To reduce PCI scope at checkout is to shrink the set of systems that can affect a card payment, so the self-assessment you sign each year gets shorter and the evidence behind it gets thinner. That is the commercial case. It is not a security case, and most coverage blurs the two. A merchant can move from SAQ D to SAQ A, cut the question count dramatically, and still lose card data on Tuesday morning because somebody with access to the content management system drew a fake payment form over the real one.
What follows is six design choices, in the order merchants usually consider them. Each one is weighed against three things that actually matter in practice: which self-assessment questionnaire it lands you on, what requirements 6.4.3 and 11.6.1 still demand afterwards, and which of the three injection routes it closes.
Those three routes, used throughout: a compromised third-party supplier whose script you load; injected first-party code after somebody gets into your admin, your repository or your server; and an unpatched or end-of-life platform.
What ‘scope’ actually means on a checkout page
The cardholder data environment covers anything that stores, processes or transmits account data. It also covers anything that can affect the security of that data. The second clause is the one that pulls your checkout page in.
A page can never see the primary account number and still be in scope, because it decides what else runs in the browser alongside the payment form. Script tags, injected tags, a tag manager, a chat widget: all of them execute in the same document. The payment form is a tenant. Your page is the landlord.
The ladder, in plain terms. SAQ D is the full standard, hundreds of questions, and where you sit if card data reaches your servers. SAQ A-EP is for merchants who never touch the data but whose site controls how the payment form is delivered: substantially shorter than D, still long enough to need a project plan. SAQ A is the short one, written for merchants who have outsourced the payment page completely. The January 2025 revision of SAQ A changed the eligibility wording again, so read the current version in the PCI SSC document library rather than a blog summary, and confirm your reading with your acquirer before you build.
Now the caveat, up front. Scope reduction cuts assessment work. It does not stop a skimmer. A fake overlay form, drawn by injected first-party JavaScript on top of a perfectly secure hosted field, harvests the card number before the real field ever receives a keystroke. The customer gets an order confirmation. So does the thief. Keep that distinction in your head while reading the rest, and keep the obligations in PCI DSS 4.0 requirements 6.4.3 and 11.6.1 in view, because both have been mandatory since 31 March 2025.
Choice 1: full redirect or hosted payment page
The customer clicks pay and leaves your domain. The payment form is rendered, served and secured by the payment service provider on their hostname. Nothing about the card entry step is yours.
This is the largest single cut to the question set available to a merchant, and it is the only arrangement that has never been seriously argued about. It also costs you something real. Branding goes, mobile back-button behaviour gets awkward, and some merchants abandon it after an A/B test shows the drop. Measure your own numbers rather than trusting a figure from a case study; the gap between a well-themed hosted page and a bad one is wider than the gap between redirect and inline. The detailed comparison sits in hosted payment fields versus redirect, including the cases where the redirect wins on conversion.
What stays yours: the page holding the button. Injected code can rewrite the destination of a redirect, post the basket to a lookalike host, or render its own card form in place of the handoff. Rare, but it happens.
- Lands you on: the shortest questionnaire, subject to acquirer agreement.
- Still owed: very little on the payment page itself, but your pre-payment pages remain worth watching.
- Routes closed: supplier and platform routes, largely. First-party injection on the button page remains open.
- Best fit: low volume, small development team, no appetite for ongoing script governance.
Choice 2: hosted payment fields in iframes
Each sensitive field, card number, expiry, security code, is an iframe served from the PSP’s domain, styled to match your theme. Keystrokes land inside the PSP’s document. The PAN never enters your DOM, never reaches your JavaScript, never touches your server. For most mid-sized merchants this is the sensible default, and the reasoning is set out in hosted payment fields as the cheapest way to remove skimming risk.
Now the honest part. Same-origin policy stops the parent page reading inside the iframe. It does not stop the parent page being reskinned, the iframe being removed and replaced, or a convincing fake form being positioned on top with a z-index and a transparent border. Three weak points, all first-party, all invisible in the payment network’s logs.
There is a second wrinkle. The PSP’s loader script sits on your page, not theirs, and it goes straight into your script inventory with your name against it. Most PSPs serve that loader from a versionless URL (Stripe’s v3 endpoint is the familiar example) and update it whenever they like, which means you cannot hash-pin it. Any vendor who tells you otherwise has not tried.
- Lands you on: SAQ A in many acquirer interpretations, A-EP in others. Get it in writing.
- Still owed: a full script inventory and change detection on the parent page.
- Routes closed: supplier route for the card fields themselves. First-party injection and platform compromise remain wide open.
Choice 3: cut third-party scripts off the payment page entirely
Run a HAR capture of your own checkout before you form an opinion. On a stock Magento or Shopware install with the usual marketing stack bolted on, the number of third-party files loading at the card entry step is rarely in single figures, and nobody in the business can name an owner for half of them.
Chat widgets, heatmaps, session replay, A/B testing, affiliate pixels, review badges. None of them belong on the step where a card number is typed. Move analytics to a server-side collector or fire it on the confirmation page instead. You lose some funnel granularity. You remove an entire category of supplier risk in exchange, which is a trade most finance leads will take once it is put to them in those terms.
The blind spot is Google Tag Manager. A content security policy that allows the GTM domain, plus Subresource Integrity on your static files, gives exactly zero protection against a tag published by a marketing contractor with container access and a deadline. The control is not technical. Audit who can publish, remove everyone who does not need it, and freeze the container for the checkout path so new tags cannot fire there without a code change. Then write down who approved that.
- Lands you on: no change on its own, but it makes a SAQ A claim defensible.
- Still owed: 6.4.3 justification and approval for every script you keep.
- Routes closed: most of the supplier route, cheaply.
Choice 4: pin and police what is left, with SRI and CSP
Whatever survives choice 3 needs holding in place. Subresource Integrity hashes each static third-party file so the browser refuses a modified copy; the maintenance cost arrives the first time a supplier ships new code without notice and your checkout breaks at 9am. That is a feature, not a fault, and the practical handling is covered in pinning the scripts you cannot control.
Content Security Policy does the rest. Use script-src with per-request nonces rather than ‘unsafe-inline’, because an allowlist with unsafe-inline in it is theatre. Then set connect-src to the handful of hosts your checkout legitimately talks to, and form-action to stop a hijacked form posting somewhere else. Those two directives are the defence that actually bites against exfiltration, and they are the compensating control for the PSP loader you could not pin.
Roll it out in that order: report-only with a live reporting endpoint for a fortnight, triage the noise (most of it will be browser extensions and your own legacy inline handlers), then enforce on the checkout path only, with the rollback already written as a single header change somebody on call can make. A broken checkout costs more in an hour than the policy saves in a year.
- Lands you on: nothing new, but it produces the evidence for 6.4.3 and 11.6.1 as a by-product.
- Still owed: periodic review of the policy as the page changes.
- Routes closed: narrows all three, closes none outright.
Choice 5: platform hygiene, because end-of-life is an injection route
British Airways in 2018 is the case that settles the argument. The script that sent card data to an attacker-controlled host was on BA’s own pages, modified in place. No unfamiliar supplier. No new entry in anyone’s vendor register. The ICO issued a £20m penalty in October 2020 over the incident, which affected the personal data of more than 400,000 customers. Supplier-focused controls would have caught none of it.
The platform route is just as blunt. Magento 1 reached end of support on 30 June 2020, and in September 2020 Sansec reported close to 2,000 Magento 1 stores skimmed over a single weekend in one automated campaign. The lesson was not “patch faster”. The lesson was that running unsupported code turns your shop into a scheduled task for somebody else.
On Magento 2: keep the security patches current, move the admin off the default path, enforce two-factor authentication on every admin account, and restrict who can deploy to the template layer. Treat an unsupported PHP version as the same problem wearing a different hat.
- Lands you on: no change, but SAQ A eligibility assumes the site is not susceptible to script attacks. An end-of-life platform makes that attestation hard to sign honestly.
- Routes closed: the platform route, which is the one that gets exploited at scale.
Choice 6: segment the shop and watch the database, not just the filesystem
Separate the people who can change content from the people who can deploy code, and log both. Most merchants do the second and forget the first, which matters because modern skimmers increasingly live in database-stored content rather than in files. A Magento CMS block, a widget instance, layout XML, the miscellaneous scripts field in the head configuration: all of them render into the checkout and none of them appear in a filesystem diff.
File integrity monitoring reports all clear. The page is compromised. Any scope review has to read the content tables and the configuration table, not just app/ and pub/.
Push the same discipline up the supply chain. Your PSP and your tag vendors should owe you notice before a script changes, a hash where one is available, and a named contact for incidents. You will not win every one of those clauses at contract renewal. Ask anyway, and record which supplier refused, because that list is useful evidence of due diligence later.
- Still owed: 11.6.1 change detection that compares rendered output, not source files.
- Routes closed: narrows first-party injection, the one route nothing else touches.
Which combination fits, and what to document
Low volume, no in-house developer: redirect, plus choice 5. Mid-sized, brand-sensitive, one or two developers: hosted fields with a cleared payment page and an enforced policy, so choices 2, 3 and 4 together. High volume on Magento or a custom stack: all six, and expect the script governance to be a standing job rather than a project.
Build the checkout allowlist once and let it serve as the 6.4.3 inventory. Each entry needs a business justification and a named approver, not just a hostname. Merchants who build the allowlist and the inventory separately end up with two lists that disagree within a quarter, which is worse than having one, because now an assessor gets to pick which one is wrong.
The evidence pack that survives review: the inventory with owners and dates, CSP violation reports from the enforced policy, SRI hashes alongside the change tickets that updated them, and dated screenshots of the payment page as rendered. The rejections that come back most often are predictable. An inventory with no approver names. A report-only policy presented as enforcement. A scope claim that a five-minute page audit contradicts, usually because a marketing tag reappeared after a theme release.
Re-check after every theme update, every extension release and every container publish. Scope creeps back quietly, and the thing that puts it back is almost never malicious.
The short version
Six choices, in descending order of effect on your questionnaire: redirect, hosted fields, strip the page, pin and police, patch the platform, segment access and watch the database. To reduce PCI scope at checkout you need the first two. To reduce skimming risk you need the last four, and requirements 6.4.3 and 11.6.1 already oblige you to do most of them, though an acquirer is unlikely to chase you for the detail. Start with the HAR capture of your own checkout, then read the working CSP setup for a checkout page before you write a single header.
Frequently Asked Questions
Does reducing PCI scope at checkout also stop card skimming?
No. Scope reduction changes which questionnaire you complete and how much evidence you gather. Injected first-party code can still draw a fake form over a hosted field, or replace the payment step entirely, on a site that is legitimately eligible for SAQ A. The controls that stop that are script governance, change detection and access control.
Do hosted payment fields qualify me for SAQ A?
Sometimes, and it depends on your acquirer’s reading of the current SAQ A eligibility criteria, which changed in the January 2025 revision. Many acquirers accept iframed fields where the PAN never enters your DOM; others still place that arrangement on SAQ A-EP. Ask in writing before you build, and keep the reply.
Which scripts are allowed on a payment page under PCI DSS 4.0?
The standard does not publish a permitted list. Requirement 6.4.3 says every script on the payment page must be authorised, its integrity assured, and an inventory maintained with written business justification. In practice that means the PSP loader and whatever you can genuinely justify to an assessor. Chat widgets, heatmaps and affiliate pixels rarely clear that bar.
Is a redirect checkout worth the conversion loss?
Test it on your own traffic rather than accepting a general figure. For a low-volume merchant with no developer and no capacity to manage a script inventory, the reduced assessment burden usually outweighs a small conversion gap. For a high-volume retailer with a design team, hosted fields plus a cleared payment page is normally the better trade.
What evidence does an acquirer want to confirm reduced scope?
A script inventory with justifications, approvers and dates; a rendered-page audit showing what actually loads at the card entry step; CSP violation reports or equivalent change-detection output for 11.6.1; and change tickets tying script updates to approvals. Screenshots help. An unsupported assertion that the page is clean does not.
