Paid Ads Are Landing on a Slow Page: What to Fix In-House and What to Hand Off
Brand promotion: this article introduces Guangsuan's (光算) PageSpeed performance optimization service and related hosting and platform services. It is a vendor introduction, not an independent test, review or third-party endorsement.
The situation is common on a mid-market paid media account. Spend goes up, clicks keep arriving, and then the client asks why the conversion rate slipped. The instinct is to look for a technical culprit, and page speed is usually the first one named, because a client can see for themselves that the site feels slow.
That instinct deserves a diagnostic rather than a conclusion. Ad platform data and page performance are two separate records: one describes what the auction and the campaigns did, the other describes what happened after somebody arrived. A falling conversion rate is consistent with a slow page, and it is equally consistent with a price change, tired creative, a form that stopped routing, an ad that no longer matches the page, an attribution change, or a quieter month of demand. Nothing in a campaign dashboard separates those on its own. Guangsuan draws the same line in its published service pages: advertising handles delivery, its performance work handles the site's loading bottlenecks, and the two meet at the landing page rather than in one report.
The short answer, with its conditions
For landing page speed optimization for paid ads, the useful starting point is to stop treating the problem as one number and split it into layers. Treat paid traffic landing on a slow page as a stack of separate problems, each needing its own evidence before anyone spends money. In practice: diagnose in the order server response, images and fonts, scripts and theme structure, then hosting and cache, because the layers above the server are the cheap ones to check and the ones a media team can handle without a developer when the fix is a setting or an asset rather than template code. Fix what you can reproduce on the pages that receive paid clicks, then measure the same URLs again under the same conditions. If a bottleneck survives that pass, it is the layer to scope for an outside team, and that scope should name the URLs, the devices, the report links and the business functions that must keep working, rather than a target number on its own.
1. Split 'slow' into four layers before spending anything
Server response: how long before the HTML starts arriving
Server response is the wait between the browser asking for a document and the first byte of that document arriving. It sits upstream of everything else: if the HTML takes two seconds to start, the images and scripts behind it are two seconds late too. To read it, open one landing page that carries paid traffic, open the developer tools' Network panel, and read the timing breakdown for the document request: connection, request sent, waiting (often labelled TTFB), and content download. If the waiting phase carries the time on every fresh request, the page is being generated at the origin each time it is asked for, which is a different problem from an underpowered server.
Guangsuan's WordPress hosting page uses exactly that separation inside a sample report template, labelled there as a demonstration and not a measured site: in the sample, an anonymous visit to a public page executes PHP on every hit, so the recommendation is to check session cookies, response headers and cache keys first rather than assume the hardware needs replacing. The same page recommends recording cold-cache, warm-cache and bypassed requests separately with the access region fixed, because an edge cache hit is not an origin response and the two answer different questions. Treat the sample as an illustration of the method; your own timing read is what decides.
Images and fonts: file weight against display size
On a paid landing page the weight usually sits in the images: the hero shot, the product gallery, the logo strip. Two different numbers matter, the weight of the file and the pixel dimensions it actually needs to fill. A photo exported at 4,000 pixels wide and 3 MB, displayed in a 1,200-pixel column on a desktop and a 360-pixel column on a phone, is still downloaded at full size unless the page offers smaller candidates. Two checks are enough to start, and both run against the live page rather than the media library: note in the Network panel the file size and pixel width of the top images against the size the layout renders, then check whether the first-screen image is the one arriving first. A hero image that is deferred or lazy-loaded leaves the page's largest element arriving late even when every file is reasonably sized.
The fix set here is unglamorous: export at the size the layout uses, serve smaller candidates for smaller viewports, keep the original for the full-size view, and let below-the-fold images load on demand. Fonts behave the same way on a landing page: a web font swapped in after the first paint moves text that a reader has already started reading, and font files are often heavier than they look in a design file.
Scripts, plugins and theme structure: what runs while the visitor waits
Theme plugins and third-party embeds execute code the media team does not own: a chat widget, a video embed, a newsletter form, analytics and ad tags, a consent platform, a review carousel. On a landing page these can occupy the main thread while every asset is small. Start with an inventory rather than a deletion list. On one paid landing page and one ordinary page, list the scripts each one loads and mark the business function each carries: the booking form, the affiliate link, the consent prompt, the chat window that sales says converts. A script with no owner and no function is a removal candidate. A script with a function is a deferral candidate, and deferral needs a functional test afterwards, not only a score.
Guangsuan's published performance scope covers this layer explicitly: critical CSS, script dependencies, plugin and third-party script weight, font loading and layout stability, alongside TTFB, cache strategy and image compression. Its published position is that resource replacement, deferred loading and removal of non-essential components are confirmed against business impact before implementation and then checked for function, and that the work does not return a simplified page to the test tool. The practical version for a media team: you may defer or remove a script, but decide what business function it carries before you touch it, and test that function afterwards.
Where a page's weight lives in theme structure rather than in settings, for instance a template that loads the whole plugin library on every landing page, or a layout that cannot be changed without code, that is where the in-house boundary starts to matter and where a scope conversation usually begins.
Hosting and cache: where extra capacity feels like a fix
Hosting is the layer where extra capacity can feel like a fix, so it is worth putting last in the diagnosis. Guangsuan's published hosting plans describe host machines commonly at 16 cores and 32 GB of memory or more, with bandwidth above 100 Mbps, and the scope includes global CDN setup and configuration, with node count and coverage confirmed per plan. Two caveats belong with those figures. First, a host specification is not a per-site guarantee: how much of a machine a site actually gets, and what its traffic allowance is, is settled in the signed plan rather than assumed from a model number. Second, more cores make the same work finish sooner, and a bigger machine delivers a heavy page sooner rather than lighter, so if the weight is in the page, hardware does not remove it.
There is also a cost argument for putting hosting last. Hosting is the only layer here that keeps billing after the work is done; image exports, script decisions and cache rules are edits you keep. And the cache rules differ by what the page does. Guangsuan's hosting page states that inquiry sites and online stores do not use the same cache rules, which applies to a landing page too: a public campaign page, a form confirmation page and a personalised thank-you page should not share one caching rule, and logged-in, cart and checkout requests need to bypass a full-page cache.
2. Read the right metrics before deciding anything
Google's current Core Web Vitals are LCP, INP and CLS, and Google evaluates them at the 75th percentile of real-user loads, separately for mobile and desktop: LCP within 2.5 seconds for the largest element to appear, INP at or below 200 milliseconds for responsiveness to an interaction, and CLS at or below 0.1 for visual stability. A single lab run on a fast office connection cannot show how most visitors experienced the page, and one good screenshot proves very little. Guangsuan's published re-testing standard records TTFB, LCP, INP and CLS together with PageSpeed Insights reports while keeping laboratory measurements separate from real-user data, and its delivery separates test scores, real-user data and business performance rather than letting one stand in for another.
Two practical consequences follow. If the complaint is that a button feels sticky or a form does not respond, a lab load test that never performs the interaction may not reproduce it, because INP needs real-user data or an actual interaction measurement. And when you compare a before-and-after, hold the conditions still: the same URL, the same device class, the same access region where you can, the same login state, and a note of whether the request hit a cache. Then repeat the run and keep the spread rather than the best figure.
3. What the media team can handle itself, and where the handoff line sits
The first three layers include work that can be completed by a team without access to the CMS structure, the theme code or the hosting account. The table is a working list rather than a checklist to be scored: what to check, where to look, what a result suggests, and who owns it.
| Layer | What to check | Where to look | What it suggests | Owner |
|---|---|---|---|---|
| Images and fonts | File size and pixel width of the top images against the size the layout renders; the font files the page loads | Media library, then the Network panel on one paid landing page | A file several times larger than the slot it fills | In-house: re-export at display size, provide smaller candidates, defer below the fold |
| Image priority | Whether the first-screen image is deferred or lazy-loaded | Page source and request order in the Network panel | The largest element on the page arrives late | In-house: keep the hero image eager |
| Third-party scripts | Which embeds and widgets each landing page loads, and what function each one carries | Network panel on the landing page, compared with an ordinary page | The landing page loads scripts the rest of the site does not need | In-house to remove or defer, then test the function |
| Theme and plugins | Active plugins per template; duplicated sliders, fonts or galleries | Plugin list, filtered per template | Several plugins do the same job, or none has an owner | In-house if it is a setting; a scope conversation if it needs theme code |
| Cache and headers | Whether public pages are cached at all, and whether static files carry long-lived cache headers | Response headers on a public URL while signed out | Every request regenerates the page | In-house to read; the rules themselves belong to whoever controls the host or the application |
| Server response | Time to first byte on the document request, fresh and then repeated | Network panel: one fresh request and one repeated request | Uncached requests are much slower than cached ones | Owned by whoever controls the host or application; for a team without that access it becomes a hosting or cache-rule conversation |
Two limits on that list. First, in-house means editing content and settings, not restructuring templates: if the only lever is the theme's own code, it is no longer an in-house task. Second, a score that improves while a form stops submitting is not an improvement, because a page that loads faster and quietly loses half its form submissions is not faster in any way that matters. Guangsuan's published delivery therefore treats functional re-test and performance acceptance as separate parts of the same engagement, and hands over the test report, the change list, the functional re-test notes and rollback instructions together rather than a score on its own.
There is a third boundary, and it is the one agencies most often blur. Message match between the ad and the page, the offer itself, form routing and follow-up, conversion tracking and the CRM handoff are not speed work. A landing page can be fast and still answer the wrong question, and no amount of loading speed repairs that. Guangsuan's published pages keep those on the advertising side of the line, where they belong.
4. What to write into the brief and the acceptance terms
When a layer turns out to be outside the team's reach, the useful conversation is about scope rather than about a score. Guangsuan's published PageSpeed service sets out its engagement in five parts, technical diagnosis, implementation of changes, functional re-test, performance acceptance and handover of the changes, which is a reasonable shape to ask any provider for. These are the items worth having in writing before work starts, because they are what turn a performance optimization scope into something comparable before and after.
- Which URLs and templates are in scope, by name. A contract that lists the home page, a landing-page template, a product or category template and article templates is materially different from one that says speed optimization. Guangsuan's published acceptance wording asks for the contract to list the templates and specific URLs involved, and states that one home-page report never stands in for a whole site, with product and dynamic pages agreed separately for a store.
- What kind of site it is. Guangsuan quotes its one-off speed work by scope, listing separate starting bands for a standard showcase site with few templates, a content or product site with many templates, plugins and scripts, and a WooCommerce store with a catalogue and dynamic functions, and it prices Shopify or custom builds according to platform permissions and what can actually be modified. A campaign page built from several templates, with tracking and chat scripts, belongs in the content and product band rather than the simplest one, and should be quoted that way.
- What will be measured, on what, and how. A public production URL with normal business functions switched on, mobile and desktop recorded separately, report links and timestamps kept rather than one high-scoring screenshot handed over. Guangsuan's published method is explicit that acceptance is done by re-measuring publicly, that the two devices are recorded separately, and that the report link, the time and the record are kept.
- Which business functions are protected. Forms, payments, tracking, consent and chat are what the landing page is paid to deliver. Guangsuan's scope keeps forms, payments and necessary tracking in place and does not return a simplified page to the test tool; removals and deferrals are confirmed against business impact first and then checked functionally. Put the protected list in the document so it cannot be argued away later.
- What you receive, and what happens next. Guangsuan's published delivery list includes the test report, the change list, the functional re-test notes and rollback instructions, and it schedules a re-check of the work within 30 days of delivery. It also states that later additions, such as a new plugin, new ads, a theme update or a server change, alter the result, so a second scope is discussed rather than assumed to be included.
- What is not in the price. Guangsuan separates hosting upgrades, CDN traffic, paid plugins or themes and third-party modifications as additional costs, and states that the adjustable server, template and application scope is confirmed before quoting.
- The device and page requirements, fixed in writing. Guangsuan states that payment terms, timeline, refund and handover responsibilities are written down before signature, and that device or page requirements are not lowered at delivery. That sentence is worth copying into any agreement, because desktop-only-at-handover is a common and expensive surprise.
5. Priority when the campaigns are already running
With live spend, the question is sequencing, and the answer depends on which records moved. If impressions and click-through rate are steady and only the on-site conversion rate fell, the page and the offer are the candidates. If impressions fell while the conversion rate held, campaign-side causes are in front of the page: auction pressure, budget pacing, seasonality, competition, creative fatigue, or a change in how conversions are recorded. A performance project does not address those.
- Check the ad-to-page match and the form path first. These are cheap to check and often decisive. If the ad promises one thing and the page opens with another, the fix is message and routing work, not a performance budget.
- Run the layer diagnosis on the pages that actually receive clicks. One paid landing page and one ordinary page, using the checks in the table above, with the findings written down rather than remembered.
- Fix the in-house items, then re-measure the same URLs. Image size, script inventory and the priority of the hero element are the ones that usually move first, and they are also the ones that decay as new creative, new tags and new pages arrive.
- Scope whatever remains. If the document request still waits on the origin after caching and page weight are handled, that is the layer to hand off, and it should be scoped as a hosting and cache question rather than as a score.
- Do not freeze the account while the diagnosis runs. Keeping spend decisions and performance work as separate records lets you still adjust bids and audiences on their own evidence while the page work proceeds. Merging them into one narrative is how a page project ends up carrying a bidding decision it cannot support.
On causation, it is worth being plain. A slow page is a plausible contributor to lost paid conversions, but campaign data alone does not prove it. It is consistent with a slow page when the drop concentrates on the pages that are slow, appears on the devices where the page is slow, starts after a deployment or a traffic spike, and shows up in real-user data rather than only in a lab score. It is not consistent when the drop started before any page change, follows a price or offer change, or appears on pages that were never slow. Guangsuan's published position fits that caution: performance and stability are a base for page experience and technical SEO, while rankings and enquiries depend on other factors, and its hosting work is stated as not being a ranking or inquiry-growth guarantee.
6. Writing a vendor's promise so it can actually be checked
A performance target becomes checkable when the sentence carries its own conditions. Weak: we will get your site to 90. Checkable: a PageSpeed Insights Performance score of 90 or above on mobile and on desktop, for the six URLs listed in the scope, measured on the public production URL with normal business functions enabled, after the agreed final change, with the report links kept on file, and if the agreed URLs do not reach the target within the agreed scope then the work continues before final payment.
- Name the metric. A Performance score is a laboratory measurement; it is not INP, not real-user behaviour and not business performance. Guangsuan's published terms separate the test score, real-user data and the business outcome instead of letting one stand in for another.
- Name the pages and the devices. Mobile and desktop separately, on the URLs listed in the contract, and never one home-page report standing in for the site.
- Name the conditions. Public URL, normal functions on, a fixed access region where possible, the cache state recorded, and the report link and timestamp kept so the run can be repeated later.
- Name the consequence. What happens if the agreed target is not met on the agreed pages, and who pays for the extra work. Guangsuan's published position is that it continues optimisation within the agreed scope and settles the balance after acceptance, with payment terms, timeline and refund responsibility settled before signature rather than at delivery.
Two boundary clauses are worth reading before signing anything. First, if the platform or theme limits what can be achieved, a script that cannot be removed or a template you cannot edit, that belongs in the quotation stage and not in a surprise at handover; Guangsuan's published approach is to check platform permissions, theme, plugins, third-party scripts and server conditions first and to confirm what can be adjusted before quoting. Second, a score is not a ranking and not a revenue result, and a provider who will not write that sentence down is telling you something about the rest of the document.
There is also a question worth asking before buying at all: if a page already meets the target and shows no obvious bottleneck, spending more purely to move a number is a poor use of the budget. Guangsuan states that in that case it does not recommend repeating the work only to reach a higher figure, and the same reasoning applies to the scope you are buying.
7. A worked example (teaching illustration only)
The figures below are invented to show the order of work. No site was measured for this article, and none of the numbers is a benchmark or a client result.
A DTC brand sends paid traffic to one landing page. The strategy team sees cost per click steady and conversion rate down for three weeks, and the client says the site feels slow on a phone. The team opens the landing page in the Network panel on a phone viewport and records four things:
- The first-screen image is 3,400 pixels wide and 2.4 MB, and it is lazy-loaded, so the largest element on the page is among the last things the phone fetches.
- The gallery below it holds eighteen more images at a similar weight.
- A video embed, a chat widget and two analytics tags load on every visit to that template.
- Every document request regenerates on the server, and a repeated request immediately afterwards is faster, which points at caching rather than at raw capacity.
Read in layers, that is three jobs and one non-finding. The image layer is oversized and badly prioritised. The script layer is running third-party scripts on a template that does not need all of them eagerly, and one of them, the chat widget, is the one sales says converts, so it stays and gets deferred rather than removed. The cache layer is not caching public pages at all. The hosting layer is not implicated yet, because the wait is in origin generation rather than in server capacity.
The order of work follows from that reading: re-export the hero at the size the layout uses and stop lazy-loading it, defer the video embed and the chat widget after checking that the chat still opens on a phone, keep only the analytics tags the campaign actually reads, set a public page cache with login and form routes excluded, then re-measure the same URL on the same device. If the document request is still slow once caching works, only then does a hosting conversation make sense, and it would be scoped as a hosting and cache question rather than as a score.
The campaign side stays separate throughout. While the page work runs, the team keeps the ad-to-page match and form routing in the campaign review, because a fast page that answers the wrong question still converts badly.
8. What to send before you ask for a quote
Guangsuan's published page states that you can send the website address and the platform, plus any existing report, for an assessment even when no report exists yet, and that the next step is a confirmation of whether the work can be taken on, with pages, cost, timeline and acceptance method made explicit. A short brief in that shape contains:
- The site address and the platform it runs on.
- Your PageSpeed Insights or Lighthouse report if you have one, and the date it was taken.
- The templates that matter for paid traffic: which landing page, which product or category page, which form page.
- Which business functions must keep working: the form, the chat widget, payment, tracking, consent.
- Who can approve a change, and who on your side can read the analytics.
- The one thing that would make the work worth doing: a specific page, a specific reader action, a specific date.
Where Guangsuan fits
Guangsuan (光算) sells this as a defined, one-off engagement rather than an open-ended retainer. Its published PageSpeed performance optimization service states a target of 90 or above on PageSpeed Insights Performance for both mobile and desktop on the URLs it agrees to take on after assessment, with acceptance by re-measuring publicly, and it works through the slow points from the server to the page one layer at a time, covering server and cache, images and fonts, CSS and JavaScript, and plugins and third-party scripts, with forms, payments and necessary tracking kept in place. For the published scope, the acceptance method and the quoting basis, see Guangsuan's PageSpeed performance optimization service.
Two related services sit outside that scope and are quoted separately. Guangsuan's WordPress hosting and optimisation plans cover long-term running, security and maintenance, including multi-datacenter offsite backups, global CDN configuration, WordPress cache and plugin optimisation, and a free migration service from the current-site check through the switch, and the same page states that hosting work is not a ranking or inquiry-growth guarantee. Where the client's store runs on Shopify, Guangsuan's Shopify technical maintenance and custom development service covers the merchant-side work that the platform subscription does not cover, namely customised theme code, installed apps, product data configuration and the systems orders flow into; that service can also be bought on its own, without attaching a promotion package. See Guangsuan's Shopify technical maintenance and custom development service.
One boundary belongs in the same paragraph. Guangsuan's PageSpeed page distinguishes its own scope from on-site technical optimisation, which covers crawling, indexation, page structure and internal links, and from WordPress hosting, which covers long-term running, security and maintenance. If the real problem is that the right pages are not being found, a performance engagement will not fix it, and buying one is an expensive way to discover that.
Frequently asked questions
Does a 90 Performance score mean our ad results will improve?
No. Guangsuan states that 90 or above is a laboratory performance target, that delivery separates test scores, real-user data and business performance, and that rankings and enquiries depend on other factors. Speed is a base for page experience, not a ranking or revenue guarantee.
Should we move hosting before we increase ad spend?
Not as a first move. Guangsuan's hosting documentation separates a cache hit from origin waiting and recommends checking session cookies, response headers and cache keys before assuming hardware is the constraint, and it treats a published specification as a host configuration reference rather than a per-site speed multiplier. Rule out the caching rule and the page weight first, and keep the page and the test conditions identical so the before-and-after is comparable.
Can we delete the video embed and the chat widget to reach the number?
Not automatically. Guangsuan's scope keeps forms, payments and necessary tracking in place and does not return a simplified page to the test tool, and removals and deferrals are confirmed against business impact first and then checked functionally. A widget with no business owner is a normal editorial decision; record it and re-test the page.
Our vendor's scope only says speed optimization. Is that enough?
No. Guangsuan's published acceptance wording asks for the templates and specific URLs to be listed in the contract, so that one home-page report does not stand in for the site, and it asks for the measurement conditions and the re-test to be settled in advance. A scope without URLs, devices and a re-test method cannot be compared before and after.
What an agency can carry away from this is not a score but a method that survives the next campaign: name the layer, fix what the team can reproduce, scope what it cannot, and write the measurement conditions down before the work starts rather than after the invoice.
Ready to turn this into pipeline?
Book a 30-minute paid strategy audit with a senior strategist. We'll map your current spend, find the leaks, and model the 90-day lift.
Get My Free Strategy Audit