pricing, and is always taken. For a bigger group the sampler takes the middle address, or evenly spaced ones if you ask for more.
Nothing leaves your browser. The box starts with an invented list of 21 addresses. For a sitemap index, paste each file in turn.
21 addresses, 8 groups, 16 pictures
Every page at the same widths would be 42 pictures. 5 groups hold 80 percent of your addresses. It fits in the 100 screenshots a month of the Free plan.
| Group | Addresses | Samples to capture |
|---|---|---|
| products / depth 2 | 6 | https://example.com/products/oak-desk |
| blog / depth 2 | 5 | https://example.com/blog/launch-notes |
| docs / depth 3 | 3 | https://example.com/docs/api/orders |
| docs / depth 2 | 2 | https://example.com/docs/intro |
| legal / depth 2 | 2 | https://example.com/legal/privacy |
| (home) / depth 0 | 1 | https://example.com/ |
| pricing / depth 1 | 1 | https://example.com/pricing |
| about / depth 1 | 1 | https://example.com/about |
Five real sitemaps
The same grouping, run once on 8 November 2026 on the public sitemaps of five sites, one of them this one.[1] The count of groups stays small while the count of addresses grows 279 times.
| Site | Addresses | Groups | Groups with one address | Groups that hold 80% | Largest group |
|---|---|---|---|---|---|
| curlshot.com | 106 | 20 | 15 | 4 | blog / depth 2 (28) |
| nodejs.org | 1,670 | 6 | 1 | 2 | en / depth 4 (1,072) |
| kubernetes.io (English) | 1,889 | 23 | 13 | 4 | blog / depth 5 (570) |
| laravel.com | 944 | 30 | 13 | 5 | blog / depth 2 (443) |
| postgresql.org | 29,613 | 22 | 9 | 1 | docs / depth 3 (28,312) |
Few groups hold most pages
From 1 to 5 groups hold 80 percent of the addresses of each site. The rest is a long tail of one-off pages.
The tail is where the surprises are
51 of the 101 groups are a single address. On curlshot.com they are the home page, pricing, about, contact and the legal pages. Each is its own template, so each needs its own picture.
It fits in the Free plan
The largest count of groups is 30. At two widths that is 60 pictures; the Free plan of CurlShot has 100 a month. See the plans.
Where the rule fails
The rule looks at the shape of an address, not at the page. Two things in the table show it. The biggest group of postgresql.org is docs / depth 3, with 28,312 addresses, and it holds the manual of every version at once. One sample cannot show that version 14 and version 19 differ.
On nodejs.org the largest group, en / depth 4, mixes release notes and blog listing pages, which sit under the same key but are two kinds of page. Pick two or three samples for a big group, and widen the rule yourself where you know better: add the second segment for a site that keeps all its sections under one folder.
For a check that must cover every page, such as a content migration, use the list route instead: read the sitemap and send it in bulk. That is what the article on bulk screenshots does.
The audit sheet
Open the pictures group by group, the 390 px one first. For each picture, go down these six lines. Most problems show up on the template, so one fix clears the whole group.
Sideways scroll at 390
- Look for
- In the 390 px picture, content runs past the right edge, or the page is wider than the picture and text is cut.
- Why it shows up
- One wide element, often a table, a code block or a fixed-width image, is enough to push a whole template.
- What to do
- Find the element with a width in pixels and give it max-width: 100%, or let its container scroll.
Empty or half-drawn blocks
- Look for
- Grey placeholders, spinners or white gaps where content should be.
- Why it shows up
- The picture was taken before the data arrived. It can be the template, or only the capture.
- What to do
- Retake with wait_until=networkidle or wait_for_selector on the block. If the gap stays, it is a real bug.
Banner over the content
- Look for
- A cookie notice, chat bubble or promo bar covers text or a button.
- Why it shows up
- Overlays are added site-wide, so they show on every template at once.
- What to do
- Ask for the capture without the cookie banner for review pictures, and check the real page at 390 for the rest.
An error inside a 200
- Look for
- A picture that says "not found" or shows an empty list, for an address the sitemap lists.
- Why it shows up
- A page can answer 200 and still show an error, so a status check alone passes it.
- What to do
- Open that address by hand. Remove it from the sitemap or fix the page.
Fallback font or missing image
- Look for
- Headings in a plain system font, or an image shown as an empty frame.
- Why it shows up
- A font or image address that works on the developer machine can be blocked or wrong for the public.
- What to do
- Open the address of the font or image in a private window and fix the path or the headers.
Sticky bar in the middle of a full-page picture
- Look for
- A header or footer bar repeated or sitting across the middle of a tall picture.
- Why it shows up
- Fixed elements are drawn once at the position the capture happened to use.
- What to do
- Treat it as a capture artefact first. Compare with the first-screen picture before you change the page.
A status check is not enough
The script
It reads a sitemap, follows an index, groups the addresses with the rule above and takes the middle address of each group at 390 and 1280 px. With --dry it only prints the plan. It waits 7 seconds between calls, which keeps it under the 10 requests a minute of the Free plan. Run it on your server or laptop, not in a web page, because the key must stay private.
// node audit.mjs https://example.com/sitemap.xml --dry (list the plan, send nothing)
// node audit.mjs https://example.com/sitemap.xml (take the pictures)
import { mkdir, writeFile } from 'node:fs/promises'
const [sitemap, flag] = process.argv.slice(2)
const WIDTHS = [390, 1280]
const PAUSE_MS = 7000 // 10 requests a minute on the Free plan
const decode = (s) => s.replace(/&/g, '&').replace(/</g, '<').replace(/>/g, '>')
async function locs(url) {
const response = await fetch(url)
if (!response.ok) throw new Error(`${url}: HTTP ${response.status}`)
const xml = await response.text()
const found = [...xml.matchAll(/<loc>\s*([^<\s]+)\s*<\/loc>/g)].map((m) => decode(m[1]))
if (!/<sitemapindex[\s>]/.test(xml)) return found
const all = [] // an index lists other sitemaps: read each one in turn
for (const child of found) all.push(...(await locs(child)))
return all
}
// One group per first path segment and path depth: blog / depth 2, docs / depth 3, ...
const groups = new Map()
for (const address of new Set(await locs(sitemap))) {
const parts = new URL(address).pathname.split('/').filter(Boolean)
const key = `${parts[0] ?? 'home'}-depth-${parts.length}`
groups.set(key, [...(groups.get(key) ?? []), address])
}
const jobs = []
for (const [key, list] of groups) {
const sample = list[Math.floor((list.length - 1) / 2)] // the middle one, not the first
for (const width of WIDTHS) jobs.push({ key, sample, width, size: list.length })
}
console.log(`${[...groups.values()].flat().length} addresses, ${groups.size} groups, ${jobs.length} pictures`)
for (const job of jobs) console.log(`${job.key} (${job.size}) ${job.width}px ${job.sample}`)
if (flag === '--dry') process.exit(0)
await mkdir('audit', { recursive: true })
for (const job of jobs) {
const response = await fetch('https://curlshot.com/api/v1/screenshot', {
method: 'POST',
headers: { 'X-Access-Key': process.env.ACCESS_KEY, 'Content-Type': 'application/json' },
body: JSON.stringify({ url: job.sample, viewport_width: job.width, full_page: true, format: 'png' }),
})
if (response.ok) await writeFile(`audit/${job.key}-${job.width}.png`, Buffer.from(await response.arrayBuffer()))
else console.error(job.key, job.width, (await response.json()).error_message)
await new Promise((resolve) => setTimeout(resolve, PAUSE_MS))
}On 8 November 2026 the dry run on the sitemap of CurlShot printed 106 addresses, 20 groups, 40 pictures. The calls to the API follow the documentation and were not sent in this run. The bulk screenshots tool and the mobile screenshot tool run a first try without code, and getting started explains the key. Sizes for the second width are in the viewport cheat sheet.
How it was run
On 8 November 2026. Each sitemap was downloaded once, with one request per file. Sitemap indexes were followed one level down; for kubernetes.io only the English sitemap was read. Every address was grouped by its first path segment and the number of path segments, with the same function that runs in the sampler on this page.
Counts are of unique addresses in the files, with no check that each address answers. A group counted as a single address when only one address had that key. The “groups that hold 80%” column is the fewest groups, largest first, that together reach 80 percent of the addresses.
What the run did not cover: whether the groups are real templates, which needs a look at the pages; sitemaps are chosen by their owners and may leave pages out; and the API of CurlShot, whose calls are written from the documentation and were not sent. Plan sizes were read from the pricing page on 8 November 2026.
Questions people ask
How do I screenshot every page of a website?
You usually do not need every page. Read the sitemap, group the addresses by their first path segment and depth, and capture one or two of each group at a phone width and a desktop width. If you do need all of them, read the sitemap and send the list in bulk: see the article on bulk screenshots.
How many templates does a site have?
In five public sitemaps read on 8 November 2026, the count of groups by path shape ran from 6 to 30 for sites of 106 to 29,613 addresses. It is a rule of thumb, not a count of real templates.
Which widths should I check before launch?
At least a phone and a desktop: 390 and 1280 px are used here. Add a tablet width if tablets are a real share of your visitors.
Does a status code check replace looking at the pictures?
No. A page can answer 200 and still show an error, an empty list or a half-drawn block, and a status check passes it. A picture shows what a visitor sees.
How much does this cost with a screenshot API?
CurlShot counts one successful render as one screenshot, and the Free plan has 100 a month without a card (read from the pricing page on 2026-11-08). Cache hits and failed renders are not counted.
Sources
- [1]sitemaps.org, sitemaps.org - Protocol. Checked .
Read next
- PlaybookHow to screenshot a list of URLs in bulkScreenshot hundreds or thousands of URLs: a queue simulator that says how long it takes, a short Node.js script, and what fails at scale.
- Cheat sheetViewport sizes cheat sheet: phones to desktopsViewport sizes in CSS pixels, screenshot sizes and device pixel ratios of iPhone, Pixel, Galaxy, iPad and laptop screens, drawn to scale, and the sizes to test.
- PlaybookWebsite change monitoring with screenshotsHow to get told when a web page changes: the capture and compare loop, three ways to detect a change, a noise checklist, a cron and Node.js setup, and tools.