How to try a new version of a weak ad landing page safely
Short answer: Do not replace a weak landing page in one go. Keep the old page live, build one new version that changes one thing, send a small share of ad clicks to it, and write down a stop rule before you start. Then judge it after a set number of visits, not after a good morning. You can do this with no special tool, with a landing page builder such as Unbounce, or with Ploy, an AI platform that builds and runs business websites and can split traffic on one section of a page if it is switched on for your account.
The page is weak, the ad money is real, and every headline you try is a guess. Here is a way to try a better version while the old one keeps earning.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
The safe way to try a new version of an ad landing page is to run it next to the old one, not instead of it. Keep the current page as your control, change one thing on a copy, send a minority of clicks to the copy, and decide in advance when you will stop it and how you will judge it. If version B loses, you have lost a small slice of traffic for a short time. If you swap the whole page and it loses, you lose all of it, and you may not even notice for a week.
This page is the "how to launch version B" guide. For finding out why a page is weak in the first place, read why ad clicks do not convert and how to fix the landing page. For protecting your bookings while a test runs, read how to run a website test without losing bookings. For showing a client that the result is real, read how to prove to a client that a page change helped. Prices and features here are as of November 2026, so check each vendor page before you pay.
This page may contain affiliate links from which Blur Studio earns commissions.
Why swapping the whole page is the risky way
Replacing a page outright feels decisive, but it hides two problems.
- You lose your baseline. If leads rise the week after the swap, you do not know whether it was the page, a better week, a holiday or a change in the ads.
- You change too much at once. A new headline, photo, form and price together leave you unsure which helped or hurt, and 100 percent of clicks go to a page that may be worse.
The fix is boring on purpose: one control, one challenger, one change, one goal, and a rule for when to stop.
Pick the page, the goal and the one thing to change
Start with the ad group that spends the most money on the weakest page. A page that gets 20 clicks a month cannot be tested properly, and a page that already converts well is a poor first target.
Then write down three things before you build anything:
- The goal. The one action that counts: a form sent, a booking made, or a call tapped. Pick the one closest to money. Page views and clicks on a button are not goals.
- The change. One of these is enough for a first test: the headline and opening line (the "hero section", the first screen a visitor sees), the main button wording, the offer or price shown, the form length, or the proof you show near the top.
- The reason. One sentence: "The ad says emergency repair but the headline says quality plumbing, so repeating the ad should get more calls." If you cannot write the reason, you are guessing.
Message match, meaning the page repeating the promise in the ad, is usually the best first change. Ad landing pages that match the ad explains how. Leave the ad, budget and targeting alone while the test runs.
Five ways to try version B, from least to most effort
| Route | What you do | Cost (as of November 2026) | Main weakness |
|---|---|---|---|
| No tool: one after the other | Run page A for two weeks, then page B for two weeks, and compare enquiries | $0 beyond your time | Weeks differ. A holiday or a rival's sale can swamp the page effect. |
| No tool: split in the ad account | Make a second ad pointing to page B and rotate evenly with page A's ad | $0 beyond your time | You are also testing the ads. Needs care so both ads are otherwise identical. |
| Unbounce, a landing page builder with testing built in | Build both pages in its editor and split traffic | Starter $29, Build $99, Experiment $149, Optimize $249 a month billed monthly; Agency from $349 a month billed annually. Yearly billing is lower. Unlimited A/B testing is listed on Experiment and above. | You may have to move your page into its builder. Starter allows 5 pages. |
| Instapage or HubSpot, other page builders for ad teams and a marketing platform with a page builder | Use the built-in testing | Instapage (public prices at instapage.com/plans, as of November 2026): Create $79 a month billed annually or $99 billed monthly, 15,000 unique monthly visitors; Optimize $159 annually or $199 monthly, 30,000 visitors, and it is the plan that lists server-side A/B testing and traffic splitting; Convert is custom, by demo. HubSpot, a marketing and sales platform, shows one pricing page of per-seat plans (Professional $90 per seat a month billed annually, $100 monthly): check that page for which plan includes testing. | Instapage ties plans to visitor caps; the free 14-day trial needs a card. HubSpot is priced per seat, so cost grows with your team. |
| Ploy | Test one section of one page against a challenger, with a 10 to 90 percent traffic split | Free $0, Starter $50 a month, Pro $300 a month, Enterprise by quote | On request only, one section at a time, and the statistics method is not published. |
There is also GrowthBook, an open-source tool for experiments and feature flags (switches that turn a change on for some visitors). Its Starter plan is free for up to 3 users and 1 project, and it needs a developer to set up. Google's own free testing tool, Google Optimize, is gone: it stopped on 30 September 2023. Optimizely and VWO, two large testing tools, do not publish prices for web testing, so expect a sales call. HubSpot and Instapage prices are in the table above. For a wider look at page tools, see best landing page builders for ads.
Choose the route by traffic and budget. Under about 300 clicks a month to the page, none of these will give a clean answer, and the sensible plan is a careful before and after. Over a few thousand clicks a month, a proper split is worth having.
The safe rollout, step by step
- Record the baseline. Last 30 days for this page: clicks, form sends, calls, conversion rate, cost per lead, and how many leads turned into real jobs.
- Build version B as a copy. Change only the one thing you chose. Keep the phone number, the form fields and the thank-you page working as before, unless the form is the thing you are testing.
- Check version B before any visitor sees it. Open it on a phone and on a laptop. Send a test form and check the message arrives. Tap the phone link. Make sure your conversion tracking records the test. Read the checklist in how to run a website test without losing bookings.
- Start with a small share. Send 10 to 20 percent of clicks to B and the rest to A for the first two or three days. That is a smoke test: you are looking for broken things, not winners.
- Raise to an even split only if nothing is broken. A 50/50 split reaches an answer fastest. Stay at 20 to 30 percent if the page carries most of the client's business and you cannot afford a bad month.
- Do not peek and stop. Look daily for breakage. Do not call a winner until you reach the sample you set (next section).
- Decide, then clean up. Keep the winner as the live page, delete the loser, and write down what you learned in one line.
Run A and B at the same time, never one after the other, wherever your tool allows it. That cancels out the weekday and week effects.
How much traffic is enough
People stop tests too early because the numbers look exciting. Evan Miller, a well-known writer on testing, showed in his article "How Not To Run an A/B Test" that if you keep checking and stop the moment the result looks significant, you will see a false winner far more often than expected: 26.1 percent of the time in his example, against the 5 percent the maths promises. His advice is to decide the sample size in advance and wait until it is reached.
The sample you need depends on your current conversion rate and the smallest lift you care about. The figures below are derived by me, not taken from a vendor. I used the standard two-sided formula for comparing two rates (a z-test), with equal groups, 95 percent confidence and 80 percent power, and rounded to the nearest hundred. They are for planning only, and other calculators may give slightly different numbers. Check your own case with a calculator such as Evan Miller's sample size calculator.
| Current conversion rate | Lift you want to detect | Visitors needed per version (approx.) |
|---|---|---|
| 2 percent | to 3 percent | 3,800 |
| 5 percent | to 6 percent | 8,200 |
| 5 percent | to 7.5 percent | 1,500 |
| 10 percent | to 12 percent | 3,800 |
Read this table as a warning. A small lift needs a lot of clicks. If your page gets 400 clicks a month, you cannot reliably find a 1 point gain, but you might find a big one, such as a jump from 5 to 7.5 percent. That still needs about 1,500 visitors per version, roughly 3,000 clicks in all, which is about seven months at 400 clicks a month. So make bold changes when traffic is low, not tiny tweaks.
Run it for whole business cycles. Optimizely says at least one full cycle, typically seven days. Convert, another testing company, says at least one to two full business cycles. The vendors differ, and Optimizely usually runs at a 90 percent significance level while Convert uses 95 percent as its reference. So "run it for seven days" is not a safe universal rule. For a service business with weekday calls and weekend forms, two full weeks is a more sensible floor.
A worked example with made-up numbers
These numbers are invented to show the arithmetic. They are not results from a real business.
A dental practice pays $3,000 a month for ads that send 1,000 clicks to one page for "Same-day emergency dentist". The page converts 4 percent into booked calls or forms, so 40 enquiries, and each costs $75. The headline says "Welcome to Riverside Dental".
The owner chooses one change: the headline becomes "Emergency dentist in Riverside, seen today", matching the ad. Goal: booked calls plus form sends. Guardrail: stop B if, after 150 visits, it has fewer than half the enquiries A would be expected to have at the same traffic. Planned sample: to detect a rise from 4 to 6 percent, the formula above gives about 1,860 visitors per version.
- Days 1 to 3: 20 percent of clicks go to B. About 20 visits reach B (1,000 clicks a month is about 33 a day). A form test shows the phone number worked on B. Nothing is broken, so the share goes to 50 percent.
- Weeks 1 to 3: with an even split, about 350 visits reach each version (about 230 clicks a week, split in two, over three weeks). A made 14 enquiries (4.0 percent) and B made 20 (5.7 percent).
- Decision at week 3: B looks better, but 350 visits is far below the 1,860 planned. At about 1,000 clicks a month, the full sample takes roughly 16 weeks.
Is that a clear win? No. At 350 visits per version, a gap this size is well within what luck can produce. The sensible reading is "B looks better, keep running". The owner can also choose a bolder change, because a bigger gain needs far fewer visits, or accept a before and after reading with its limits. That is a fair outcome of a small test, and better than a false claim of a win. The point is the habit: written plan, equal footing, a sample set in advance, and no victory lap on day four. For a different, tool-specific route to a split test, see how to test two versions of a landing page to get more leads, which covers Netlify's split testing and manual methods.
Where Ploy fits (one option of several)
With Ploy, you can run an A/B test on one section of a live page. In its docs, you pick the page, name the section, write a hypothesis, choose a goal event and a guardrail, and ask Ploy's Agent to draft the challenger. You then set a traffic split between 10 and 90 percent for the challenger and start the experiment. It can be paused and resumed. This fits the safe rollout above: you can start at 10 or 20 percent and raise it.
The limits matter more than the features:
- It is on request. Experiments are enabled per workspace, and Ploy's docs say to email sales@ploy.ai to get it. It is not switched on by default.
- One section of one page. The docs describe testing a section, not a full page split against a different page.
- A plan limit applies. The number of experiments you can run at once is capped by plan. Ploy's pages do not publish the number, so ask before you buy.
- The statistics are not published. Ploy's docs do not name its method or significance level. Treat its result as one input and apply your own sample rule from the table above.
- Analytics and goals first. The site must be published, Ploy analytics must be on, and your goal events must already be tracked.
- Cookie refusers are not counted. Visitors who refuse cookies always see the current version and are left out of the results.
Plans, per Ploy's pricing page as of November 2026: Free $0, Starter $50 a month, Pro $300 a month, Enterprise by quote. AI credits, which pay for the Agent's work, do not roll over, according to Ploy's docs. Read Ploy pricing explained before you pick a plan. AI-written pages and copy still need a human to read them before they go live.
Who should skip Ploy for this
- You need full-page split tests. If you want to test a completely different page layout against the old one, a dedicated page tool like Unbounce is built for that. Ploy's own comparison pages concede Unbounce is the pick if testing page variants is the main job. See Ploy vs Unbounce and Ploy vs Instapage.
- Your traffic is tiny. No tool fixes a lack of clicks. A careful before and after will serve you better.
- The offer is the problem. If the price or the promise is weak, a new hero section will not rescue it.
Copy-ready test sheet
Paste this into a note or a sheet before every test.
- Page and ad group:
- Goal (one action):
- Baseline for the last 30 days (clicks, conversions, rate, cost per lead):
- The one change in version B:
- Traffic split and start date (for example 20 percent to B for 3 days, then 50 percent):
- Sample to reach before deciding (visitors per version) and minimum days:
- Stop rule: stop B at once if the form or phone link fails; stop B early if conversions are under half of A's after a set number of visits:
- Pass: B beats A on the goal at the sample, and a check of 10 leads shows they are real enquiries:
- Fail: no gain at the sample, or more leads of poor quality:
How I'd approach this
I would start with no new tool. Pick the ad group with the highest spend and the weakest page, change only the headline so it repeats the ad, and run it at 20 percent for three days beside the old page. If the form and phone link work and nothing looks odd, move to an even split and wait for the sample you wrote down. The tool matters much less than the plan.
First-week plan: day 1, record the baseline and write the test sheet. Day 2, build version B and check the form, phone link and tracking on a phone. Day 3, start at 20 percent. Days 4 to 6, check once a day for breakage only. Day 7, move to an even split if all is well, then leave it alone until the sample is reached. Decision rule: use a split tool if the page gets several thousand clicks a month and you will test more than once, and use the free before and after route if it gets under a few hundred. If you want one platform that builds the pages and can run a section test, try Ploy's free plan and ask for experiments to be enabled, and keep your own sample rule.
If you need tests tied to your own booking or order system, or a build that records which version produced each booking, Blur Studio builds that custom version. See what that includes.
For a closer look, see testing a landing page headline.
Questions people ask
How do I test a new landing page without losing leads?
Keep the old page live, send a minority of clicks (10 to 20 percent) to the new one, and check the form, phone link and tracking first. Raise the share only if nothing is broken. The full steps are in how to run a website test without losing bookings.
What should I change first on a weak ad landing page?
Usually the headline and opening line, so they repeat the promise in the ad. Change one thing at a time, or you will not know what worked. If the page already matches the ad, try the form length, the offer or the proof near the top.
How long should I run a landing page test?
Until you reach the sample you set in advance, and for at least one to two full business cycles. Optimizely says at least one full cycle, typically seven days, and Convert says one to two cycles. For a service business, two full weeks is a sensible floor. Do not stop early because it looks good.
Can I do this without a testing tool?
Yes. Run version A for a few weeks, then version B, and compare. It is weaker because the weeks differ. A better free route is two ads with the same wording pointing to the two pages, rotated evenly, though that also needs care. Both are fine for low-traffic pages.
Does Ploy support A/B testing?
Yes, for one section of one page, with a 10 to 90 percent traffic split and a goal event. It is enabled per workspace on request (sales@ploy.ai), the number of experiments is capped by plan with no number published, and the statistics method is not documented. Check Ploy's current docs for your plan.
Is Google Optimize still available?
No. Google says Google Optimize and Optimize 360 stopped on 30 September 2023. Google's help page lists AB Tasty, Optimizely and VWO as testing providers it works with on Google Analytics integrations. GrowthBook has a free Starter plan for small teams with a developer.
Will a landing page test hurt my Google search rankings?
It should not if you follow Google's guidance on testing: do not show Googlebot different pages from people (that is cloaking), use rel="canonical" on alternate URLs, use a 302 temporary redirect rather than a 301, and remove test elements promptly when the test ends. Google's page also warns that running a test longer than needed may look like an attempt to deceive search engines. Check Google's current guidance, and see the bookings guide.
More in this topic: Websites: testing and proving page changes (4 guides) →
Sources
- Ploy docs: Experiments
- Ploy pricing
- Unbounce pricing
- Evan Miller: How Not To Run an A/B Test
- Evan Miller: Sample size calculator
- Google Search Central: Website testing and Google Search
- Google Analytics Help: Google Optimize sunset
- Optimizely: How long to run an experiment
- Convert: How to interpret A/B test results
- GrowthBook pricing