Website security audit: what your live site shows anyone who looks
An external audit, reviewed by a person, of twelve things anyone on the internet can examine without logging in: your hostnames, certificates, headers, cookies, leftover files, admin screens and the code your site sends to every browser.
Each result comes back with evidence, business impact and a fix, and the report states what an outside view cannot reach.
On this page
What a website security audit looks at
A website security audit examines a live site from the position of a stranger on the internet.
There is no password, no source code and no admin access, only what the site, its DNS records and its hosting return to an ordinary request. That turns out to be a lot. Old hostnames, forgotten backup files, keys bundled into JavaScript and staging copies of the site can all be reached from outside, and none of them needs an account to find.
Clever Ops sells this as the website security check. A person confirms every finding and keeps the evidence behind it, and each check that found nothing is recorded too, with the hostnames and date it covers. Other providers call similar work a website security assessment or a website security review.
Whatever the label, ask two things before you buy: does the work stay outside the login, and does a person confirm each result before it reaches you?
Website vulnerability testing from outside has a firm limit, and a careful audit keeps to it: it shows that a weakness exists without using it. A backup file is identified from its first bytes rather than downloaded, a leaked key is recognised by its format rather than tried, and a subdomain that someone else could take over is reported rather than claimed.
That keeps the audit low-risk for a site with real visitors, and it is why some findings arrive rated on what was visible, with the deeper question handed to a penetration test.
The twelve checks and what a good result looks like
Each check names the OWASP Web Security Testing Guide v4.2 test it follows, so a developer can look up the method.
Where a check goes further than that test, the extra part is marked, with its own reference where one exists. The good-result lines double as a website security audit checklist you can work through yourself before booking anything. The worked example on this page shows what each check records.
The audit starts by finding everything that responds on your domain, because a forgotten hostname tends to be the least maintained one. It reads robots.txt and sitemaps, which anyone can fetch and which crawlers are free to ignore, and looks at what search engines have already indexed. It then lists hostnames from public DNS and certificate records and checks where each one points.
- Check 1, search results, robots.txt and sitemaps (WSTG-INFO-01, WSTG-INFO-03). Good result: nothing private appears in search results, and robots.txt names no path that serves private files, because it is a public list rather than a lock.
- Check 2, hostnames and dangling DNS (WSTG-CONF-10). Good result: every hostname has an owner who can say what it is for, and no record points at a third-party service you have stopped using.
These checks read the certificates and headers your server sends with every response. They are quick to fix and easy to get subtly wrong, so the audit records the exact value seen on each hostname rather than one pass or fail for the whole domain.
- Check 3, certificates and TLS (WSTG-CRYP-01). Good result: a valid certificate matching every hostname that serves pages, and only current protocol versions accepted.
- Check 4, HSTS (WSTG-CONF-07), plus the HTTP redirect as an extra. Good result: a Strict-Transport-Security header that tells browsers to keep using HTTPS for the number of seconds its max-age sets. The extra check: plain HTTP requests redirect straight to HTTPS.
- Check 5, framing (WSTG-CLNT-09), plus a Content Security Policy review against the OWASP Content Security Policy Cheat Sheet. Good result for framing: other sites cannot load your pages inside a frame, which the clickjacking test checks through the X-Frame-Options header. Good result for the policy: an enforced CSP listing the script sources the site actually uses, with a frame-ancestors directive, which the cheat sheet describes as replacing X-Frame-Options.
- Check 6, cookie attributes (WSTG-SESS-02). Good result: Secure on every cookie, HttpOnly on any cookie that identifies a session, and a SameSite value chosen deliberately rather than left to the browser.
- Check 7, cross-origin rules (WSTG-CLNT-07). Good result: endpoints either send no Access-Control-Allow-Origin header or name the origins allowed, and none echoes back whatever origin asks.
Most sites carry traces of how they were built and changed: version strings, files left behind by a migration, a staging copy nobody switched off. None of it needs a login to find. Versions are matched against the vendor's own advisories, and files are confirmed with a single careful request.
- Check 8, visible software and versions (WSTG-INFO-02, WSTG-INFO-08). Good result: no component visible from outside is older than its vendor's latest security release.
- Check 9, old, backup and unreferenced files (WSTG-CONF-04). Good result: no archives, database exports, logs or copies of configuration files at any path the web server will serve.
- Check 10, hidden administrator interfaces (WSTG-CONF-05), plus staging copies as an extra. Good result: no admin screens or functions turn up beyond the ones your team knows about and uses. The extra check looks for staging and test copies of the site; good result: they sit behind a password or cannot be reached from the internet at all.
Anything a browser receives can be read by the person using it: page source, JavaScript bundles, source maps and error pages. The last two checks read that material for things that were meant to stay on the server.
- Check 11, keys and secrets in page source and JavaScript (WSTG-INFO-05). Good result: the only keys in the browser are ones designed for browser use, each with the restrictions its provider offers switched on.
- Check 12, error pages and debug output (WSTG-ERRH-01). Good result: unexpected input produces a plain error page, with stack traces and debug detail kept in server logs.
How an observation becomes a finding
A raw observation is not yet a finding.
A missing header on a page that sets no cookies and runs no scripts is a hardening note, while the same gap on a login page is a real exposure. Before anything is reported, four questions are answered, and the answers go into the report.
- Is it real? The observation is reproduced and the evidence kept: the request, the response and the time, with secrets removed. Anything that cannot be reproduced is dropped rather than passed on as a possible issue.
- Where does it apply? Each finding names the hostnames and paths affected, because one setting often differs between the main site, a booking subdomain and a staging copy.
- What could someone do with it? The business impact is written for the owner: whose data or which action is exposed, and what an attacker would still need.
- How will we know it is fixed? Every fix comes with retest criteria, so your developer knows in advance exactly what the second look will check.
Two ratings, and why they can disagree
Each rated finding carries a CVSS v3.1 score from 0.0 to 10.0 and a business-risk rating. The CVSS score states the technical severity: its Base metrics describe how the weakness can be exploited and the direct impact if it is. The business rating says what that severity means for your business in particular, given the data and customers involved.
The two can point in different directions, so the report shows both. Where a rating depends on something an outside view cannot confirm, the finding says so, and the open question joins the list a penetration test would settle.
Why this differs from a scanner-only report
A scanner lists every pattern it matched, including items that do not apply to your site, and leaves the sorting to you. An audit report lists what was confirmed, what was checked and found clean, and what could not be checked from outside, each tied to the hostnames it covers.
A clean result is a negative result for that hostname on that date, not a statement that the area is sound everywhere.
What an outside view cannot see
An unauthenticated audit stops at the login screen, and that line decides how much it can tell you.
When customers have accounts, make payments, upload documents or view records that are theirs alone, the most serious risks usually sit behind that screen, where only an authenticated penetration test reaches. The last column of the worked example marks the line check by check. The lists below cover the areas the twelve checks never touch.
Covered from outside
Everything in the twelve checks: hostnames and DNS, certificates and transport security, headers and cookies on public responses, cross-origin settings on public endpoints, the software versions a site reveals, files and admin screens at linked or guessable paths, secrets in code sent to the browser, and errors that public inputs can trigger.
Needs an authenticated penetration test
These questions can only be answered with test accounts, usually two for each role, so behaviour between users can be compared.
- Whether one customer can read or change another customer's records, and whether staff roles stay within their limits.
- Business rules the server is meant to enforce, such as prices, discounts, booking limits and payment steps.
- API routes that answer only signed-in requests, including whether they share a weak setting that a public route revealed.
- Account recovery, session handling after sign-in, file uploads, and stored input that other users later see.
- Pages and scripts served only after sign-in, which an outside audit never receives.
Outside both
Some areas need a different kind of review or a different provider altogether, whichever test you choose.
- Source code, unless a code review is added. A flaw that never shows in a response stays hidden from both tests.
- Your hosting provider's platform, your payment provider's hosted checkout and other services run by third parties, which belong to their owners.
- Load, traffic and denial-of-service testing, which Clever Ops does not perform.
- Office networks, staff devices and email accounts.
Worked example: twelve checks on an illustrative site
The example site, site.lab.test, has an enquiry form, a booking subdomain and a CMS login. The site and every result in this table are an illustration written to show how each check becomes a finding; they are not a client's site or a client's results.
How to use itHide the instructions
Read along a row to follow one check from evidence to fix, use the first column as a checklist for your own site, and use the last column to see where an authenticated penetration test would go further.
12 rows
Check
What the example site showed
1. Search results, robots.txt and sitemaps (WSTG-INFO-01, WSTG-INFO-03)
- What the example site showed
- robots.txt asks crawlers to skip /print-quotes/. That path lists 40 PDF quotes, and a search engine has indexed two of them.
- Finding and rating
- High. Customer quotes are public in a listed directory.
- Business impact
- Names, addresses and quoted jobs are readable by anyone, and search engines already hold copies. robots.txt advertised the path instead of hiding it.
- Fix and retest
- Remove the files and the listing, keep quotes behind a login, request removal of the indexed copies, and read the OAIC's explanation of how the Notifiable Data Breaches scheme applies. Retest: the path returns 404.
- Audit versus authenticated test
- Same either way, because the files are public. A signed-in test would also check who can reach quotes inside the application.
2. Hostnames and dangling DNS (WSTG-CONF-10)
- What the example site showed
- Public DNS and certificate records list six hostnames. shop.site.lab.test still points at a hosted store service where the store has been deleted.
- Finding and rating
- High. Subdomain takeover candidate, reported and not claimed.
- Business impact
- If the provider does not verify ownership, anyone who opens a store under that name could publish pages on your hostname that visitors would trust as yours.
- Fix and retest
- Delete the DNS record or point it at a service you control, and add DNS clean-up to the steps for retiring any service. Retest: the hostname no longer resolves to the provider.
- Audit versus authenticated test
- The audit is usually the wider sweep here. A penetration test scoped to one application may never look at the shop hostname.
3. Certificates and TLS (WSTG-CRYP-01)
- What the example site showed
- www and booking hostnames: valid certificates matching each name, TLS 1.2 and 1.3 accepted, older versions refused.
- Finding and rating
- No finding. Recorded as a negative result for those two hostnames on the audit date.
- Business impact
- None found. The result covers only the hostnames and date tested, so a hostname added later is not covered by it.
- Fix and retest
- Nothing to fix. Keep certificate renewal automatic and include any new hostname in the next audit.
- Audit versus authenticated test
- Identical from outside and signed in, because the same certificates serve both.
4. HSTS (WSTG-CONF-07), plus the HTTP to HTTPS redirect as an extra check
- What the example site showed
- http:// redirects to https:// on every hostname. No Strict-Transport-Security header on any response.
- Finding and rating
- Low. HSTS not set.
- Business impact
- A visit that starts on http://, from an old link or a typed address, can be intercepted on a hostile network before the redirect. HSTS sends the browser straight to HTTPS on later visits.
- Fix and retest
- Send Strict-Transport-Security on every HTTPS response, starting with a short max-age and raising it once stable. Add includeSubDomains only after check 2 confirms every subdomain serves HTTPS. Retest: header present with the agreed values.
- Audit versus authenticated test
- Same from outside. An authenticated test also confirms the header on signed-in pages and API responses.
5. Framing (WSTG-CLNT-09), plus Content Security Policy (OWASP Content Security Policy Cheat Sheet)
- What the example site showed
- No Content-Security-Policy header and no X-Frame-Options. A test page on another origin loaded the CMS login inside a frame.
- Finding and rating
- Medium. Login and enquiry pages can be framed, and no CSP is set.
- Business impact
- A hostile page can hide your login inside its own and trick a visitor into clicking or typing through it. With no CSP, any script injected elsewhere on the site would run unchecked.
- Fix and retest
- Set frame-ancestors 'self' in a policy, with X-Frame-Options SAMEORIGIN alongside it until the policy is live. Run the CSP in report-only mode, then enforce it. Retest: framing from another origin is blocked.
- Audit versus authenticated test
- The audit sees headers on public pages. An authenticated test checks pages behind the login and looks for the injection flaws a CSP is there to contain.
6. Cookie attributes (WSTG-SESS-02)
- What the example site showed
- The login page sets a pre-login session cookie with Secure and HttpOnly but no SameSite. A chat widget cookie set on every page lacks Secure.
- Finding and rating
- Low. SameSite left to browser defaults, and one cookie without Secure.
- Business impact
- Protection against cross-site requests depends on which browser a visitor uses rather than on the site. The widget cookie can travel over plain HTTP if a page is ever loaded that way.
- Fix and retest
- Set SameSite=Lax on the session cookie, or Strict for admin-only cookies, and add Secure to every cookie. Retest: each Set-Cookie header carries the agreed attributes.
- Audit versus authenticated test
- The audit reads cookies issued before sign-in. An authenticated test follows the session through sign-in, use, sign-out and timeout.
7. Cross-origin rules (WSTG-CLNT-07)
- What the example site showed
- /api/availability on the booking hostname echoes any Origin it receives in Access-Control-Allow-Origin, with Access-Control-Allow-Credentials set to true.
- Finding and rating
- Medium, rated on what was visible. This public route returns only open time slots.
- Business impact
- Low on this route. If signed-in routes share the setting, another website could read a logged-in customer's bookings through that customer's own browser.
- Fix and retest
- Allow only named origins, and never echo the request's Origin alongside credentials. Retest: an unlisted Origin receives no Access-Control-Allow-Origin header.
- Audit versus authenticated test
- The audit can test public routes only. Whether customer routes share the setting is a question for an authenticated test.
8. Visible software and versions (WSTG-INFO-02, WSTG-INFO-08)
- What the example site showed
- A generator tag and a response header name the CMS and its version. The vendor's advisory list shows two security releases since that version.
- Finding and rating
- Medium. Outdated CMS core, matched to the vendor's advisories. No exploit attempted.
- Business impact
- The advisories publish what was fixed, which gives anyone a starting point. Extensions are probably behind as well, but that cannot be confirmed from outside.
- Fix and retest
- Update the core and every extension, turn on security updates where the platform allows, then remove version strings, since hiding a version does not patch it. Retest: reported version at or above the fixed release.
- Audit versus authenticated test
- The audit sees only what the site reveals, which can be incomplete. An admin-level review reads the exact installed versions.
9. Old, backup and unreferenced files (WSTG-CONF-04)
- What the example site showed
- /site-backup-2026-03.tar.gz returns 200. The first bytes confirm a compressed archive, and nothing more was downloaded.
- Finding and rating
- High. A full site backup can be downloaded by anyone.
- Business impact
- A backup can hold a copy of every file on the site, including configuration files with database and email passwords, and possibly a database export.
- Fix and retest
- Delete it from the web root now, store backups away from the web server, rotate every password and key it could contain, and check access logs for earlier downloads. Retest: the path and common backup names return 404.
- Audit versus authenticated test
- Both find files only at linked or guessable paths. Neither lists the server's folders, which your host or developer can do directly.
10. Hidden administrator interfaces (WSTG-CONF-05), plus staging copies as an extra check
- What the example site showed
- staging.site.lab.test serves a full copy of the site and its CMS login with no password in front, and carries no noindex tag.
- Finding and rating
- Medium. A staging copy and its admin login are open to the internet.
- Business impact
- Staging copies are easy to miss when updates are applied, and they may hold real customer data copied across. Anyone can try the login, and search engines may list the pages.
- Fix and retest
- Put staging behind a password or an IP allowlist, add noindex, and limit the production admin screen to staff using multi-factor sign-in. Retest: staging asks for credentials before any page loads.
- Audit versus authenticated test
- The audit finds login screens but does not sign in or guess passwords. An authenticated test checks what each role can reach once inside.
11. Keys and secrets in page source and JavaScript (WSTG-INFO-05)
- What the example site showed
- The enquiry form script holds a key in the format the email provider uses for secret server keys, beside a maps key. Neither key was used.
- Finding and rating
- High. A secret email-sending key is shipped to every visitor. The maps key needs its restriction confirmed.
- Business impact
- Anyone could send email through your account under your domain, for spam or phishing that looks like yours. An unrestricted maps key can be used by other sites at your expense.
- Fix and retest
- Revoke and replace the email key, move sending to the server, and review the provider's logs. Restrict the maps key to your domain. Retest: no secret key in any served file, and your developer confirms the old key is revoked.
- Audit versus authenticated test
- Both read public code. An authenticated test also reads scripts served after sign-in, and code access adds the repository and build settings.
12. Error pages and debug output (WSTG-ERRH-01)
- What the example site showed
- A malformed date in a booking URL returns a stack trace naming the framework, its version, server file paths and a database table.
- Finding and rating
- Low. Debug output is switched on in production.
- Business impact
- It hands an attacker the internal map they would otherwise have to guess, and suggests other debug features may be switched on too.
- Fix and retest
- Turn off debug mode in production, return a plain error page, and keep the details in server logs. Retest: the same request returns a generic error with no internals.
- Audit versus authenticated test
- The audit can only trigger errors through public URLs and forms. An authenticated test reaches the inputs behind the login and in the API.
Who an external audit suits
The audit fits sites where the public surface is most of the risk: brochure and marketing sites, lead-generation sites built around enquiry forms, and sites whose only logins belong to the people who edit them.
It also fits the moments when that surface has just changed and nobody has looked at it as a whole.
- Before a new site or rebuild goes live, while hostnames, redirects and headers are still easy to change.
- After a change of agency, developer or host, when nobody is certain which subdomains, test copies and DNS records the last team left behind.
- Before you answer a customer security questionnaire that asks about HTTPS, headers or exposed services.
- After a specialist has cleaned up a compromised site, to see what it exposes before it starts taking enquiries again.
When it is the wrong depth
If customers sign in, pay inside your site or upload files, start with a web application penetration test instead. WordPress and WooCommerce sites get more from the WordPress security audit, which adds the logged-in roles and a plugin inventory read from the admin side. Live Shopify storefronts are outside this service.
Security reviews of sites we built or host
The same checks then run as a security review rather than an independent audit, and the report names who built the site and who hosts it. When the person relying on the report wants independent evidence, choose a tester who had no hand in building or running the site.
What the audit report contains
The report is written so the owner can decide and the developer can act without a meeting in between.
The worked example above is the format in miniature. For a full finding at length, the penetration test report example shows the same structure applied to a signed-in lab application.
- The scope as audited: each hostname in scope, the dates of the checks, and anything excluded, such as a checkout page hosted by your payment provider.
- A hostname inventory, including names you may not know about, with where each one points and who appears to run it.
- A results table for all twelve checks, so you can see what was clean as well as what was not.
- Each finding with its evidence, business impact, fix and retest criteria, written for your platform where the platform is known.
- The questions the outside view raised but could not answer, ready to scope a deeper test if you decide you want one.
- One consolidated retest of the reported findings once your fixes are live, with each finding then marked fixed, partly fixed, still reproducible or not retested.
What the audit needs from you
An external audit needs no test accounts, which makes it simple to set up.
The preparation is about authority and boundaries: who can approve the work, and which hostnames belong to someone else.
- Signed authorisation for every hostname in scope, from a person with the authority to approve testing of it.
- Every domain you hold, including old ones that only redirect, so the inventory can be compared with what you expected to find.
- The name of your hosting provider, so its testing policy can be checked before any request is sent.
- A list of subdomains that point at services other businesses run, such as a booking or newsletter platform. Those are excluded unless their owner agrees.
- A contact who can answer questions during the audit and confirm what an unfamiliar hostname is for.
- Any questionnaire or contract wording that prompted the audit, so the report answers the questions actually being asked.
When an audit is not the right purchase
Some requests that arrive as a website audit need a different provider.
Each of these is referred on rather than stretched to fit.
- External vulnerability scans for PCI DSS Requirement 11.3.2, which an Approved Scanning Vendor performs. Clever Ops is not a PCI Approved Scanning Vendor. A merchant who needs those scans is referred to one, and an audit report does not stand in for them.
- A tender or panel that asks for a CREST-accredited provider. CREST awards accreditation to organisations that show, through an application and a review of their evidence, that they meet its standards, and penetration testing is one of its accredited disciplines. Clever Ops is not CREST accredited, so that work goes to a firm that is.
- A site that is compromised now. Containment and cleanup come first, from an incident response specialist, and the audit earns its place after the cleanup is finished.
- Load or denial-of-service testing, which needs a specialist and your host's written approval.
- A certificate or compliance sign-off. An auditor, insurer or customer weighs an audit report when they make their own decision, and the report does not certify a site.
Frequently asked questions
The difference is where each one stops. The audit never signs in, so it covers what a stranger can reach: DNS, certificates, headers, cookies, leftover files, admin screens and code sent to the browser. A penetration test works with test accounts for each role and probes the rules behind the login, such as whether one customer's records stay hidden from another. If only your own editors sign in, the audit covers the surface that matters most. If customers sign in or pay, book the penetration test.
It is planned not to. The audit sends ordinary requests at a modest pace, does not sign in or guess passwords, does not download a found backup in full or use a found key, and leaves out load and denial-of-service testing entirely. If the site or its firewall starts refusing requests, testing pauses and your contact hears about it.
Most of them, because the checks read what the site sends back rather than how it was built. Live Shopify storefronts are the exception and sit outside this service. On hosted site builders much of the configuration belongs to the platform, so the report separates what you can change from what only the platform can.
The authorisation comes from you as the site owner. Some hosting providers also ask for notice or approval before security testing, so their policy is checked for each hostname before the audit starts. Hostnames that point at services run by another business, such as a hosted booking page, stay out of scope unless that business agrees.
It can. Questionnaires may ask whether HTTPS is enforced, which security headers are set and when the site was last assessed, and the report answers those with dated evidence. Read the wording first. If it asks for a penetration test, an independent tester or a named accreditation, an external audit may not satisfy it, and the scoping call will tell you so.
Tie it to change rather than the calendar. Run the checks again after a rebuild, a new subdomain or third-party service, a change of host or agency, or a major platform update, and before a customer questionnaire falls due. Between audits, keep the hostname inventory from the report up to date, because several of the twelve checks start from it.
Your developer or agency works through the fixes in the priority order the report sets, or Clever Ops handles them in a security fix sprint agreed separately from the audit. When the fixes are live, one consolidated retest checks each reported finding against its retest criteria.
Scope the right test for your application
Tell us what you have built, who uses it and when it needs to launch. You will get a plain-English recommendation: a website audit, a web application test or a referral to a specialist.